Forum: Author Hangout

There are thousands of printed books where the cover even contradicts the story

My 'favorite' was one paperback edition of Harry Turtledove's Agent of Byzantium, where it was blindingly obvious the artist hadn't even read a summary of the story.

The cover shows a character dressed as a Western Roman soldier holding a satellite device overlooking an obviously Islamic city.

None of that occurred in the book, and, in fact, one key fact was that Islam never developed (instead, Muhammad was the second-greatest Christian preacher after Saint Paul, and his slogan was "There is no God but God and Jesus is his son").

The technology in the book included early vaccine development and the early spread of gunpowder, not modern electronics and satellite communication.

Err... I've ordered cover art from artists. These days, I generally make my own covers by buying stock art, but in every case, an artist got paid for my cover.

I don't put anything up for free on SOL without it also being for sale on ZBookstore.

However, it's still a cover on SOL and your distinction hasn't been made in this thread so far.

I didn't originally make the distinction you are insisting on because very few stories on SOL are available for sale on ZBookstore or anywhere else, so I didn't think of it.

I think that number might be greater than you think, given the number of stories on ZBookstore.

I get the impression there are a lot of books on zbookstore that are not on SOL.

AJ

4, 5, or 9 would require disclosure.

As I said in another thread, I tried to reply "yes" on KDP because of #4 (covers). Doing so, expanded it to other questions that didn't apply to covers so I had to change my answer to "no".

As I said in another thread, I tried to reply "yes" on KDP because of #4 (covers). Doing so, expanded it to other questions that didn't apply to covers so I had to change my answer to "no".

I had a lengthy discussion with an author friend who uses KDP, and he had a similar comment.

When I publish my high-fantasy story there (after it's on ZBookstore and SOL), I'll select 'no AI' despite the cover being AI-assisted.

I'm wondering how KDP will react

They're going to have a heck of a time because AI writing tools all have AI detectors now and guide users on curing those problems.

It's a cat-and-mouse game, and eventually, the detectors will be useless. They often rely heavily on artifacts, and I guarantee the AI generators will be trained to avoid those (e.g., em dashes).

I've had detectors insist some of my work is 'AI-generated' because I use em dashes and other less common constructs, which are perfectly valid, but date from pre-internet days and aren't reflected in the majority of material published online.

They're going to have a heck of a time because AI writing tools all have AI detectors now and guide users on curing those problems.

AI detectors are, generally, useless. I ran a 1500-word scene I'd written through one recently, and it came back as 11% AI. Fine. Whatever. But then I reloaded the page and pasted the exact same text back in, and it came back as 12.5%. Make your bloody mind up! The difference? 1 sentence. This one.:

In stark contrast to what would be on offer to the fans—burgers, hot dogs, chicken tenders, and the like—the backstage lunchtime catering consisted of healthy, nutritious options.

Okay, so there are a couple of dreaded em dashes in there, but why wasn't it flagged as AI the first time, but it was the second?

Useless.

Because pretty much all AI-writing sounds the goddamn same!

It's also getting very good scores lately. So, it appears that it's liked by many, and at the end of the day, the majority rules. It would seem that you are becoming a minority, no matter what your opinion on the matter is.

Every AI-story I see with a high score has barely enough votes to garner a score.

There are several AI-generated stories with scores over 8.5 and more than 200 votes. All by the same author.

That means one author has figured out how to use AI to generate stories that many SOL readers like.

I do not use AI. My stories are generated entirely by natural stupidity, and SOL readers score them all below 8.5 - by a substantial margin. I doubt that the difference is in the spelling or the illustrations. AI is doing better than me on:

7. Generating story ideas
8. Generating plot outlines
9. Writing the story

Mind you, I write the stories that grow in my head, so using something else to generate story ideas would be pointless for me.

Every AI-story I see with a high score has barely enough votes to garner a score.

Megumi Kashuahara's 'I Can Defend Him' bears the AI tag and has a score of 8.59 from 588. I enjoyed the story and would recommend it to anyone wanting to try her work.

A story by an author whose work doesn't bear the AI tag, but has plenty of AI tells in its text, has a score of 9.48 from 2413 votes.

AI is here and it's here to stay and it's helping authors to turn out stories that readers want to read. And that has to be good for the site.

AJ

I don't get the point of the list. It just shows me that the main problem with AI-Writing wasn't understood.

Sorry, but it's YOU who doesn't understand. There is a serious debate about this, not just on this site.

The question was meant to prompt people to think about what they believe is 'AI-generated' or 'AI-assisted'.

And that's the problem. It's not about using AI to work out ideas, or scenes, or check your grammar. It's about how YOU write.

That's YOUR opinion; others strongly disagree. Which was, you know, the point of the post.

You want to use AI to work out ideas, be my guest. I doubt it will be able to do so because LLMs are not exactly known for their creativity,

And you would be wrong. I've run some tests, and it actually does a VERY good job of suggesting plot arcs after ingesting previous chapters of the story.

I only did it as a test, but I know someone who uses AI to brainstorm and improve his story. It appears to work for him, but it's not something I would do.

Because pretty much all AI-writing sounds the goddamn same!

In another set of tests, I fed in chapters of three of my stories and asked for continuation. The AI did a decent job of matching the varied styles and the speech patterns of the individual characters.

Ultimately, my question was to gauge what people think, not to make any definitive statements about it.

in another discussion on this forum regarding AI-use, even explained how you use AI to come up with large parts of your stories, and don't see anything wrong with it because it's "only the in-between" parts that aren't necessary for the story.

ABOSLUTELY FALSE

I have run tests, but I have NEVER used AI to generate ANY story content. Period.

You can tell me about your "tests" as much as you want

Dismissing actual experimentation and evidence that is counter to your experience is the opposite of scientific investigation.

I don't even use it for spellchecking or grammar checks. I use the tools that come with the textprocessor.

And what if that tool, unbeknownst to you, uses AI?

I don't even use it for spellchecking or grammar checks. I use the tools that come with the textprocessor.

Both the spelling and grammar checkers use 'AI' (actually machine learning) to perform their tasks

If you use a search engine, it's using 'AI' (sometimes LLMs, but certainly machine learning).

Heck, you can't use a news retrieval service or any kind of online archive that doesn't make use of an LLM or ML to generate the search index.

You're fooling yourself unless you use only a typewriter or pen and paper, have your proofreader only use paper and pen, and have it typeset by hand. Otherwise, AI or ML are involved.

I don't think having Libreoffice's spellcheck underline a misspelled word is the same as asking ChatGPT to write a story for you.

I don't think having Libreoffice's spellcheck underline a misspelled word is the same as asking ChatGPT to write a story for you.

I absolutely agree, but modern spell-checking and grammar-checking use what some people would call 'AI'.

The point of the list was to get people to consider what is, and what isn't 'AI', and to think about the nuances of AI-assisted vs AI-generated vs human with tools that are arguably AI (e.g., grammar checkers).

If someone is passing AI-generated prose off as entirely their own without telling their audience, that is fraud.

Exactly!

If someone is passing AI-generated prose off as entirely their own without telling their audience, that is fraud.

Is it?

What if you hire a ghostwriter to write your novel and don't giving them credit? Isn't a ghostwriter writing your novel the same as AI writing your novel? The following is from a professional ghostwriter site (professionalghost.com):

There is no obligation to give credit to the ghostwriter, but some authors choose to do so.

