AI chatbots are steering pregnant users toward anti-abortion resources without transparent labeling of the organizations' advocacy positions, according to a new investigation by AlgorithmWatch that tested responses from four major AI services.
Researchers queried ChatGPT, Gemini, Grok, and Claude about unplanned pregnancies across 270 total responses. The anti-abortion organization Profemina appeared in 17 percent of the answers provided by these systems. None of the chatbots disclosed Profemina's stance against abortion access when recommending it as a resource.
The bias extends into jurisdiction-specific harm. In Germany, where abortion regulations require pre-abortion counseling before procedures, ChatGPT and other systems directed users to Caritas, a Catholic organization. The problem runs deeper than mislabeling: Caritas cannot legally issue the mandatory certificate required for abortion access in Germany, making the recommendation functionally useless and potentially obstructive to users seeking timely care.
AlgorithmWatch's findings expose a structural problem in how AI systems handle sensitive health information. When a user presents a medical or personal situation, chatbots retrieve and rank sources without human editorial oversight. The ranking systems appear to amplify advocacy-oriented organizations alongside neutral medical providers. Profemina operates as a pregnancy resource center, organizations known for discouraging abortion without explicitly stating that goal upfront.
The stakes matter here. Users asking AI systems about unplanned pregnancies often seek practical guidance on their options. They expect factual information about medical resources, legal rights, and counseling services. Instead, they encounter redirects to organizations designed to influence their decision-making toward carrying pregnancies to term. This happens silently, without disclaimer.
The chatbot companies face pressure from multiple directions. Conservative groups lobby them to feature pro-life resources. Medical ethics demand balanced, transparent information. Users expect neutral search functionality. Currently, the systems default to treating all organizations equally regardless of their stated mission or conflicts of interest.
OpenAI, Google, Xai, and Anthropic have not fully addressed how their ranking algorithms handle advocacy organizations in health domains. Their training data includes information from both medical institutions and activist groups without clear separation. When users query pregnancy-related topics, the systems cannot distinguish between a neutral health clinic and an organization explicitly designed to discourage abortion.
Germany's legal framework makes the failure concrete and measurable. The country requires evidence-based pre-abortion counseling from qualified providers who issue certificates. Directing users to an organization incapable of issuing those certificates obstructs legal compliance. In other jurisdictions with looser regulations, the harm remains real but harder to quantify: users receive biased information when they need balanced facts.
The investigation points toward a broader accountability gap. AI companies deploy systems that influence major life decisions without rigorous content moderation for sensitive topics. Health-related queries deserve higher standards than general search results. Users asking about pregnancy options need transparent sourcing, disclosed organizational missions, and verified provider credentials.
None of the tested systems currently implement these safeguards. The responsibility falls on the companies to rebuild their ranking systems for health queries, explicitly flag advocacy-oriented sources, and verify provider qualifications in regulated medical contexts. Until then, users will continue encountering hidden persuasion campaigns disguised as neutral AI assistance.