Policy Update AI Editing Update

[JamesMartin]

Evening Sanctum,

We've had a couple of members ask questions regarding this in the last month, and I figured it made more sense to answer in one place than separately.

Especially since, upon reflection, we think this policy needs some added safeguards.

First, for those of you who use no AI at all, a quick disclaimer so you can skip the details which don't apply to you: Normal checks with Grammarly are not AI (their rewrite options are), spelling and grammar checkers on Google Docs or Microsoft Word are not AI, nor are the built-in ones that come with browsers. Google Translate is also not AI. None of what follows applies to any of these, at present (this might change in the future if these services try to force AI in, but we can hope for the best).

Now, for the actual policy.

1. As our rules already state, AI is not allowed, under any circumstances, to write posts for you.

2. We allow the use of AI tools for checking grammar, for translation and for basic editing

3. The use of these tools must be selective. Asking it to check for grammar errors is fine, asking it to reword a sentence or two you can't make work is fine. Putting an entire post of several hundred words into it and saying "rewrite this" is not. At that point, it is no longer an aid to improve your writing; it's just doing the writing for you.

4. The big change: Because there is very little way to tell the difference between an AI-written post and a sufficently AI-edited one, we've decided that we need to introduce a requirement that for any post where AI has been used in a meaningful way (meaning it has edited more than 25% of the total word count), you must keep a copy of the unedited version available for at least 3 months. How and where is up to you. You can put it in Google Docs, you can put it in your personal notepad here on site (yes, this feature exists), you can put it in a separate thread or anywhere else you care to name. What matters is that it is available to demonstrate that there was an original, pre-editing work that was your own.

5. We strongly encourage (but will not currently require) putting the unedited version of roleplay posts in a spoiler below the posted version. Right now, we feel like this is a fair option for people who want to use these tools but still hold themselves accountable.

6. Event posts and Challenge entries will be checked as a matter of course. While they are at little risk of winning, given the poor quality of AI writing, it's still fundamentally unfair to award the same badges to AI entries as to those who participated and put effort in

7. These changes are not retroactive. They will apply only to things posted from tomorrow onwards.

As a personal aside, beyond the policy changes: For those who do want to use these tools for grammar and rewording, while I understand the impulse, as someone who has been writing most of his life and running this site for nearly a decade, I strongly recommend learning from the limitations of these tools. They are, without exception, bland and uninteresting writing that obeys the technicalities of grammar with no understanding of how to work within those rules to create something greater than the sum of its parts. If your goal is to become better (and I sincerely hope that is the goal of anyone who writes on this site), I would suggest that, whatever your use of AI, you take what it gives you as a bland, technically competent baseline and improve on it from there.

Spoiler: And to emphasize the point, here is the rewrite of what I just wrote with AI

As a personal note separate from any policy updates: if you’re using these tools for grammar fixes or rewording, I get why—they’re convenient. But as someone who’s spent most of his life writing and nearly a decade running this site, I’d urge you to pay attention to what these tools can’t do. Their output is consistently flat, technically correct but devoid of the nuance, rhythm, and intention that make writing memorable. They follow rules without understanding how to bend or combine them to create something with real voice.

If your aim is to grow as a writer—and I genuinely hope that’s true for anyone posting here—treat AI output as a neutral, competent starting point. Then push past it. Shape it into something with personality, texture, and thought. That’s where the actual improvement happens.

This will hopefully be our last word on the subject for a while, but the tech landscape is changing and we might need to update our rules further in the future. Unfortunately, this is very much a situation where a few bad apples will spoil the barrel.

As always, questions, comments, concerns and the various ways in which you wish to #BlameJames can be posted below.

 

[Scribblez]

Question and clarification of point #4 with regards to using AI to transcribe pages of handwritten drafting into a typed Word document.

Technically, if the AI program is calibrated correctly by your input commands, it should be a 1:1 ratio, with no, to limited minor, errors due to sloppy handwriting.

Would that be considered using AI to alter written work by 25% and therefore need a proof of originality to be posted?

I don’t want to out myself, but I do use AI in a transcription function to streamline my weekend editing sessions once I finish with handwritten draft #X.

 

[JamesMartin]

It wouldn't be, because the AI isn't writing, it's just transcribing words you already wrote. That would never appear as AI writing because the AI isn't changing the words.

 

[Hawkwood]

I'm maybe over-lawyer-y here but:

"you must keep a copy of the unedited version available for at least 3 months"

I assume the purpose of this is that if I, as the co-writer, think "this sounds like co-pilot; look at all those em-dashes" I can ask for proof? My only worry is that if I were being devious, I'd just provide the AI draft again.

And as a side note: speaking as someone who was trained to write in a bland, technically competent baseline for my profession (lawyer), I've spent the time since desperately trying to unlearn that style! It's tedious and horrible! Ahem.

 

[Semi-Sweet]

What about the use of AI-generated images? I may have missed an earlier post about it, but I can't find it.

Basically, my question is whether we need to disclose somehow that an image is AI.