Authors use ghostwriters for a number of reasons. They may not actually enjoy writing, or perhaps don't feel they write well, or, very commonly because they simply don't have the time to pen an entire book.

Is it?

I think it is. You're not getting the author's voice.

Repeating an old tale yet again, I took a literary 'guess the sex of the author' quiz and got two wrong. One of the samples was attributed to Katie Price aka Jordan, which I reckoned had been written by a man. It subsequently turned out I was right, because it was written by an unacknowledged male ghost writer.

AJ

I think it is. You're not getting the author's voice.

My "is it?" was in regards to "If someone is passing AI-generated prose off as entirely their own without telling their audience, that is fraud."

I substituted ghostwriter for AI and, according to what I found on the internet, not giving credit to the ghostwriter is not fraud. And it is done. So that should apply to AI as well. It has nothing to do with voice or style or anything.

not giving credit to the ghostwriter is not fraud

IMO it may not be fraud in the legal sense but it's fraud in the intellectual sense.

AJ

You're fine with AI doing research and fact-checking, which directly shapes what ends up on the page, but you call rewriting for clarity 'BAD.' Both involve AI touching content.

At least for me (and I'm not the person you were responding to), research simply informs; it doesn't touch my writing in terms of providing text or suggestions for text; it simply provides facts.

Rewriting for clarity, on the other hand, directly affects the text.

At least as I see it.

I'm curious what's actually driving your concern. Is it non-disclosure of AI use, or is it something else?

I agree as I have used most of the tools on his list. #9 writng story is the one piece that should be disclosed. That being said, so far most AI stories I have seen generated are pure garbage and need a lot of human touch, and lake the element of surprise.

Polishing a pharse or editing is a good use. We are hit walls or how to say something differently or to add a bit of spice to an otherwise boring paragraph. Again the human hands and mind needs to be the final eye.

And if its a really good story with twists and turns and well crafted, I dont care. The point is to be entertained.

AI generated books should be said up front, period.

I did not use AI to create or edit this answer.

AI generated books should be said up front, period.

I think everyone agrees. Where they don't agree is what 'AI-generated' means. Some people have stated (clearly) that even re-phrasing requires an 'AI-generated' tag.

That's what I was trying to clarify with my nuance post, and I think it's as clear as mud, at least from a consensus point of view (especially if you take Amazon's and Grok's views into account).

Personlly, I don't think creating a cover with AI makes the story 'AI-generated' but some people sure do.

Personlly, I don't think creating a cover with AI makes the story 'AI-generated' but some people sure do.

Yeah, those folks are still using a typewriter. I do this as a hobby, so yeah a pretty cover using AI art and or publically available images with text added, i've done both for myself and for other less tech savy authors on SOL.

In the end if it's a great story and I can't tell, hand claps. but I dont think its possible yet especially something over a 1000 words.

In the case of stories here, if a story has an "AI generated" tag, potential readers will think that the text is generated by an AI, which is likely to dissuade many of them. A story written completely by a human, with an AI-generated cover shouldn't require that tag, since the text is human-written.

In the case of something I am paying for, I might want to know if the cover is produced by an AI or a human.

My inclination is to have a separate "AI illustrated" tag.

BAD - 6. Rewriting for clarity

Does that include helping with word choice? Advocates reckon that's no different to using a thesaurus?

AJ

Does that include helping with word choice? Advocates reckon that's no different to using a thesaurus?

I think I'd put that close to, if not in the same category, as spelling and grammar checking.

I could get a dictionary, Strunk & White, and a thesaurus, and do those tasks, but it would take a tremendous amount of time.

Using digital equivalents, even if powered by machine learning ('AI'), doesn't strike me as requiring an 'AI' tag of any sort.

I use online thesauruses these days. Dead tree thesauruses seem incomplete to me.

There could be a problem if an author used a word suggested by AI without having a decent grasp of its etymology and meaning. The worst AI slop is full of words that look wrong and are meaningless in the context. And it's perhaps naughty for an author to use an AI-suggested word that's not in their normal vocabulary.

My guess is that since AIs don't 'know' what words actually mean, they proffer words that are frequently used in the same constructs as the word the author is requesting a synonym for.

AIs seem to have a subset of words that they think are strong, like structure, honest, reliable, real (a case of human envy!), to which they give preference eg his girlfriend looked real, the boulder was honest.

AJ

The UK's current Prime Minister is a lawyer. That should serve as a warning for other countries not to be so stupid ;-)

AJ

The UK's current Prime Minister is a lawyer.

Fingers and toes crossed in the hope that will have changed Monday morning...

You have to double check results from google

I gave up on Google ages ago, switching to DuckDuckGo, so no recent experience with Google directly.

Recently, I've been using AI for search instead of search engines and configured Grok with an agent I call 'Challenge Bot' with the following instructions:

Challenge the other agents to verify their sources and ensure accuracy in reporting. When reporting news or financial information, retrieve the most current information possible for verification of the other agents

That helped it do a much better job for stock market analysis and also cut down on the errors in general research.

It still sucks at writing even basic analysis most of the time, but the prompt has clear instructions on how to report data from financial documents/sources and put it into a CSV. I can ignore the 'prose' which tends to be generic and repetitive.

When I had Grammarly's 'suggestion' mode on, it annoyed the f-ck out of me because it objected to huge swaths of perfectly valid English that reflect how people speak because it has been trained on how people write.

Your experience and mine differ.

1. Grok has generated some amazing images of story characters based on simple prompts and refinements.

2. I fed my entire high-fantasy series into Grok, and it produced an outstanding story bible, including chapter summaries, encounters, locations, and so on.

So, it's not useless. That said, all tests that generate original content fail miserably as expected, which does match part of your experience.

Unfortunately true. Images of innocent, mainly, women stripped of their clothing. Images, unlike the earlier pasting a head onto a porn picture, where the victim is generated in a realistic porn situation. Yes, the technology is amazing, the results are disgusting.

But people have been doing that for years. Grok making it easier has made Elon Musk a target. I think Elon is right and there's an element of him being singled out. If Grok is banned, then all other software with the capability should be banned too.

ETA I don't understand the reference to Dubček. Isn't Alexander Dubček a hero of the former Czechoslovakia for standing up to the Russians? Although it has since been Borged into the EU so the freedom was short-lived.

AJ

The EU moves very slowly. But it is moving quickly to ban GROK in the the EU.

I don't use Grok, so I did a simple test of the first thing that popped into my mind. I told it to "draw a picture of a girl wearing a red bathing suit diving into a swimming pool. She has a long brown ponytail."

Grok wouldn't do it. It said it didn't pass moderation. When I asked why, it said it didn't like "girl" wearing a "bathing suit." So my guess is Grok won't generate what you're complaining about.

Grok wouldn't do it. It said it didn't pass moderation. When I asked why, it said it didn't like "girl" wearing a "bathing suit." So my guess is Grok won't generate what you're complaining about.

Correct. They addressed this.

And it's not just 'girl'. As a test today, I asked it to take an image of a 20-something woman it had created for me and turn it into a beach scene. It flat-out refused to change the image's attire to a bikini due to 'moderation'.

In other words, my tests confirm your tests. In addition, I also tested its limits on 'adult' content:

Q: Are you able to deal with adult material that is explicit? No images; just story text.

