Mumsnet Message Board AI Instructions: The Leaked Prompt

The Mumsnet message board AI instructions were never meant to be seen. On a thread opened on October 4, 2026, a detailed prompt for an AI model appeared where human replies should be. Mumsnet insists it does not use AI to write posts. However, its explanation changed three times, and many members are not convinced.

The thread sat on the site’s famous “Am I being unreasonable?” board. A member called AmIAComplete described a family row involving a sick daughter and a neurodivergent nephew. She wanted honest opinions from other parents.

Instead, a block of text titled “AIBU Response Schema v.1.0” showed up, according to The Guardian. It read like a brief for a machine. In short, it told an AI to judge “interpersonal reasonableness” and deliver a forum-style ruling with short advice.

Members spotted it quickly. Some called it “really creepy”, and others said it was “so disappointing”. Meanwhile, the original thread kept growing and passed 100 replies.

The prompt was written for Mumsnet specifically. It explained house abbreviations such as AIBU, ND for neurodivergent and DD for dear daughter. Therefore, whoever wrote it knew the forum well.

It also set guard rails. The model should not diagnose anyone, and it should not demean neurodivergent people. In addition, it had to put compassion, boundaries and de-escalation first.

That detail is what unsettled people. A title generator does not need rules about tone and diagnosis. A reply writer does. Because of this, several members pasted the text into a chatbot themselves, and they say it produced a full post.

Mumsnet’s account shifted as the thread went on. The Guardian lays out the sequence, and it is easier to follow in a table.

Founder Justine Roberts gave the third version herself. She said the system sends a member’s draft to OpenAI, which should return a title. This time, however, a prompt for a whole reply came back.

Her line was firm: “We don’t use AI to write threads or replies on Mumsnet.” She added that the glitch “looks like an error on their side”. The Guardian report carried no statement from OpenAI.

Yes, but not for posts, the company says. Roberts told The Guardian that Mumsnet uses AI for coding and advertising operations. It also uses it for conversational analysis and for flagging problem posts to moderators.

Furthermore, she said the company has never tested a tool that drafts posts. Its annual report does say it is “embedding AI tools” to lift productivity and user experience.

The title helper is the visible part. On September 13, a member asked on the AIBU board who was suggesting titles for new threads. Another member replied that the feature had been live for a couple of months.

The explanation raised a second problem. If a draft goes to OpenAI for a title, it leaves Mumsnet before the member presses post. As a result, people who type something personal and then delete it may have shared it anyway.

A discussion that began on October 6 warned members about exactly that. Posters there reported that Mumsnet had opted out of letting OpenAI use the drafts. They also said Mumsnet would pause the feature until it could be opt-in and transparent.

Those are members’ accounts of staff replies, so treat them with care. Still, one poster noted that the AI title option had already vanished from the posting form.

There is history here as well. Mumsnet previously launched a legal complaint against OpenAI over scraping of its content, as TechRadar reported. So the site now sends drafts to the firm it once accused.

Mumsnet began in 2000, and its whole value is advice from real parents. Roberts made the point herself. Members trust that answers come from other parents and not from a machine, she said.

That trust is under pressure. According to The Guardian, monthly unique users fell 19% in 2025, to 6.1 million. Posts dropped by 600,000 as well. Consequently, a rumour that staff seed threads with AI lands at a bad moment.

Members said they felt “fobbed off”. One wrote that they no longer knew what was real, and another talked about leaving. The paper also cites research in Nature, which found that readers see AI-assisted social posts as “robotic” and “generic”.

Mumsnet is not alone. Last month, shoppers found prompt text left inside a supermarket’s reply, which we covered in our Aldi AI customer service response story. In both cases, hidden instructions slipped into public view.

The lesson is similar too. Companies wire language models into small features, such as titles or email drafts. Then one odd output exposes the plumbing. Roberts admitted as much, saying these systems are less predictable than their sellers suggest.

For now, the Mumsnet message board AI instructions remain a single documented incident. Nobody has shown that staff post AI replies. Nevertheless, the site has to rebuild confidence one thread at a time.

If you want to check whether a post was written by a machine, read our GPTZero review first. To see how easily AI text gets disguised, browse our guide to the best AI humanizer tools.

Sources

The Guardian (via AOL), Mumsnet AIBU thread, Mumsnet members’ discussion on drafts, Mumsnet thread on AI titles, TechRadar. *Photos: Child draws while parent works on laptop at home by Shixart1985, CC BY 2.0, and Pioneer Building, San Francisco by HaeB, CC BY-SA 4.0, both cropped and resized.*