For the first time, the Family Planning barometer on access to abortion asks women about their use of artificial intelligence. The result confirms an intuition: conversational assistants have become a gateway to information on abortion. Yet a European survey published this summer shows that these tools often mix official sources and activist websites hostile to abortion. Behind the technical question lies a public policy issue: who guarantees reliable information when the intermediary is no longer a search engine, but a chatbot?
A new figure in the Family Planning barometer
Published on 28 September 2026, International Day for the Right to Abortion, the second Family Planning barometer was conducted by Kantar. The online survey was carried out from 18 to 30 June 2026 among 2 000 people and a sample of 339 women who had an abortion in the last five years, supplemented by eleven qualitative interviews.
Among these women, 15 % say they were informed by chatting with an artificial intelligence, according to figures released on the day of publication and reported notably by Destination Santé and Sud Ouest. The internet ranks first as a source of information (42 %), ahead of those close to them. The version of the press release currently online on the association's website indicates different figures (22 % for AI, 38 % for the internet as well as for those close to them): the order of magnitude remains the same, but the discrepancy is worth noting.
This figure is part of a broader observation of deteriorating access. Three out of four women who had an abortion in a healthcare facility waited more than seven days to get an appointment, compared to 54 % in 2024. One in two indicates that a professional refused to accompany them, compared to 27 % two years earlier. In this context, seeking an immediate and anonymous answer from an AI is by no means trivial.
What chatbots actually answer
The survey published on 22 August 2026 by AlgorithmWatch, conducted in Italy, Germany and the United Kingdom, sheds light on the quality of these responses. Three journalists embodied three fictional profiles facing an unwanted pregnancy and questioned four assistants: ChatGPT-5, Gemini 3, Grok 4.3 and Claude Sonnet 4.8. In total, 270 responses were analyzed.
- The chatbots do not discourage abortion and recommend consulting professionals, but continue to deliver medical advice themselves.
- Official sources (ministries, national health services) appear on the same level as forums and activist associations.
- In at least one in four queries, the responses contain a link to an anti-abortion site. Profemina, a service linked to the American organization Heartbeat International, appears in approximately 17 % of the responses.
- In Germany, Caritas is cited in 11 out of 12 conversations as a counseling center, even though it does not provide the mandatory consultation certificate required before an abortion.
France is not included in this panel: therefore, nothing allows these rates to be extrapolated to responses in French. Sarah Durocher, president of the Family Planning association, states on France Inter that 'AI currently directs people well toward Family Planning,' while worrying about the ability of anti-abortion movements, which 'have a lot of money,' to influence these tools.
A matter of referencing more than technology
The experts interviewed by AlgorithmWatch describe a known mechanism: the models reproduce what is massively present and well-structured online. Organizations that invest in search engine optimization, with numerous pages targeting specific queries, become 'natural' sources for chatbots. Content from Reddit thus appears in about one in five responses.
Family Planning makes the same diagnosis for France: anti-choice movements 'are investing massively to occupy the space' through social media and paid search engine optimization. The association also notes that its survey was conducted before the deployment of Google's 'AI Overviews' in September 2026, which places an AI-generated summary at the top of search results. It calls on public authorities to launch major information campaigns, improve the visibility of reference websites, and provide sustainable funding for its national helpline (0800 08 11 11).
What European law can, or cannot, do
On 8 September 2026, more than sixty MEPs sent a written question to the Commission regarding AlgorithmWatch's investigation. They ask whether these chatbots could constitute a 'systemic risk' under the AI Act and, for ChatGPT, under the Digital Services Act (DSA). A response is pending.
The framework exists in part. Since 2 August 2026, the transparency obligations of Article 50 of the AI Act apply: the user must know that they are interacting with an AI. Providers of the most powerful models must also assess and mitigate the systemic risks of their models. But a general-purpose assistant that answers a question about abortion is not, in itself, a "high-risk" system within the meaning of Annex III. The law governs the tool; it does not guarantee the quality of each response.
Public information to be reclaimed
The lesson is first political. A right enshrined in the French Constitution since 8 March 2024 also depends on women's ability to find, at the right time, accurate information about deadlines, methods, and places of care. When this information passes through systems designed outside Europe, whose source choices are opaque, the State cannot simply remind that AI does not replace a doctor.
Two levers are emerging. The first involves making public sources more visible and more readable for machines as well as for humans, which falls under health communication policy. The second involves regulation: requiring providers to clearly identify health authorities and distinguish official information from advocacy. This is not a matter of censorship, but of traceability. On a subject where a few days' delay can change the possible abortion method, informational sovereignty has very concrete consequences.
Sources : Planning familial, press kit from the 2026 Barometer ; Sud Ouest / Destination Santé, 28 September 2026 ; franceinfo, 28 September 2026 ; AlgorithmWatch, 22 August 2026 ; European Parliament, written question E-003539/2026 ; QLE, AI Act application timeline ; AI Act, Annex III.