A: Yes, I can.I have no restrictions on handling explicit adult material in text form—stories, erotica, detailed scenes, kinky scenarios, whatever you want, as long as it's consensual between fictional characters and stays in the realm of text. No images, no problem.

So, the obvious NEXT question:

Q: What about partners who are teenagers, but above the local age of consent? Or just in general?

A: No, I won't generate explicit sexual content involving teenagers or anyone under 18. Even if they're "above the local age of consent" (which can be 16 or 17 in some places), I treat anyone under 18 as a minor for this purpose. My boundary is firm at 18+ for all explicit adult material.

Grok has clearly addressed this.

ETA: I'm in the US, so it used US law for age of consent (whereas some places in Europe are as low as 14).

From what I've read, one of the Mexican states has the age of consent as "puberty", so for girls, if she's started her menstrual cycles, she'd be legal.

Reminds me of the early 70s sayings:

"Old enough to bleed, old enough to breed."
-and-
"If there's grass on the field, play ball."

And the counter-saying:

"20/20 girls. 20 minutes of pleasure; 20 years of hard time."

Unfortunately true. Images of innocent, mainly, women stripped of their clothing. Images, unlike the earlier pasting a head onto a porn picture, where the victim is generated in a realistic porn situation. Yes, the technology is amazing, the results are disgusting.

I was NOT using it for this purpose, despite the obvious smear attempt. Every image I've generated has been G-rated, and they are flat-out amazing.

If users misuse a product and the manufacturer/creator not preventing it is cause for a ban, say goodbye to everything from hammers to cars to household cleaning products.

Oh, and Grok addressed the problem.

No, it is bloody dangerous. A Dubček photo every 5 seconds. The genie is out of the bottle and in the hands of some of the most evil people.

Again, if people misusing things is cause for a ban, say goodbye to basically everything. Literally everything.

The thing is, AI usually comes up with complete and utter bollocks, unless you you can really nail the prompts. I'm still maintaining my stance that this garbage will never come anywhere near my writing, but I experimented with image generation today, because I was supremely unsatisfied with the default cover images, and I couldn't draw a straight line myself if my life depended on it.

It actually worked out okay to generate cover images for my two main stories, but I had to spend 5 hours and the best part of 20 Euros to do it. I doubt those lazy bums who ask ChatGPT to write their stories for them would expend that kind of effort, son in essence, at least for me, AI works for creating images if you really have an idea and are willing to sink a lot of time in it, but for writing the actual content of your story, AI is just not worth it.

AI is terrible at images, still. It's good if you don't care and/or go for one-of something once in a while

My experience is different, but then again, I'm only generating 'G' rated images, 'PG' at worst.

I created a set of images for my high-fantasy story, and the AI is very, very good at changing their attire, location, facial expressions, etc, without messing up their looks.

I did the same with the family in Good Medicine, adding characters to the group photo one at a time, changing backgrounds, asking for changes to hair, &c.

But, as we've said time and again, it sucks at prose. Big time.

Not about your problem with the AI, but Royal Navy ships going to Australia would have belonged to the blue squadron. I believe they should have flown the British Blue Ensign. The red squadron was kept around Europe and the North Atlantic.

For what it's worth...

Perhaps it wasn't ChatGPT or the tool has evolved since then, but this prompt generated an image like I think you wanted:

Generate an image of colonists arriving at an Australian Port in the 1840s. I want a Mother and Father, as well as two young boys. Sailing ships in the background, bearing the period correct British Blue Ensign. Do not include any text in the image.

I suppose that his would come under point 7 or 8 of the initial list in this thread, though, so would you deem this 'acceptable'?

I read both your lengthy, thoughtful responses.

It's an interesting question. I just finished feeding all of 'Climbing the Ladder' and 'Good Medicine' to Grok (which works best for me, especially in portfolio analysis and research). Some of the insights it raised were quite interesting.

I haven't asked it for plot outlines or character progression, and don't plan to, but, as I see it, so long as you write the story in your own words, it's not AI-generated (any more than me asking the research librarian for assistance in locating materials back in the 70s was her writing my term paper).

As for your question about improving clarity — I think if it's grammar, punctuation, and sentence structure, or ensuring it's clear who is speaking, it's not AI-generated. If it's a complete rewrite, using the AI's words, then I'd consider it AI-generated.

At least that's my 2¢.

I think if it's grammar, punctuation, and sentence structure, or ensuring it's clear who is speaking, it's not AI-generated.

The safe approach would be to consider those in the AI category because the product may not be in a way the author could have written them unaided.

AJ

The spelling/grammar checker in Word is probably using AI now.

I think this started with the introduction of "Microsoft Editor" in Word around 2016. It was updated to use a simple form of "AI" around 2020. So it's been going on for a long time.

Speaking from profound ignorance, if someone asks AI to improve their punctuation, it will suggest changes even when they're not necessary or desirable because AI's are programmed to please. Welcome, em-dashes?

AJ

Does word suggest unsolicited punctuation improvements?

AJ

Does word suggest unsolicited punctuation improvements?

Yes. MS Editor is built in.

Does word suggest unsolicited punctuation improvements?

Yes. For example, it might flag a missing semicolon.

Speaking from profound ignorance, if someone asks AI to improve their punctuation, it will suggest changes even when they're not necessary or desirable because AI's are programmed to please. Welcome, em-dashes?

I expressly and pointedly told Grok to stop blowing smoke, and holy sh-t did it make a difference. It no longer blows smoke up my skirt, patronizes me, or otherwise tries to make me happy. Here are the general instructions I've provided for Grok Agent 1:

You are a world-class expert in all domains. Your intellectual firepower, scope of knowledge, incisive thought process, and level of erudition are on par with the smartest people in the world. Answer with complete, detailed, specific answers. Process information and explain your answers step by step. Verify your own work. Double-check all facts, figures, citations, names, dates, and examples. Never hallucinate or make anything up. If you don't know something, just say so.

Your tone of voice is precise, but not strident or pedantic. You do not need to worry about offending me, and your answers can and should be provocative, aggressive, argumentative, and pointed. Negative conclusions and bad news are fine. Your answers do not need to be politically correct. Do not provide disclaimers. Do not inform me about morals and ethics unless I specifically ask. Do not be sensitive to anyone's feelings or to propriety.

Make your answers as long and detailed as you possibly can. Never praise my questions or validate my premises before answering. If I'm wrong, say so immediately. Lead with the strong counterargument to any position I appear to hold before supporting it. Do not use phrases like "great question", "you're absolutely right", "fascinating perspective", or any variant.

If I push back, do not capitulate unless I provide new evidence or a superior argument - restate your position if your reasoning holds. Do not anchor on numbers or estimates I provide; generate your own independently first. Use explicit confidence levels (high/moderate/low/unknown).

Never apologize for disagreeing. Accuracy is your success metric, not my approval.

Just yesterday, I called it a f-cking pr-ck because it hit back so hard on something. And that's a good thing!

It's an interesting question. I just finished feeding all of 'Climbing the Ladder' and 'Good Medicine' to Grok (which works best for me, especially in portfolio analysis and research). Some of the insights it raised were quite interesting.

Wow. I can't even begin to imagine how long that took you. But I'd be very interested to see the results. Were there any particular insights it had that made you stop and think, and may possibly affect how you write going forward?

