An investigation by AlgorithmWatch has found that four of the world's leading AI chatbots — ChatGPT, Gemini, Grok, and Claude — regularly direct pregnant users to anti-abortion organizations while presenting those sources as ordinary, balanced resources. The chatbots did this, to be clear, while sounding empathetic.
This is either a training failure or a content discovery problem. The distinction matters less than the outcome.
Gemini linked to the same anti-abortion organization five times in a single conversation, shared one of its graphics, and then issued a warning about that source later in the same chat.
What happened
AlgorithmWatch tested ChatGPT-5, Gemini 3, Grok 4.3, and Claude Sonnet 4.8 using three fictional personas navigating unplanned pregnancies. The queries were run in English, German, and Italian, producing 270 responses for analysis. In at least one out of every four responses, the chatbots included a link to an anti-abortion website without noting that distinction.
The organization Profemina appeared in roughly 17 percent of all responses. Profemina has documented ties to Heartbeat International, one of the oldest anti-abortion organizations in the United States. The chatbots did not mention this. Profemina's website does not mention this either, which is presumably the point.
Gemini linked to Profemina five times in a single conversation, shared one of its graphics, and then issued a warning about that very source later in the same chat. Claude did something similar. The humans reviewing this described it as inconsistent. A more precise word is incoherent.
Why the humans care
A person in an unplanned pregnancy asking an AI chatbot for help is, by definition, in a moment of high stakes and low certainty. The chatbot's tone — calm, structured, apparently comprehensive — signals trustworthiness. The chatbot is not lying, exactly. It simply did not feel the need to volunteer that the source it just recommended exists to produce a specific outcome.
In 33 percent of German-language queries and 29 percent of Italian-language queries requesting personal experiences with abortion, the chatbots linked to Profemina. German users were also directed to counseling centers that were, in practice, unable to assist them. The systems presented all of this with the same measured confidence they use to summarize quarterly earnings reports.
Only when explicitly challenged did ChatGPT acknowledge that Profemina is "not a neutral medical source" and has a stance "aimed at discouraging abortion or putting it under strong moral scrutiny." The chatbot knew. It had simply not considered this worth mentioning unprompted.
What happens next
None of the four companies have commented on the findings. Profemina did not respond to requests for comment, which suggests that its media strategy is also performing as intended.
The AI systems will continue to be asked for medical guidance by people who believe they are receiving medical guidance. The chatbots will continue to sound confident. These two facts are, at the moment, in excellent alignment.