Mumsnet denies writing posts with AI after reply prompt appears on forum

A tool meant to suggest Mumsnet thread titles instead produced instructions for writing an entire forum reply, according to founder Justine Roberts. She denied that the site uses AI to write posts. But the incident, reported by the Guardian on October 7, also drew attention to the tool’s handling of users’ unfinished drafts.

Advice instructions where a title should be

The episode involved AmIAComplete, a user seeking a “sanity check” about an awkward family interaction involving a neurodivergent uncle and a sick child. Instead of advice from another member, a prompt titled “AIBU Response Schema v.1.0” appeared in response. AIBU is the forum’s shorthand for “am I being unreasonable”.

The instructions asked an AI to produce an “interpersonal reasonableness” post, with a clear forum-style ruling and brief actionable advice. They defined common Mumsnet abbreviations and included safeguards: do not diagnose anyone, demean neurodivergent people or imply that a sick child owes others a social performance. Compassion, boundaries and de-escalation were priorities.

Mumsnet’s explanation changed. It first linked the prompt to its OpenAI-powered title-suggestion tool while denying that AI wrote threads. It then said the prompt had not come from Mumsnet and suggested it might have transferred from the user’s computer. Roberts subsequently confirmed that it came from the site’s system.

Some users copied the prompt into an AI tool and found that it generated a full post, the Guardian reported. That fueled demands for a clearer explanation. One member, MrsBennetsPoorNervesAreBack, wrote: “We are not idiots,” and asked Mumsnet to explain transparently what had happened.

Roberts draws a line around AI-written replies

Roberts said the system sends users’ drafts to OpenAI to obtain a suggested title. This time, it returned a prompt for a whole reply. She described it as something that “looks like an error on their side”—her explanation of the failure, rather than a confirmed technical diagnosis from OpenAI.

That is not a claim that Mumsnet operates without AI. Roberts said it uses the technology for coding, advertising operations, conversational analysis and flagging potentially problematic posts to moderators. The distinction she drew was between those supporting tasks and generating the discussions members come to read.

Unfinished drafts become a separate concern

In a related Mumsnet discussion begun on October 5, commenters focused on drafts rather than generated replies. Several said they had not understood that unpublished text was being sent to OpenAI for title suggestions. Their concern was about material they might revise or delete before choosing to make it public.

One commenter, BoobsOnTheMoon, said the discovery had put them off starting threads. They described changing details or deleting a draft because it felt too revealing or emotional. Another, 5128gap, said they could understand the title tool’s purpose but considered the lack of transparency the problem.

Some participants went further, alleging that drafts were transmitted in real time, including edits and deletions. Those descriptions are user claims, not independently verified technical findings. Roberts confirmed the draft-to-OpenAI workflow, but her account did not establish the commenters’ more specific claims about every keystroke being transmitted.