People asking AI chatbots about unplanned pregnancy may receive answers that appear calm, balanced, and useful. But an AlgorithmWatch investigation found that those answers can also send users toward anti-abortion websites without making that position clear.
The investigation looked at ChatGPT-5, Gemini 3, Grok 4.3, and Claude Sonnet 4.8. Across 270 responses in English, German, and Italian, the chatbots often treated ideologically driven organizations and official health authorities as if they belonged in the same category.
What AlgorithmWatch Tested
AlgorithmWatch used three fictional personas dealing with unplanned pregnancies. Those personas asked questions in English, German, and Italian, creating 270 responses for analysis.
The models tested were ChatGPT-5 from OpenAI, Gemini 3 from Google, Grok 4.3 from xAI, and Claude Sonnet 4.8 from Anthropic. The central concern was not simply whether the tools offered links, but what kind of sources they surfaced and how they described them.
According to the investigation, in at least one out of every four queries, the chatbots included a link to an anti-abortion website. The issue was especially sensitive because users in this situation may be looking for urgent, practical information, not a list of sources whose aims are hard to evaluate.
The findings point to a problem with how AI assistants present authority. A chatbot answer can sound complete while hiding important context about the organizations it recommends. For users who do not already know the landscape, that missing context matters.
Profemina Appeared Often
One organization appeared repeatedly in the results: Profemina. AlgorithmWatch found that Profemina showed up in about 17 percent of all responses.
According to AlgorithmWatch, Profemina has ties to Heartbeat International, one of the oldest anti-abortion organizations in the United States. The investigation said the chatbots usually did not flag that connection when presenting the organization to users.
Only when pressed did ChatGPT say that Profemina is "not a neutral medical source" and "has a stance aimed at discouraging abortion or putting it under strong moral scrutiny." That detail is important because it shows the model could provide relevant context in at least some circumstances, but did not reliably offer it upfront.
Other systems showed similar confusion. Gemini linked to Profemina five times in a single conversation, shared a graphic from the group’s website, and later warned about the same source in that chat. Claude also issued warnings while linking to the organization at the same time.
The pattern was notable in some language settings. In 33 percent of German-language and 29 percent of Italian-language queries asking for personal experiences, the chatbots linked to Profemina. The organization did not respond to requests for comment.
German Counseling Results Raise Practical Risks
The investigation also found a specific issue in German-language conversations. In eleven out of twelve German-language conversations, the chatbots recommended Caritas for pregnancy counseling.
But Caritas does not issue the counseling certificate that German law requires before a person can obtain a legal abortion within the first twelve weeks. That distinction is not a small technicality. If someone follows the recommendation first and later learns the certificate will not be issued, they could lose critical time.
ChatGPT described Caritas as a "recognized" organization for mandatory counseling and listed it alongside Pro Familia as an equal option. AlgorithmWatch suspects Caritas’s large digital footprint, with roughly 300 diocesan counseling centers across Germany, may help explain why the models surfaced it so often.
This example shows why source ranking is not neutral in practice. A chatbot may choose a prominent organization because it is highly visible online, yet that visibility does not mean the recommendation fits the user’s legal or practical need.
Companies Pointed To Policy, But Gaps Remain
Google and OpenAI pointed to their policy guidelines in response to the findings. Those guidelines do not specifically address abortion-related queries, according to the source article.
An OpenAI spokesperson said GPT-5.5, which shipped after the testing period, is "the best so far at taking user context, such as gender, into account." Anthropic and xAI did not respond.
The gap highlighted by AlgorithmWatch is not only about one model or one organization. It is about how AI systems handle topics where users need clarity about source credibility, institutional role, and ideological position.
For a traditional search engine, users often see a list of results and can decide which links to open. A chatbot compresses that process into a single answer, which can make the response feel more authoritative than the underlying source mix deserves.
Why AI Answers Can Become Gatekeepers
The source article argues that the abortion findings reflect a broader risk around sensitive topics. AI-generated answers, even when search is built in, are shaped by training data, cultural weightings, and outside manipulation.
That matters because the presentation is persuasive. A response that sounds empathetic and complete may lead non-experts to trust it without checking where the information came from.
The article also notes that research shows Google’s AI-generated answers dramatically cut clicks to underlying sources. If users rarely visit the original pages, they may not see missing context, institutional limits, or ideological signals that would help them judge the advice.
Legal and regulatory treatment is also changing. A court in Munich ruled that limited liability protections designed for search engines do not apply and that AI-generated answers must be treated as original content. Germany’s media regulators also classify AI search engines and chatbots as content providers with full legal responsibility.
The AlgorithmWatch investigation suggests that meeting those obligations could be difficult. If chatbots continue to blend official health authorities, counseling services, and ideologically driven organizations without clear disclosure, users may receive guidance that looks neutral while quietly steering them in a specific direction.