 

[JamesMartin]

I assume the purpose of this is that if I, as the co-writer, think "this sounds like co-pilot; look at all those em-dashes" I can ask for proof? My only worry is that if I were being devious, I'd just provide the AI draft again.

In principle, it's so that staff could check, but yes, it would also provide a means for partners who feel like they are writing against AI to ask for proof.

The main idea would be that we would check the new draft for AI flags and run it through AI checkers (these are flawed, but can be instructive).

In theory, someone could fabricate the draft after the fact—but in practice, this seems unlikely. Someone who is on a writing site and getting AI to write for them is unlikely to also be willing to take the effort to credibly forge a draft. Especially since we would, if things were bad enough, ask for drafts of multiple posts. It also runs into the fact that well, we're all writers (I can hear my partners laughing at me saying that) The biggest tell of AI writing is that it has no consistent voice. There are no idioms, turns of phrase or odd quirks that will tell you who wrote something.

What about the use of AI-generated images? I may have missed an earlier post about it, but I can't find it.

Basically, my question is whether we need to disclose somehow that an image is AI.

Right. We did announce a policy on that, but it was a long time ago.

Basically, we're not an art site. We're not going to ban AI images in general, because it's just not relevant.

However, as with using AI for editing, we do require disclosure. Using an AI image for a character is fine, claiming that you're the one who drew it is not. If you're just using the image, no disclaimer is needed; you just can't claim it's yours.

 

[dramatis personae]

What matters is that it is available to demonstrate that there was an original, pre-editing work that was your own.

I have a question about the practical side of this. What happens if someone is accused of using AI when they haven’t? I have seen all kinds of "ways to figure out if your writing partner is using AI" comments on the internet, and most of it is nonsense. People are starting to vilify em-dashes and proper punctuation, or really thoroughly edited prose. At what point do those of us who value delivering something as close as possible to technical perfection (hello hyper-fixation) have to alter our writing just to prove our humanity?

Maybe I just haven't stumbled upon the obvious AI writing I've read rants about on Reddit (posts that lack the basic acknowledge/react/respond/add structure it seems) but even then, isn't it about the ability to recognize it and step away from such a RP partner if you find it unsatisfactory?

My understanding of the culture here thus far has been that we’re adults, we choose partners whose style we enjoy, and if something doesn’t click, we quietly move on. The moment such an opinion or misunderstanding can escalate into a rule-breaking report to the administration and then lead to having to prove you're not an elephant...That scenario kinda stresses me out to think about to be honest.

 

[JamesMartin]

I have a question about the practical side of this. What happens if someone is accused of using AI when they haven’t? I have seen all kinds of "ways to figure out if your writing partner is using AI" comments on the internet, and most of it is nonsense. People are starting to vilify em-dashes and proper punctuation, or really thoroughly edited prose. At what point do those of us who value delivering something as close as possible to technical perfection (hello hyper-fixation) have to alter our writing just to prove our humanity?

It's a fair concern, but the thing I want to be clear on is we are not going to be throwing accusations around lightly or based on quirks that AI has stolen. AI can pry my em-dashes from my cold, dead hands.

In my experience, you're right, the guides you'd get online are nonsense—but we're not using those guides, we're not assuming someone is an AI because they know how semi-colons work. We're also advantaged because, since we are a forum, we have full access to someone's writing history on-site. We don't just need to decide if one specific post has the hallmarks of AI; we can also look at their other work and see if there is a specific voice they maintain or quirks of their writing that are consistent across months. AI models update so frequently and are so inconsistent that the result is technically competent posts that read like they were written by entirely different authors. They also make large errors in continuity that no human writer would make, typically because they are writing a response to a single post without even having access to the full story.

There are other indicators as well—but I'll be deliberately vague here and just say that we have ways other than the punctuation or grammar of a post to figure out if someone is having an AI write for them.

Maybe I just haven't stumbled upon the obvious AI writing I've read rants about on Reddit (posts that lack the basic acknowledge/react/respond/add structure it seems) but even then, isn't it about the ability to recognize it and step away from such a RP partner if you find it unsatisfactory?

My understanding of the culture here thus far has been that we’re adults, we choose partners whose style we enjoy, and if something doesn’t click, we quietly move on. The moment such an opinion or misunderstanding can escalate into a rule-breaking report to the administration and then lead to having to prove you're not an elephant...That scenario kinda stresses me out to think about to be honest.

Odds are, if we get an AI report based on a single roleplay, the person being reported will never even know about it. We'll look into it, triple-check and only act if we're certain beyond any doubt.

Part of why this is needed, is site events and the writing challenges. We've caught at least two AI challenge entries that I recall just since 2024. And we had an AI image incident in the Secret Santa. Those are fundamentally different from a roleplay, because it can't just be dropped—it actively makes the event less fun and makes people less likely to participate if they get the same reward as the person who put the prompt in Copilot and said "write something."

 

[Oreodaqueen]

Thank you for the clarification on the rules, James! It makes sense. I personally do not use AI but always undestanding boundaries of the site, rules, and my partners is important. As such, I very much love the common sense approach and look forward to future writing ventures on here with my inner sanctum friends!