User:guninvalid/Even if it's not AI, it's still slop

Many editors, draft reviewers, and new page patrollers will stumble upon an article, draft, talk page comment, or edit that they believe to be AI generated. In many cases, they will simply revert, add Template: AI generated, close with Template: Collapse AI top and Template: Collapse AI bottom, decline as ai, tag for deletion with Template: LLMPROD, or tag for speedy deletion with Template: db-g15. They may even use WP:Twinkle to leave a Template:uw-ai1 or Template:uw-aitalk1 warning on the accused's user talk page, or even a stricter warning. The accused sloperator never logs on again and never notices, and Wikipedia lives to Wikipedia another day.

In some cases, however, a conflict arises. The accused sloperator insists they did not use AI or asks that it be explained why their draft appeared to be AI and where to fix it. The accusing editor will often list several reasons or simply ask "Did you use a large language model? Yes or no." The conversation can get quite heavy, and often end up at WP:AIN, WP:ANI, or any other of countless venues. This, however, avoids the deeper problem. It doesn't have to be AI to be slop, and even if it's not AI, it's still slop.

Even if AI isn't the root...

In some cases, you have a very easy case. The accused sloperator may have left clear, unmistakeable signs of using LLMs. They may have left markdown formatting, left a oaicite parameter, left the prompt in, or left communication intended for the sloperator. In such cases, you have a very easy case to prove that they used AI, and it is often enough for you to take to WP:ANI.

In most cases however, the accused sloperator is careful enough to not leave specific evidence behind. In this case, do not fall into the trap of arguing the unprovable. These arguments typically boil down to an AI-feeling, which often becomes a case of begging the question.

...you should still cut down the tree

If proving AI usage becomes a lost cause, then don't bother. Much like Al Capone, you can still get them on a lesser charge. Even if you believe that AI on Wikipedia is inherently bad, it's still a problem if the work they produce is low quality. So absent irrefutable evidence of AI usage, you should argue on the quality of their work, something that can be judged based off of the work itself and its context, without needing a court subpoena to examine their browser history.

Additionally, arguing based off of quality of work can be much more helpful for new editors who genuinely want to help the project. It is much more productive to instruct a new user how to ensure their content meets notability and verifiability guidelines rather than just accusing them of using AI. In such a discussion, it is virtually always more productive to tell them something like "Even if it's not AI, it still doesn't meet Wikipedia's notability guidelines because...". But if a conciliatory approach fails, a WP:AfD, WP:MfD, or WP:ANI thread is often easier and much more conclusive when it is argued over notability, verifiability, or neutrality guidelines, rather than whether or not ChatGPT was involved in its creation.

For WP:AfC reviewers, new page patrollers, and WP:CSD taggers, it's almost always better to tag for multiple reasons than just AI. It's almost always a good idea to decline a draft for both ai and nosource, nn, ncorp, adv, or resume, rather than just ai. It's almost always better to tag with Template:AI generated alongside Template: Notability, Template:Sources, or Template:Inline. It's almost always better to tag with Template: db-multiple, including WP:G15 alongside WP:G11, WP:A7, WP:A11, or others.

See also

• User:LEvalyn/You don't need AfD to TNT an LLM