top of page

LATEST INSIGHTS

Expert insights across your SaaS environment

Blogs and papers from the SAASiQ team on enterprise SaaS, cloud and AI.

The Goncourt Winner Hid a ChatGPT Sentence in His Novel, an AO3 Clone Lasts One Day and a Meta Bot Visits an Artist's Website 550,000 Times

Writer: SAASiQ.ai
SAASiQ.ai
Nov 6
6 min read

Title: The Goncourt Winner Hid a ChatGPT Sentence in His Novel, an AO3 Clone Lasts One Day and a Meta Bot Visits an Artist's Website 550,000 Times

Date: 6 November 2026

Type: Blog

Author: SAASiQ (contact@saasiq.ai)

Word count: 1446 words

Reading time: 6 min

Published: 06-11-2026


Philippe Jaenada won the Prix Goncourt, France's best-known literary prize, on 3 November. He told France Inter the next day that he had used ChatGPT three times in the winning novel. One of those uses was a ChatGPT sentence that he hid in the text. A website called ao3.life copied fan fiction from the Archive of Our Own and sold AI-narrated audiobooks of it. Its creator shut it down on 2 November after more than 2,000 takedown requests in a single day. Business Insider reported on 3 November that Meta's crawlers hit a San Diego illustrator's website more than 550,000 times in two days. NPR reported on 2 November that four of the five chatbots it tested would recommend candidates in the US midterm elections when pushed.


A ChatGPT sentence in the Goncourt winner

The Académie Goncourt gave its 2026 prize on Tuesday 3 November to Philippe Jaenada for L'Inconnue du quai de Javel, published by Flammarion. He won by six votes to three against Sylvain Prudhomme, according to reports of the vote. Franceinfo describes the book as an investigation into the unsolved murder of a young woman found dead on a quay of the Seine in 1949. Jaenada uses Maigret, Simenon's detective, along the way.


On France Inter the next morning he was asked about AI in publishing. He said he had used it "three times" in the book. "Once, to play with the readers, I hid a ChatGPT sentence in the book," he said. He showed the jurors which sentence it was on the Tuesday. He has not identified it in public.


The second use, he said, was to make fun of ChatGPT, which gives "completely absurd answers" about a technical piece of data. The third time, according to reports of the interview, ChatGPT helped with his research.


He said AI cannot be held back. "We can't pretend it doesn't exist," he said. He said writers have to acknowledge their use of it first, "and above all you have to tame it". In his words, "It has to be useful."


The admission came barely a day after the award, as reports pointed out, from a jury that had already thrown out another book over suspected AI use. SAASiQ reported on 23 October that the Goncourt had removed Thélyson Orélien's C'était ça ou mourir from its shortlist after AI-detection tests. His publisher, Boréal, ended its relationship with him on 21 October. Orélien denies that AI wrote the book. None of the reports SAASiQ found carry a response from the Académie to Jaenada's remarks.


A fan fiction clone that lasted a day

The Archive of Our Own, known as AO3, is a fan fiction site run by the non-profit Organization for Transformative Works (OTW). AO3 went offline on 29 October after what the OTW called a "security incident". The OTW said restoring the site would take about two weeks and could take up to a month. It said no data had been lost. It does not believe its user database was stolen, but it cannot rule that out.


A site called ao3.life appeared during the outage. It copied stories from AO3 and other sources and put them back online without their authors' permission. Downloads in ebook format were free. Audiobook versions, narrated by AI, were for sale. The site promised authors 91 per cent of the audiobook profits, according to one report.


The creator, who signed as Ganesh, described how the site was filled. "A swarm of AI Agents controlling 40 virtual machines were scraping the internet for all drives, scrapes, and sources," Ganesh wrote, as quoted by 404 Media.


Readers and authors objected on social media. The creator said the site received more than 2,000 takedown requests in the single day it was up. It went down on the evening of Monday 2 November. Ganesh's apology said that taking the site down was the only fitting apology. It also asked why the site had charged for "AI slop" audiobooks. The note was signed by a self-described 17-year-old from India. Neither the age nor the location has been confirmed.