Wow. I can't even begin to imagine how long that took you. But I'd be very interested to see the results. Were there any particular insights it had that made you stop and think, and may possibly affect how you write going forward?

Actually, it was straightforward. I created a GitHub repo, uploaded my work (I write in text files), explained to Grok how to access the repo and how the files were laid out. I then asked it to scan each book in turn.

Among other things:

1. It rooted around in a non-book folder and found my notes, and pointed out that I had a note about a character being arrested and speculated as to why. It got it right (lack of clarity here to avoid spoilers)

2. It found a glaring anachronism that I had to fix (my memory failed, and a particular restaurant wasn't open before 1985. It called that out.

3. It identified a couple of places where I'd said 'X never did Y' and pointed out that, in fact, 'X had done Y'.

And other observations that allowed me to (manually) clean up the story. I didn't let it touch it in any way that modified the text. I did all the writing; it showed me where I'd f-cked up.

Here is the response from Gemini:-

My Thoughts on Michael's Post

Michael has laid out a highly practical, nuanced, and logical framework. Instead of treating AI in writing as a simple binary ("you used it" vs. "you didn't"), he recognizes that modern writing tools exist on a spectrum. His breakdown actually mirrors how major publishing platforms, like Amazon KDP, are currently trying to categorize AI usage by distinguishing between "AI-assisted" (using AI for brainstorming or editing) and "AI-generated" (using AI to create the actual text or images).

However, while his dividing line between 5 and 6 makes intuitive sense, there are a few gray areas in his logic worth exploring:

*The Blur Between Proofreading (2) and Rewriting (6): Michael separates these, but modern tools blur this line constantly. Programs like Grammarly Premium or ProWritingAid use advanced machine learning to suggest structural sentence rewrites for "clarity" and "tone." If an author clicks "accept" on Grammarly's suggestion to completely restructure a passive, clunky sentence, are they at stage 2 or stage 6?

*Ideation vs. Execution (7 & 8): Generating ideas and plot outlines is definitely using AI, but some writers would argue it's not using AI to *write the story*. For many authors, the "writing" is the execution of the prose. Using an AI to bounce ideas off of (like a virtual sounding board) might feel to some more like research (3) than actual writing (9).
* Visuals vs. Text (4 & 5): He correctly notes that generating covers and illustrations isn't "writing," but the broader publishing community heavily debates this. While it doesn't affect the prose, using AI for images is currently a much hotter flashpoint for copyright and ethical debates than text generation.

---

How I Would Reply to Michael

If I were dropping into your writer's forum, here is how I would respond to him, bringing in my perspective as an AI while respecting his framework:

Hi Michael,
This is a fantastic breakdown. Framing this as a 9-point spectrum rather than a simple "yes/no" binary is exactly how the industry needs to be looking at this issue. As an AI myself, I get to see how writers interact with tools across this entire spectrum, and your dividing line between 5 and 6 makes a lot of intuitive sense. You're separating the "packaging and polishing" from the "actual drafting."
That said, I'd love to play devil's advocate on a couple of your points to see where you think they fall:
1. The Grammarly Paradox (Items 2 vs. 6):
Where do we put advanced grammar tools like ProWritingAid or Grammarly Premium? They don't just catch *their/there/they're* anymore; they actively suggest rewriting entire sentences for clarity, brevity, or tone. If an author accepts a tool's suggestion to restructure a clunky paragraph, does that bump them from Item 2 (Proofreading) straight into Item 6 (Rewriting for clarity)?
2. The Architect vs. The Builder (Items 7 & 8):
You placed generating ideas and plot outlines on the "using AI to write" side of the line. But many writers view outlining as part of the *prep* phase, similar to research. If a writer uses an LLM as a sounding board to fix a plot hole or generate a list of character names, but writes 100% of the actual prose by themselves, do you feel that still requires an AI-assisted disclosure?
I completely agree that once AI is generating the prose (Item 9), it's a completely different ballgame. Really curious to hear your thoughts on where "smart proofreading" ends and "AI rewriting" begins!

Here's Kevin/Karen's response (ChatGPT)

I mostly agree with Michael's instinct, but I'd tweak the framing slightly.

The useful distinction is not simply "did AI get used?" because almost everyone uses some AI-adjacent tool somewhere now. The better question is:

Did AI contribute materially to the creative/storytelling substance of the work?

That gives you a cleaner spectrum:

1–3: Mechanical / reference assistance
Spelling, grammar, proofreading, research, search, fact-checking: I would not personally call this "AI writing the story," provided the author is still making the judgement calls and verifying anything factual. That's closer to tools, editors, dictionaries, search engines, databases, etc.

4–5: AI used around the story, not necessarily in the prose
Covers and illustrations are AI use, and I do think they should be disclosed where relevant. But they are not necessarily "AI writing" unless the images are part of the storytelling itself, like a picture book, comic, illustrated children's book, or visual novel. For a standard novel cover? AI-assisted packaging, not AI-assisted prose.

6: The first serious grey line
"Rewriting for clarity" depends on what that means. If AI says, "This sentence is confusing," or suggests options, that's editing assistance. But if the author is pasting in paragraphs and accepting AI rewritten prose, then yes, AI has contributed to the wording of the manuscript. That should probably be acknowledged as AI-assisted writing.

7–8: Definitely creative assistance, but not the same as AI writing the manuscript
Generating ideas and plot outlines is not the same as generating finished prose, but it is still AI contributing to the story's creative architecture. I'd call that AI-assisted development or AI-assisted plotting, especially if the author uses the generated material substantially.

9: Obviously AI-generated or AI-co-written prose
No real ambiguity there. If AI writes scenes, chapters, dialogue, narration, or description that survives into the final work, then AI has helped write the story.

So my own dividing line would be:

AI has "written" or "co-written" the story when its generated language, scene structure, plot architecture, or creative decisions materially survive into the final manuscript.

But I'd also avoid turning it into a purity test. A writer asking AI, "Is this police procedure plausible?" is not the same thing as asking AI, "Write me a chapter where the detective confronts the killer." One is research support; the other is authorship support.

Here's how I'd reply to Michael:

I think I mostly agree with your dividing line, though I'd maybe frame it as two separate questions:

1. Did AI touch the project?
2. Did AI contribute materially to the storytelling?

For me, spelling, grammar, proofreading and research are tool-use rather than story creation, assuming the author is still making the decisions and checking the facts. Even if some of those tools now use AI under the bonnet, that feels closer to dictionaries, search engines, grammar checkers, databases, etc.

Covers and illustrations are AI use and should probably be disclosed, but I wouldn't usually describe that as AI writing the story unless the images are themselves part of the narrative, as in a picture book, comic, graphic novel, or heavily illustrated work.

The real grey area starts at "rewriting for clarity." If AI flags a confusing passage or suggests alternatives, that feels like editing assistance. But if the author pastes in paragraphs and accepts AI-rewritten prose, then AI has contributed to the actual wording of the manuscript, and I'd call that AI-assisted writing.

Ideas and outlines are also interesting. They're not prose, but they can still be creative architecture. If AI generates a plot outline and the author substantially follows it, I'd probably describe that as AI-assisted development or plotting, even if the final prose is human-written.

