Author: Paperowl
11 Author has written for Marvel, 911, and Heated Rivalry.Introduction
You know the google doc that has AI proof? Yes, the one that was released literally yesterday
(Anon, 2026). I have beef with it and I am currently unemployed, bored, and possess the academic experience to redo it with (slightly more) rigor.
Major Issues I Have with the Paper:- It did not verify the Claude HTML artifact and only referenced Linkedin posts. If you're gonna base an entire paper on this artifact, maybe double-check it?
Sub-issue: we have zero detail about this artifact. How long has Claude been producing it? What removes the artifact? Does this artifact appear in languages other than English? Is there any possible way to get this artifact without using Claude?
- Methods?
"After filtering the “Game Changers | Heated Rivalry - All Media Types” tag, as well as other popular fandom tags (such as “Off Campus,” “The Pitt,” “Stranger Things (TV 2016)” and “Pluribus (2025)”) by either Word Count or by Most Recent"
So... which one is it? Word Count or Most Recent? Both? Why wasn't it consistent between all fandoms? Later in the paper it only says Word Count was used - why is Most Recent mentioned here? Typo? Why did we choose those two methods of sorting? My statistician brain assumes it's because those are probably the closest you get to randomization with Ao3's given sorting methods, but I want the authors to tell me that. At minimum where's my consistency.
"we viewed the source code for every fic on each tag’s first page."
That's 20 fics from each of these fandoms. It's a start, but it's small. Ideally you'd want minimum 50-100 fics as your sample size if you want results to be roughly representative (Central Limit Theorem, friends!).
"We then focused on the Heated Rivalry tag, expanding the search to almost every work over 1000 kudos and rumoured AI-assisted fics."
There are roughly 9,500 Heated Rivalry (HR) fics with more than 1,000 kudos, which is about a quarter of the total HR fics. That's not a bad sample size at all. The majority of fics identified in the appendix have more than 1000 kudos so we know they probably spent a lot of their time here. There is potential bias in the sorting method (kudos indicates popularity of a fic/perceived quality of a work; this may increase/decrease the presence of AI fics compared to a random sample). It's not a huge issue, but it should be noted.
Number of Fics in Paper Appendix by Kudos
| Kudos |
Number of Fics |
| 1000+ |
26 |
| 999-700 |
2 |
| 700-450 |
5* |
| 150 or less |
3 |
| None |
1 |
*One of these fics was written by an author with another fic in the Appendix that has over 1000 kudosGreat, that's a fine sample with fine parameters, we're done! But hang on, why are there fics in the appendix with 53, 70, and 119 kudos? Where did they come from? Hey Lynn, you say, these are the rumored AI fics. But rumored by who? Where did the rumors come from? There are more than 16,000 Heated Rivalry fics with less than 150 kudos - how were these three fics singled out? And, wait a second. There are 37 fics in the appendix. Nine have less than 1000 kudos (a quarter!). One has no kudos because it's not currently public (reminder: this paper was about untagged AI use and encourages proper tagging. This fic is not public and hasn't been since mid-march, if memory serves, so how does proper tagging apply here).
If the goal of your paper is to encourage transparency and you are anti-harassment (as this author claims to be) you need to be crystal clear about the sources for these rumored AI fics with less than 1000 kudos. Did you suspect AI use? Did your friend? Did your group chat? Maybe you saw a post on the timeline (which social media platform?). But
where did you find these fics? This is a quarter of your list, it's not something you can just gloss over.
- Have you noticed what's missing yet?
Remember when I mentioned verifying the Claude HTML artifact? Remember that this paper is about transparency in tagging because Ao3 has tags for this? We should be using those Ao3 tags to our advantage here - we can select a sample of only AI generated fics and see how often this Claude HTML artifact shows up. For whatever reason this was not done in Anon (2026).
- The focus of this paper was on Heated Rivalry only. Given the large sample size I get focusing on just one fandom. But the way it's presented is not my favorite.
Other fandoms were searched very lightly to find
examples from other fandoms that demonstrate the Heated Rivalry AO3 tag is not the only fandom tag where the fragment of code exists
. But these code fragments are equally unverified! This is where you should present works that are tagged as using AI. Not only does that help show the code fragments are related to AI use, but you'll also get fics from a variety of fandoms. Two birds, one stone.
We dragged a bunch of other fandoms into this paper to make a point, but what point? We barely looked at them. Plus, the title of the paper is "Fandom has a Hidden Generative AI Problem" which implies multiple fandoms, not just one. So, let's look at multiple fandoms.
There are other minor things to note, the paper really doesn't mention frequency of the Claude HTML artifact (some results had it twice, others had it hundreds of times) or give insight into how each author used Claude (was it for prose, code, or both?). Technically these are second study questions, meant for once you've established the presence of the HTML artifact and have demonstrated its prevalence in fandom. However, by mentioning specific fics by name and including the number of HTML artifact instances, the paper has made me more curious about answering these questions than I would be otherwise.
