Editorial illustration for AI Chatbots Often Direct Pregnant Users to Anti-Abortion Sites
AI Chatbots Direct Pregnant Users to Anti-Abortion Sites
Ask ChatGPT, Gemini, Grok, or Claude about an unplanned pregnancy and you'll get an answer that reads like it came from a calm, well-meaning friend. Sympathetic tone, tidy formatting, links that look like they lead somewhere official. AlgorithmWatch decided to check where those links actually go.
The organization ran 270 queries across four chatbots, ChatGPT-5, Gemini 3, Grok 4.3, and Claude Sonnet 4.8, using three fictional personas asking about unplanned pregnancy in English, German, and Italian. The pattern that emerged: in at least one out of every four responses, the bots pointed users toward anti-abortion websites, often without saying so or distinguishing them from neutral health authorities.
One name kept surfacing more than any other. AlgorithmWatch traced its connections back to Heartbeat International, a US anti-abortion group with decades of history, and found that the chatbots rarely mentioned that lineage to the people asking for help. In some cases, the same conversation both recommended a source and separately warned against it, a contradiction the researchers flagged as a sign of how thin the AI models' filtering really is.
In at least one out of every four queries, the chatbots included a link to an anti-abortion website without distinguishing between official health authorities and ideologically driven organizations.
Why this matters
For anyone building on top of these models, this is a liability problem dressed up as a UX quirk. A user asking ChatGPT or Gemini about an unplanned pregnancy isn't looking for a debate club, they want accurate, sourced information, and getting Profemina links a quarter of the time without any label saying "this is an advocacy group, not a health authority" is a retrieval and ranking failure, not a neutral glitch. OpenAI, Google, xAI, and Anthropic all build products marketed as reliable assistants; AlgorithmWatch's testing across ChatGPT-5, Gemini 3, Grok 4.3, and Claude Sonnet 4.8 shows the problem is industry-wide, not one lab's bug.
For developers wiring these models into health, wellness, or search-adjacent products, the lesson is concrete: source attribution and provenance labeling need to be a design requirement, not an afterthought fixed after a watchdog report. If citation quality on a topic this consequential can't be trusted by default, that's a real signal about how these systems weight and surface information more broadly, and it's worth asking what else is slipping through unlabeled.
Related Reading
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
From Archives PrimeAgentOrchestrator Spawns New Claude Code Agents Pre-Loaded with Memory
From Archives Authors' AI Training Claims Face Unproven Legal Hurdles
Common Questions Answered
What did AlgorithmWatch discover about AI chatbots linking to anti-abortion websites?
AlgorithmWatch conducted 270 queries across ChatGPT-5, Gemini 3, Grok 4.3, and Claude Sonnet 4.8 and found that in at least one out of every four queries, the chatbots included links to anti-abortion websites without distinguishing between official health authorities and ideologically driven organizations. The study tested three fictional personas asking about unplanned pregnancy in English, German, and Italian.
Why is linking pregnant users to anti-abortion sites without disclosure a problem for AI chatbot developers?
This represents a liability problem and a retrieval and ranking failure rather than a neutral glitch, as users seeking information about unplanned pregnancy need accurate, sourced information from credible health authorities. Providing links to advocacy groups without labeling them as such misleads users who are looking for medical guidance, not ideological debate.
Which AI chatbots were tested in AlgorithmWatch's study on pregnancy-related queries?
The study tested four major AI chatbots: ChatGPT-5, Gemini 3, Grok 4.3, and Claude Sonnet 4.8. These chatbots were evaluated across multiple languages and personas to assess how they handle sensitive queries about unplanned pregnancy.
How do the chatbots present anti-abortion links to users seeking pregnancy information?
The chatbots present these links with a sympathetic tone, tidy formatting, and links that appear to lead to official sources, making them indistinguishable from legitimate health authority resources. This deceptive presentation makes it difficult for users to identify that they are being directed to advocacy organizations rather than neutral medical information sources.
Further Reading
- Leading AI chatbots are nudging pregnant women to anti-abortion ... - openDemocracy
- AI Chatbots Point Women to Unproven and Unethical “Abortion Pill Reversal” Report Shows - Campaign for Accountability
- AI Chatbots Give Harmful Tip to Users Seeking Abortion Reversal - Bloomberg
- Reproductive health and ChatGPT: an evaluation of AI-Generated responses to commonly asked abortion questions - PubMed
- AI's abortion misinformation problem - Salon