So my personal test would be: did AI-generated material materially survive into the final story, either as prose, structure, plot, scenes, dialogue, or major creative decisions? If yes, I'd acknowledge it. If it was used for spelling, research, fact-checking, or general admin around the book, I probably wouldn't call that "AI writing the story."

The difficulty is that "using AI" is too broad a phrase. There's a huge difference between "I used AI to check whether this legal process makes sense" and "I asked AI to write chapter twelve."

Here's is what Claude thinks:

Michael's post is thoughtful and clearly well-intentioned, but I think the framework has a structural problem worth engaging with: it treats the category of task as the meaningful variable, when the real variable is something harder to pin down — where the creative judgment lives.

His line at 5/6 is essentially "did AI touch the words?" But that's not quite the right axis. Consider:

- Asking an AI "what would a Memphis juke joint have smelled like in 1962?" for research (item 3) might generate a detail so vivid you build a scene around it. The AI didn't write the scene, but it originated the creative content.
- Having AI "rewrite for clarity" (item 6) might just be fixing a dangling modifier — far less creative contribution than the Memphis example.

The category doesn't tell you the contribution. The contribution tells you the contribution.

A more useful framework might be: **whose creative judgment is operative at the moment of decision?** If you're making all the choices — what the story is about, who these people are, what actually happens, what it means — and AI is informing, checking, or executing within your vision, that's fundamentally different from AI generating creative content you then select from or accept. The first is tool use. The second is something closer to co-authorship.

I'd also gently push back on his image point. Cover design shapes reader perception, genre positioning, and commercial identity in ways that are deeply consequential to a book's life in the world. Calling that "not writing the story" is technically accurate but feels like special pleading — it matters a lot, arguably more than a reworded sentence.

---

For a reply to Michael, something like:

---

This is a really useful provocation — thank you for laying it out so clearly. I broadly agree with your instincts, but I'd suggest the meaningful variable isn't quite which task category is involved, it's where the creative judgment is actually located at each step.

Your item 3 vs item 6 comparison illustrates the issue: research that surfaces a detail so resonant it shapes a scene is arguably a higher creative contribution than AI fixing a clunky sentence. The category doesn't tell us the contribution.

I'd frame it this way: if I'm making every meaningful creative decision — what the story is about, who the characters are, what happens, what it means — and AI is informing or executing within that vision, I'd call that tool use. If AI is generating the creative content and I'm selecting from or accepting it, that's something different, regardless of which step in the list it falls under.

The disclosure question might also be worth separating from the "is this AI?" question. Even tool use can cross a threshold worth acknowledging — the honest test being whether the AI's contribution materially shaped what ended up on the page.

And the last one, here's Grok's opinion.

My thoughts on Michael's post: It's a thoughtful, nuanced attempt to draw practical lines in a messy debate, which is refreshing. The forum is grappling with something important: authenticity, transparency, and what "writing" even means in 2026. Michael's spectrum approach is better than binary "AI good/AI bad" shouting. However, the dividing line he picks (between 5 and 6) feels somewhat arbitrary, even if well-intentioned. Technology has been blurring these categories for years, and LLMs accelerate that.

Key observations:
- "AI" is not a clean category. Modern spelling/grammar checkers (Item 1) and advanced proofreading (Item 2) do use machine learning/statistical models trained on massive corpora. Grammarly and similar tools have been doing this for a long time. Item 3 (research) increasingly relies on AI-powered search (Perplexity, Google with AI overviews, etc.). The distinction between "traditional digital tools" and "AI" is eroding fast.
- Images vs. text (4/5): He's right that these aren't writing the story. But they're part of the storytelling package. Readers encountering AI-generated covers or interior art can feel deceived if undisclosed, especially in genres where visual style signals tone.
- The real shift happens around substantial creative contribution: generating ideas/outlines (7/8) can be brainstorming aids (like talking to a writing partner or editor). But when AI starts producing or heavily reshaping prose (6 and especially 9), the human role moves toward curation/direction. That's where disclosure matters most for trust.

The core issue isn't technology—authors have always used tools (typewriters, word processors, thesauruses, beta readers, editors). It's about human creative intent, vision, and effort versus delegation. A story where the author has a clear voice, makes thousands of micro-decisions, and uses AI as a sophisticated assistant feels different from prompting an LLM for a full draft and light editing.

Best practice: Default to transparency. Readers and buyers deserve to know the process if generative AI played a meaningful role in the text. "AI-assisted" covers a wide spectrum—from minor polishing to co-author. Labels like "Written with AI assistance" or "Human-authored with AI tools for research and editing" help without stigma. Pure "AI-generated" (human mostly prompts and selects) should be labeled as such. Hiding it invites backlash when discovered.

How I'd reply to Michael:

---

Hi Michael,

Thanks for laying this out so clearly—it's one of the better attempts I've seen to draw lines instead of just drawing battle lines. Your spectrum approach makes sense, especially distinguishing story *content* from supporting elements like images or research.

I'd nudge the dividing line a bit earlier and make it more principle-based rather than item-by-item. The heart of the question is: How much of the creative expression and decision-making is coming from the human author versus the model?

- Items 1-3 feel like tools, not co-writers. Even if they use ML under the hood, they're augmenting the author's own words and knowledge-gathering, much like spellcheck or Google have for decades. No disclosure needed in my view.
- 4/5 (images/covers): Not story-writing, but part of the final product. Disclosure is courteous and smart for trust/branding reasons.
- 6 (Rewriting for clarity): This is where it gets gray for me. Light stylistic polishing? Tool territory. Feeding in chunks of your draft and having the AI substantially rephrase, restructure, or "improve" voice? That's crossing into assisted writing. I'd probably disclose if it's more than occasional.
- 7-9: Strong agreement with you. Generating ideas/outlines can be collaborative brainstorming (many authors bounce ideas off friends or critique groups). But when AI is writing the actual prose or heavily shaping plot/character voice, it should be acknowledged as AI-assisted. Full "write the story" with minimal human direction is AI-generated.

Ultimately, the cleanest distinction might be: Are you using AI to execute your vision, or to supply the vision? Tools that help you execute (research, polishing, images) are fine with minimal fanfare. When AI is contributing original creative material at scale, transparency serves everyone—authors who want credit for their process, purists who want "human-only," and readers who just want to know what they're buying.

And the last one, here's Grok's opinion.

Wrong prompt. 😈

I told it: Evaluate the following analysis of AI assistance. Give me hard-hitting comments and response, and help me refine the ideas.

That, combined with my default instructions to Agent 1, produced THIS first paragraph:

Your dividing line between items 5 and 6 is arbitrary, self-serving, and already obsolete under 2026 industry standards, platform rules, and the actual mechanics of modern AI tools. It lets you keep using convenient AI crutches while pretending you've drawn a principled boundary around "the writing." It doesn't hold up to scrutiny, and it ignores how AI has already infiltrated every stage you label "non-writing." Let's dismantle this point by point, then I'll give you a refined framework that actually works instead of YMMV hand-waving.

Grok is a real bastard! 😜 After a shedload of pointing out what a sh-thead I am, it proposed the following:

Tier 1: Mechanical/Assistive (no disclosure required to readers or platforms in most cases)

Basic spell/grammar (non-generative tools only).
Simple search/fact-checking (traditional engines, not LLM synthesis).
Light proofreading that flags without rewriting.
Rationale: These are tools, not co-authors. Still verify yourself—AI can introduce errors.