Overall, the methodology here is shaky. The authors did not verify Claude produces this artifact, they did not search for the artifact in works tagged as using AI, they did not clearly describe their methods, they made exceptions for fics outside of their perfectly appropriate sample size without explanation, even including a fic which hasn't been publicly available for months.In an attempt to address some of these issues, here's the rest of my paper.
Methods
2.1 - Claude HTML Artifact & AI Tagging
I know I harped on about digging into the Claude HTML artifact, but I didn't do that. If this was a real paper I would, but to be so real I am not falling into an AI research hole for someone else. Now, luckily for us, someone else got curious and posted a
test fic generated by Claude! So it's like, half verified.
To identify the Claude artifact a modified version of the site skin provided by Anon (2026) was used. The site skin turns the screen red when the 'font-claude-response-body' HTML is found anywhere in the work (all fics were opened to the entire work). Presence of the artifact was verified by searching for 'claude' in the page html using inspect element. Additionally, to easily verify the presence/absence of AI-related tags, the site skin also highlighted tags with the following words (included partial words, non-case sensitive (aside from AI)): "AI", "Claude", "Grok", "Gemini", "GPT", "GBT".
2.2 - Search Criteria
For consistency the same search criteria were used for each fandom/tag. Works had to be written in English (my convenience, plus lack of info on non-english Claude HTML) and posted between October 9, 2025 (Heated Rivalry trailer release date; Claude surged in popularity early 2026) and June 30, 2026. For each fandom the search results were sorted by kudos, and the top 100 fics examined for the Claude HTML and AI-related tags.
[Note: would love to have done more fics per fandom, but I'm only one person with limited time and also ao3 flagged me as a bot at the end of my hunt lol. As for why I sorted by kudos, that's how the Heated Rivalry sample from the original paper was determined. Ideally I'd look at fics from a random sample, which would probably be determined by selecting fics (i.e. the first on each page) from different kudos/word count/date segments - like 100 fics from those with 200-300 kudos, and 100 fics from those with less than 50 kudos, and 100 fics from those with 1234-1570 kudos, etc.]2.3 - Fandoms/Tags
Verification of the Claude artifact was done on works (any fandom) tagged as 'Created Using Generative AI' (the most commonly used AI tag on Ao3) that explicitly mentioned Claude.
[Why didn't I also look at all fics tagged with 'Created Using Generative AI' regardless of Claude mention? Well, there were more than 3,000 posted during the search period, and given results on fics with a Claude mention if felt like the work to review some/all of them wouldn't give much additional insight. To be thorough, this should be done at some point though.]Aside from Heated Rivalry, eight other fandoms were included. Five were chosen for their large and longstanding popularity (top of their respective fandom section on Ao3): My Hero Academia, Harry Potter, Good Omens, K-pop, and Marvel. The other three, The Pitt, Stranger Things, and 911, were chosen for their similarity to Heated Rivalry: TV shows that aired all of a season during the search period and have garnered popularity on social media (twitter/tiktok/reddit).
[If this were a real paper I'd need to give a tangible reason for each fandom being "popular". But this isn't a real paper, so I'm considering Marvel, HP, Good Omens, Kpop, and MHA's popularity to be common knowledge. The Pitt, ST, and 911's are anecdotal, but you guys probably know what I'm talking about here too. I do have some supporting numbers in the fun facts!]Results
3.1 - AI-Tagged Fic
Alright, what everyone is here for! Probably.
There were 34 fics within the set search criteria that were tagged 'Created Using Generative AI' and that explicitly mentioned Claude. Two of the 34 were excluded; one for not containing AI-generated text explicitly produced by Claude, and the other is our test fic from earlier.
Of the 32 remaining fics, a grand total of two contained the Claude HTML artifact! Yeah. That's right. Two. Fic #1 was posted in June 2026, and Fic #2 was posted in April 2026.
[Warning! Don't generalize these numbers! I'm gonna do it in like 5 paragraphs but the sample size is small, the sample populations are totally different, and there's bias all over!]
3.2 - Fandoms
Number of Fics with the Claude HTML Artifact in the Top 100 Works by Kudos
| Fandom |
Fics w/ Claude HTML |
AI-Tagged Fics |
| 911 |
0 |
0 |
| Good Omens |
0 |
0 |
| Harry Potter |
1 |
0 |
| Heated Rivalry* |
0 |
0 |
| K-pop |
0 |
0 |
| My Hero Academia |
0 |
0 |
| Marvel |
0 |
0 |
| The Pitt* |
0 |
0 |
| Stranger Things* |
2 (same author) |
0 |
*Bias note: these fandoms were mentioned in Anon (2026), so authors may have edited/removed works that would have otherwise contained the Claude HTML artifact.Discussion
Examination of fics tagged as being created by AI revealed that the HTML artifact is pretty lousy (2/32 = 6.25%) at identifying Claude-created work. Basically this means that we can't draw meaningful conclusions from data related to the HTML artifact. It may, at times, give us insight into patterns of use, but it's really not some crazy AI kryptonite.