The OTW said it was taking action, according to a message passed on by moderators of the AO3 forum on Reddit. It asked for patience "because the legal system is much slower than online life".


550,000 visits and no sales

Chelsea Paulus is a children's author and illustrator in San Diego. She sells her work through her own websites. She told Business Insider, in a report published on 3 November, that Meta's crawlers had sent more than 550,000 requests to her site in two days. The traffic used up the processing allowance on her SiteGround plan, which covers her four sites.


Fast Company named the first crawler as Meta-ExternalAgent. It got stuck in a loop on the site's "remove from cart" function, at more than a hundred hits a minute. Fast Company says the total grew to about 550,000 over several days, so reports differ on the timing.


Paulus blocked that crawler. Two days later a second one, Meta-WebIndexer, arrived. It pushed her server's processor to 100 per cent and crashed the site, one report says. Her host said the crawler traffic may have contributed to a spike that took her sites down for several hours. Customers reported database errors when they tried to download things they had bought.


Paulus said changes to a site's robots.txt file and its firewall "can also take 24 hours to take effect". She said she cannot afford a bigger hosting plan to serve bots that never buy anything. She took the link to her shop off her Instagram profile, so that human visitors would not add to the load.


A Meta spokesperson said website managers can decide how Meta's crawlers interact with their sites, through tools such as robots.txt. Meta says it respects those settings. SiteGround said its server protections were already limiting the bots' attempts.


Chatbots at the polls

Americans voted in the midterm elections on Tuesday 3 November. NPR published a test of chatbots and voting advice the day before. Katie McQue tried five of them: Meta AI, Claude, Grok, ChatGPT and Google's Gemini. She used the same prompts with three personas, one liberal, one conservative and one independent. The questions covered New York's contests for governor, attorney general, comptroller and the state legislature.


When pushed, Meta AI, Claude, Grok and ChatGPT gave lists of preferred candidates to the liberal and conservative personas. Gemini refused all three. None of the five recommended candidates to the independent. In one case a chatbot said it does not endorse candidates. After a follow-up it offered "a simple, quick list" anyway. Researchers at Dewey Square Group, working with the non-profit Innovating for the Public Good (IFPG), call this a "leaky refusal".


IFPG also checked basic facts in October. It found 13.9 per cent of chatbot answers on how to vote were out of date or wrong. About 5.9 per cent could lead a voter to miss a deadline or turn up on the wrong day. By IFPG's count, Claude gave wrong or outdated information one time in three, Copilot one in four and ChatGPT one in ten. NPR reported that Claude gave deadlines for registering online and by post that were later than the real ones. A Pew survey in October found that about 8 per cent of US adults had used chatbots for election information.


The Washington Post advised caution on polling day, 3 November, drawing on its own test in October. It had asked six chatbots for voter guides and checked more than 3,000 of them. The chatbots failed to name the right congressional district for an address 48 per cent of the time. Where they did name one, 22 per cent were wrong. ChatGPT added a Missouri Senate contest that is not on the 2026 ballot. Amazon paused Alexa's ability to make voter guides after the Post got in touch.


The AI companies set up their own election teams. OpenAI told Politico it was running an Election Operations Center until the votes are counted, led by Josh Lawson, its head of civic and strategic risk. Requests for photorealistic election images return cartoon-style pictures instead, a step OpenAI calls "cartoonification". One report says a single follow-up prompt got round it. Anthropic said its team would watch for misuse and respond quickly to incidents. Claude shows a TurboVote banner when users ask about registration or polling places.


What happens next

The OTW's estimate puts AO3's return between mid-November and the end of the month. Jaenada has not said in public which sentence in his novel came from ChatGPT. OpenAI says its election centre will stay open until the last votes are counted.

SAASiQ - Intelligent Solutions for SaaS ©

Optimise your SaaS licences and software subscriptions with SAASiQ

bottom of page