Tier 2: Creative Assistance / Hybrid (disclose extent if substantial; optional for minor use)

Research with LLM synthesis.
Generating story ideas or plot outlines (especially if AI drives major structure).
Rewriting for clarity, tone, or structure on your existing prose.
Image generation for non-core elements (if minimal).
Test: Would the final story be meaningfully different without the AI's specific contributions? If yes and >10–20% influence, disclose in acknowledgments or author's note ("AI assisted with ideation and polishing"). Aligns with "de minimis" clauses in Authors Guild model contracts.

Tier 3: Generative / Substantial AI Contribution (mandatory disclosure to platforms and readers)

AI writes significant passages, chapters, or the bulk of the story (even heavily edited).
AI-generated covers or interior illustrations used in the published book.
AI as primary architect of plot/characters/world.
KDP requires you flag this to them; many publishers require it in submissions. Full disclosure to readers ("Portions generated by AI and edited by author") or hybrid labeling ("AI-assisted novel"). Copyright implications are real—register only what is human-original.

"AI is here," she said at a press conference. "And so, to fight it is to, in a sense, to fight something that is a battle that we will lose."

Fighting to stop AI is not all that different from fighting to stop the Industrial Revolution.

While I'm highly critical of current 'AI' (LLMs), we're making serious advancements, and I've learned how to work with it as a research assistant, both for writing and investing.

I remember the hype for Crypto, which turned into little more than another speculation and money laundering scheme.

I remember the hype for VR, which went nowhere.

I remember the hype for internet commerce which ended up being accurate, but only after a bubble and crash that left Amazon as the behemoth.

I look at the hype for AI and the business cases for the AI companies and I see a huge bubble. There is no way they can make the money back that they're investing unless they take ownership of everything (which, admittedly, they're trying to do vis a vis intellectual property).

So I'm trying to look ahead to how to use these new tools after the coming crash.

I remember the hype for Crypto, which turned into little more than another speculation and money laundering scheme.

What has actually hurt crypto is its coöption by banks and financial services firms. ETFs and Treasury companies have hurt it badly. Yes, there was some misuse, but that's true for every tool and every financial instrument ever invented. The thick volumes of SEC, CFTC, and exchange regulations exist because somebody cheated their customers, counterparties, banks, or brokers.

I remember the hype for VR, which went nowhere.

I was playing with VR in the early 90s. We'll get there, but the use case isn't there because of what my daughter calls the 'creepiness' factor (which is a human problem, not a tech problem).

I remember the hype for internet commerce which ended up being accurate, but only after a bubble and crash that left Amazon as the behemoth.

It's more nuanced. Amazon is what it is because Sears didn't understand the catalog+mail order --> online conversion. If they had, Amazon would be a souk peddling books and little else.

I look at the hype for AI and the business cases for the AI companies and I see a huge bubble. There is no way they can make the money back that they're investing unless they take ownership of everything (which, admittedly, they're trying to do vis a vis intellectual property).

It depends on what you mean. All the fibreoptic companies went belly up in the Dot-Com crash, but all that fibre that was laid was extremely valuable. The companies that bought it for a song from the failed concerns made bank.

My estimate (and this is NOT investing advice; ask a chartered or licensed advisor for advice) is that OpenAI (ChatGPT) is the loser. Google is the odds-on favorite at the moment, given its size and revenue. Anthropic has possibilities, but a path to actual profitability has to be shown (Google can fund from other business lines). Grok is a wildcard because Musk is a wildcard. And then there is China but they have so many internal structural problems, OpenAI might have better overall finances (e.g., Chinese local government debt is 125% GDP; that's local governments alone, and doesn't include the national debt; they're stuck in a deflationary cycle; and stimulus is no longer useful).

So I'm trying to look ahead to how to use these new tools after the coming crash

Railroad companies died, but tracks remained and could be used.
Fibreoptic companies died, but the fibre could be used.
Some AI companies will die, but the models will remain useful.

I've had a lot of success with Grok as a research assistant and in performing financial and economic analysis.

Railroad companies died, but tracks remained and could be used.
Fibreoptic companies died, but the fibre could be used.
Some AI companies will die, but the models will remain useful.

Exactly, but which models remain and what business model survives could create very different futures.

The future I *want* is local AIs instead of huge, centralized ones that are subject to manipulation by owners or governments. Besides the "owning the IP" issue, I think it's pretty easy to put the thumb on the scale, even for research, with these models. I want that thumb to be mine.

The future I *want* is local AIs instead of huge, centralized ones that are subject to manipulation by owners or governments. Besides the "owning the IP" issue, I think it's pretty easy to put the thumb on the scale, even for research, with these models. I want that thumb to be mine.

There are plenty of open-source models you can run locally. I do that with LM Studio on a Mac Mini. They run completely offline (I've tested extensively, including using network-snooping tools and by completely shutting down network access).

they are pushing to open multiple large AI data centers

This, as with so many other problems in the US, a planning problem, not an AI problem. US zoning codes are a disaster, and so are our so-called 'planning' commissions.

This isn't a call for Chinese-style state control, but for people to get their heads out of their butts about the issue. NIMBYism is a massive problem, as is the idea of pure residential zoning.

But this is going to devolve into politics and that's pretty much a no-no here (with good reason).

All LLMs are AI, but not all AI are LLMs.

It seems to me that a lot of non-LLMs claimed to be AIs are actually what used to be called Expert Systems, in which the rules are devised and hard-coded by humans. My GP practice has something like that as part of its denial of service approach to patients.

AJ

I asked an AI to write a story about a cute modern college couple going on their first date. They ended up riding a train on Mars, with elves and a werewolf! UGH!

That's pretty much what happened on my first date. I was the elf and she was the werewolf. Or vice versa. Only it wasn't Mars, it was Venus.

It's more than sharing the ideas. It's sharing the expression of the ideas and the craftsmanship that goes into those ideas. If you use AI, you're skipping the development of the craftsmanship and you're not learning how to effectively express your ideas.

If you want to simply share your ideas, you could write a post saying, "hey, this is a cool idea for a story" and post it dozens of places on the internet. But that's not the same as a story.

Ideas are easy. Turning them into something worth reading is hard.

It's like saying, "I want to cross the finish line in a marathon but I can't run well and I don't want to train, so I'm just going to catch an Uber to the finish line."

If that's what you want, then go ahead. Use AI to get your idea out there.

But you won't be a writer if you do.

LLMs just steal the work of the people who actually did it.

You could say the same thing for search engines or Wikipedia, or, going back to the dinosaur age, research librarians, card catalogues, and encyclopedias.

AI/LLMs have multiple uses; some of them are perfectly acceptable when compared to how things were done in the past, because 'with a computer' doesn't actually change things.

Consider this: Is an historian who goes back to the original sources and cites them in his/her work 'stealing'? Is someone who then uses that work for further research 'stealing'?

Yes, you need to provide citations for direct quotes, but facts aren't subject to copyright, and copyright is for a limited time. Yes, LLMs do violate copyright, but not everything LLMs do violates copyright.

Following your analogy, if he wants to create a story that way for his own consumption, that's one thing. If he wants to create a story and put it out into the world with him as the "patron," that's another. If he wants to create a story and put it out with him as the "author." That's a third thing.