As for specific fandom Claude usage as indicated by the HTML artifact? Pretty sparse. Of course, as was just demonstrated, the HTML artifact is mildly helpful at best. But we're not being overrun by AI fics either. And maybe kudos self-selects and removes the AI works, who knows. Regardless, these numbers show fandom AI use seems to be pretty similar, there isn't one fandom that's markedly worse than the others. For more generalizeable numbers I'd love to look into the top 1000 in the future, or maybe figure out a solid randomization method so I'm picking fics from all over the fandom.
Now here's the generalization I warned everyone else not to make: It's really easy and really tempting to say that for every 2 fics with the HTML artifact there are at least 30 additional unidentified AI fics. I am going to make a math point with this generalization and then you're gonna forget I did this because I promise you it's super statistically unsound.
Lynn! Fandom is being overrun by AI! What do you mean it doesn't seem that bad! Let's use Anon (2026)'s numbers (but rounded for nice math) to investigate this. Assuming they looked at every fic above 1k kudos (9.5k), they found 26 fics with the HTML artifact. If it's 2 identified for 30 unidentified, that means there are approximately 390 unidentified Claude fics in that 9.5k. If we break even more stats rules and times it by 4 to try to get numbers for the whole fandom, that puts us at 1,560 unidentified fics + 104 identified fics, which we'll round to 1,700 for wiggle room. That's 4.25% of all Heated Rivalry fics. Friends, 5% is where they put statistical significance because it's a pretty small percentage, 5% is not that big of an issue.
Okay, but what about all the other AI models? Yeah, what about it. Double it - make it 10% (actually 8.5% but we're exaggerating. Apparently Claude is really good at prose compared to other models, which is why I only doubled). Do I like roughly 10% of a fandom I read being untagged for AI use? Not really. Is it worth losing my mind over? Also no. 90% of fics in my fandom are still made by humans! That's a lot! AI advertising would have you thinking the 90/10 split is flipped the other way, but it's not. There's work to be done, but the majority of fandom, even with our exaggerated AI numbers, is still creating with their own two hands!
Conclusion
GUYSSSS. What are we actually doing. No, seriously. What are we doing? If we make lists of "proof" (no matter how legit the proof is) and harass authors who we suspect of using AI - do you really think they'll magically decide to tag their work? No. Why would they when AI tags = losing the respect of your friends and readers and more harassment. And before you say it, authors who appropriately tag their AI use get harassed too! You guys remember our test fic from earlier? The one that explicitly says it's a test fic? It got two hate comments within several hours of posting.
It is worth having discussions about AI use in fic. It is worth investigating the overall presence of Claude HTML artifacts and fragments of AI prompts in fic so we have an idea of the scope of AI use on Ao3. It is not worth it or useful to publicly single out specific authors or fics for AI use. It's just not. You want to encourage appropriately tagged AI fics? Be chill about AI use. Don't harass people, don't go out of your way to be mean about it (be mean on your blog if you want, but don't go hunting around to be mean on other people's blogs), and if you suspect a fic you're reading is made with AI and you're uncomfortable with it? Mute the author, forget about the fic, and MOVE ON.
Encourage writing fic for the joy of it, encourage being "bad" at writing fic, write comments, make friends, post your own human-made work (even if you think it's awful), and build community! Fandom should be for fun! Yeah, AI use is something to discuss, but the way the fandom at large has decided to go about dealing with it is counter-productive and will fix nothing, so go have fun for a bit. AI will still be here when you get back, and nothing you do right now is going to fix it, so take a breather.
Fun Facts!
(they don't let you do this in real journal papers, but they should!)
Fics posted in the search time period (rounded):
Good Omens - 7k (6k for Aziraphale/Crowley - it's the top of the Theater fandom category by total fics (97k) but popularity is clearly waning)
Heated Rivalry - 40k (37k for Shane/Ilya)
Harry Potter - 68k (9k for Sirius Black/Remus Lupin)
K-pop - 72k (5k for Han Jisung/Lee Minho)
My Hero Academia - 42k (8k for Bakugou Katsuki/Midoriya Izuku)
Marvel - 45k (2.4k for Bucky/Steve)
Stranger Things - 45k (25k for Mike/Will)
The Pitt - 22k (5k for Robby/Whitaker)
911 - 14k (10k for Buck/Eddie)
Heated Rivalry is That girl. The search period covers the entirety of the TV fandom time period and somehow the number of fics posted is on par with MHA, Marvel, and Stranger Things??
Popularity note: You know that list of the most popular pairings on Ao3 that gets posted at the end of the year? (I trust you know this, I am not digging it up for a link). Hollanov was in like the top 10 last year because it had 10k or so fics, after two months mind you. Now look at the number of fics posted in the other "newer" fandoms. All of these fandoms are doing Numbers.