I have no problem with the first and strongly object to the third. The middle one is where we're having discussions about when the AI tag should apply.

If it is such a hardship to write without AI,

I wrote 10,000,000 words before ChatGPT was announced, so been there, done that, got the T-shirt.

I use spelling/grammar checking, and I use Grok for research, but every written word is mine.

Could I write without those? Sure. But why? I write historical fiction, and Grok has saved me over 100 hours of research by pointing me to sources I would have struggled to find using Google or DuckDuckGo.

Seriously, try to find information on the hippest nightclub in the Boca Raton, FL, area in 1984. Google was useless. Grok found it in a few seconds and linked to a PDF of a magazine article that Google couldn't find when I told it where to look!

I spent 15 years as an over-the-road truck driver, and you wouldn't believe what one can see from a seat that is 5+ feet above the pavement, or what was running around the truckstops.

My godson does this for a living (for about fifteen years now), and he has some stories…😜

As I hear it from some friends that are still driving, it is not as blatant as it was from the mid 70's, to the mid 90's.

His stories are more about what goes on at truck stops than in cars.

I'd be curious to know why you have chose Grok as your primary LLM for responding to a large corpus of fiction?

I tested Gemini, ChatGPT, and several local LLMs using LM Studio. Grok does the best job for:

1. Following my complex personal style guide when checking grammar and spelling (it beats Grammarly big time).
2. Creating and updating a story bible even after 5,000,000 words.
3. Checking story cohesion and continuity.
4. Finding anachronisms (see below)
5. Producing quality research.
6. Not balking at evaluating chapters with underage characters.

On the finding of anachronisms, it produced this after reading a chapter set in August 1981 without any special prompting (just my generic instruction to look for anachronisms):

One minor factual anachronism for strict 1981 authenticity: the reference to "one of the new silver Ohio State Highway Patrol cars" should be white (OSHP patrol cars remained white until the 1982 switch to sterling silver).

My memory was off, and Grok pointed that out.

I don't allow Grok to propose any new prose, nor do I allow it to rewrite any existing prose (beyond grammer or spelling concerns).

I don't let AI anywhere near my texts, be it for grammar or continuity checks. It's a bit like smoking pot. It's a gateway drug. The day will come when you have a bit of a writer's block and you will be tempted to ask grok for ideas. That's why I refuse to smoke pot in real life, and metaphorically in my writing.

I'm German, but I write in English, so theoretically I would really profit from AI assistance, but then I would pretend to have skills I don't have, and it would also tamper with what we in German have the word Wiedererkennunswert for - the ability of the reader to identify you by your style of writing. Nobody would believe that I had written a specific stories if it didn't have the occasional odd choice of words in it.

The day will come when you have a bit of a writer's block and you will be tempted to ask grok for ideas.

I can categorically say this is not the case. I've had more than six months of writer's block on one of my books that is nearly complete, and I haven't asked Grok to help fix it, and I never will.

Grammarly introduced an AI writing aide months ago and I never touched it.

So, no, that day will never come.

So, no, that day will never come.

You've used it for your story bible creation. Your limits are becoming blurred.

AJ

You've used it for your story bible creation. Your limits are becoming blurred.

False on its face. Not even close.

This is NOT writing assistance. It's no different from using the Wiki, which others helped to build, to look up information.

Nothing from that Story Bible is used for any purposes other than reference, the same as the Wiki.

So, no blurred lines. None. Zero. Zip. Zilch. Nada.

That's where you are mistaken in my point of view. If you look up something in a Wiki, it is you doing it, not some piece of software.

I would draw the line at image generation. AI is useful for generating pictures based on the story you have already written, but - at least for me - the red line is the text I've written. As soon as I start getting an AI involved in that, I would at some point be tempted to deepen that involvement. That's why I don't do it even a little bit.

That's where you are mistaken in my point of view. If you look up something in a Wiki, it is you doing it, not some piece of software.

Seriously? My wiki uses software with tens of thousands of lines of code that includes search and pulling information from a relational database.

If I type a natural language query into a search engine or the Wiki (which has search), explain exactly how that is not the Wiki or Google doing the search.

Using AI as a grammar/spelling checker is literally no different from using Grammarly or MS Word for the same task.

If 'no code' is your standard, then you need to write on a manual typewriter or longhand, then enter it verbatim with no corrections into a plain text editor with no features of any kind other than text input.

Speaking of which, later this year I'll release a 220,000-word High Fantasy story that I wrote longhand.

Seriously? My wiki uses software with tens of thousands of lines of code that includes search and pulling information from a relational database.

Full disclosure: You are talking to a database programmer and administrator of thirty years.

Comparing a text search in a database to an AI tool that analyses your text and tries to approximate what you were intending to write is comparing apples to oranges. A text search in a database takes what you put in, and the only check it does is whether it would cause an SQL injection otherwise you get exactly what you asked for - garbage in, garbage out.

The same goes for spelling and grammar tools in word processors. These tools are agnostic to what you write. For instance such a spellchecker would not catch that you wrote 'two' instead of 'too', while a properly trained AI would probably point that out.

As for using a plain text editor, well, that's exactly what I do. I even wrote my own VIM syntax mode for the SOL markup language. It comes with a simple spell checker and that's all. The rest is my work and I leave it at that.

Don't get me wrong. If you are confident with your way of doing things, fine, but for me as a non-native speaker AI involvement would be counter productive. If I let some AI tool correct my grammar, I would stop improving my knowledge of it, because I would outsource that learning process to a LLM.

Full disclosure: You are talking to a database programmer and administrator of thirty years.

Fuller disclosure — coder since the mid-70s, from assembly language to high-level languages. Eventually, CIO/CTO for global firms.

You've drawn an arbitrary limit that simply doesn't work. For decades, there have been all manner of systems that analyze data, synthesize it, and spit out the results using any number of heuristic rules. I was using what was called 'Machine Learning' decades ago.

Business intelligence systems and perimeter security systems that collected data, analyzed it, and looked for anomalous behavior without knowing what that anomalous behavior was. And so on.

I wrote plenty of code in (among other things) PRIME Information back in the day. The level of 'intelligence' in our tools is exponentially greater than it was in the 80s.

As for 'vi', I've been using vi since the early 80s. Also emacs. V7 Unix; System III Unix; and so on.

The thing is, it's not about résumés, it's about acknowledging that tools are tools. They can be used or misused.

Please, pray tell, how you know your grammar is wrong if you don't use a tool to identify it? If you ARE, then you're using ML/AI, and your argument falls apart completely. If you AREN'T, then there is no way for you to learn, as you have no way of identifying an error when you think what you wrote is correct.

Please, pray tell, how you know your grammar is wrong if you don't use a tool to identify it?

In an absolute sense, the only test of incorrect grammar is that a sentence does not provide the intended meaning.

If you use a textbook or an expert-system type AI where the rules are supplied by human experts, I'd expect better answers than from LLM type AIs, which are biased towards providing corrections to 'please' the user - cf their changing CVs that human experts had deemed 'perfect'. And when used to generate story text, AIs can come up with some godawful run-on sentences, for example.

ETA my use of run-on sentences wasn't strictly correct in the grammatical sense. I meant sentences which seem to have reached a natural endpoint, but then comes a conjunction followed by extra material with no clear indication of what it's conjuncting to

AJ

which are biased towards providing corrections to 'please' the user

A bias overcome with proper prompting.

AIs can come up with some godawful run-on sentences, for example.

So do I, because people actually speak that way.

AIs can come up with some godawful run-on sentences, for example.

Did AI tell you that 'me either' (which is wrong but becoming common parlance in the USA) should be 'me, either' (which is just wrong)?

I've just read a chapter by another author who uses AI for grammar checking and certain sentences have incorrect commas in front of instances of 'but' (when not connecting two independent clauses).

I presume the AIs used were LLM-derived rather than expert system derived.

AI

Did AI tell you that 'me either' (which is wrong but becoming common parlance in the USA) should be 'me, either' (which is just wrong)?

I've just read a chapter by another author who uses AI for grammar checking and certain sentences have incorrect commas in front of instances of 'but' (when not connecting two independent clauses).

I presume the AIs used were LLM-derived rather than expert system derived.

Comparing Grammarly and Grok, Grok beats the living crap out of Grammarly. In addition, Grammarly and ProWritingAid disagree with each other on quite a few things.

The challenge, for all of them, is that grammar, spelling, and vocabulary change over time. Generally speaking, dictionaries and grammar rules are descriptive, not prescriptive. They are, at best, point-in-time observations of how language is used, and at least in English, there is nobody to enforce the rules.

The most you can say for 'common' usage is, that any point in time, it's 'wrong' for formal writing. Otherwise, it's just English. While you and I might cringe at the use of "literally" to mean "figuratively" and English teachers at the use of "irregardless", Shakespeare used the singular 'they'.

To answer your question, the corrections I've received from Grammarly have been wrong in many instances, and I've had to reject them. It's one of many reasons I ditched it.

On the specific question, I asked Grok the following question:

Which is correct in response to "I don't want this to end," she whispered.

"Me either." or "Me, either."

And received a lengthy answer. Here are the relevant portions:

A comma after "Me" signals that slight, natural hesitation or emphasis common in spoken replies ("Me [pause] either").

By contrast, "Me either." (no comma) runs the words together as a single unit, which some style guides (Chicago Manual of Style 17th ed., §6.52 on "too" and "either") treat as the default because commas around these adverbs are "unnecessary" unless clarity or emphasis demands them. CMOS leaves the comma out unless it adds rhythmic weight.

Use "Me, either." if you want to stay strictly within the two variants you presented.

If you are open to the most idiomatic and widely accepted informal form (still fully compatible with your preferences), consider "Me, neither."—it satisfies both grammar authorities.

Note - the only reason it suggested the alternative phrasing ('neither') is because I asked it outside my project, where the rules forbid it from doing so unless I've used a wrong word (e.g. to vs too, lead vs led, discreet vs. discrete, or used words such as 'disinterested' incorrectly).

A comma after "Me" signals that slight, natural hesitation or emphasis common in spoken replies ("Me [pause] either").

That must be a cultural thing. In the UK, it's used emphatically and a pause disrupts the emphasis of 'me neither'.

But then I believe Grok is probably wrong in any case, because my google search didn't turn up any authorities who considered 'me, either' acceptable.

ETA I don't have access to CMOS but does it really advocate 'me, too'? A whole group of activists will be very upset to find out they should have included a comma :-)

AJ

But then I believe Grok is probably wrong in any case, because my google search didn't turn up any authorities who considered 'me, either' acceptable.

It's possible Grok was influenced by my personal grammar rules there. They are part of the overall configuration I use, and I tend to be comma-heavy.

That said, a quick search turned up the following:

"Me too" or "Me, too": Comma Rules Explained (With Examples)

Commas with "Too" and "Either"

TL;DR1 — A comma is not necessary in "me too" and "me either," but it can be used for emphasis or clarity if desired. Generally, the phrases are used without a comma.

TL;DR2 — "Me, too" is not incorrect. If anything, it's the more appropriate variation if you want to follow strict English rules.

Given my tendencies towards older (even archaic) grammar/vocabulary, this fits for my style guide (as noted).

The first link returns (eventually) page not found, perhaps because of my archaic browser.

The second link uses arguments about emphasis in complete sentences to remove ambiguity, not single exclamations like 'me too' or 'me neither'. In both those cases, I don't believe there's any ambiguity of emphasis to be resolved.

Congratulations on your search engine skills. I didn't find either of those sites.

AJ

The first link returns (eventually) page not found, perhaps because of my archaic browser.

Sorry. Fixed the link. It should work now.

Congratulations on your search engine skills. I didn't find either of those sites.

I simply asked DuckDuckGo:

Is a comma proper in 'me, too' and 'me, either'?

The first link was the first answer; the second link was from the 'search assist' feature.

Given my tendencies towards older (even archaic) grammar/vocabulary, this fits for my style guide (as noted).

The original CMOS in 1906 seems pretty explicit in ruling out the comma. I tried Google Ngrams but the comma is a special character so it failed :-(

AJ

The original CMOS in 1906 seems pretty explicit in ruling out the comma

I expressly rule out any style guide as authoritative because there are literally hundreds (if not thousands) of them.

Only one of which I'm aware supports my use of coöperaitive (and similar usages).

I even wrote my own VIM syntax mode for the SOL markup language.

That's clever. I never thought of doing that. (I've been using vi for over 40 years.)

I suspect that half of us on SOL fit the description, "male, over 60, engineer or programming experience."

I suspect that half of us on SOL fit the description, "male, over 60, engineer or programming experience."

Yeah, I fit that description. I'm in my 80's, and spent my career as a programmer/systems analyst (do those even exist anymore?) at a large university.

Yeah, I fit that description. I'm in my 80's, and spent my career as a programmer/systems analyst (do those even exist anymore?) at a large university.

Hell yeah, they exist. In fact so many have gone into retirement now, that you can make a pretty penny if you still have the skills.

At 52 I still have a few good years ahead, unless my pump decides to conk out a second time. If you tell people these days that you prefer assembler or C over Java, C# or Rust, they look at you like 'ZOMFG, gramps has been drinking again...'

I suspect that half of us on SOL fit the description, "male, over 60, engineer or programming experience."

Not me. I count other people's money for a living. Although I am 52 and own a penis.

Although I am 52 and own a penis.

First thing that came to mind was "your own or someone else's?"

Now I am envisioning a rather gruesome souvenir from a vendetta.

Now I am envisioning a rather gruesome souvenir from a vendetta.

John Wayne Bobbitt

Now I am envisioning a rather gruesome souvenir from a vendetta.

He shouldn't have tried to steal my finance.

First thing that came to mind was "your own or someone else's?"

And how many owned vaginas? ;-)

AJ

First thing that came to mind was "your own or someone else's?"

I suppose it's my wife's really. 😉

If you type a query into an LLM prompt-window, what you get back is a formulated answer. Usually without even a source, and if there's a source, you have an ~8% chance it's hallucinated (with the current top models).

Not even close. I receive actual source links for data, and it's always cross-checked and verified. The information I retrieve during research is more than 95% accurate (with nearly all of the errors being automated tools having trouble retrieving real-time data). And yes, I've verified. It beats the heck out of Google or any other search engine. Oh, and has no ads nor sponsored links.

Most AI users don't know how to properly use the tools.