AI Chatbots Routinely Direct Pregnant Users to Anti-Abortion Websites, Investigation Finds

The study, published on August 24, 2026, highlights a troubling pattern of bias and incomplete information in AI-powered tools that millions rely on for personal health guidance. AlgorithmWatch examined how four leading chatbots handle queries about unplanned pregnancies, finding that they often default to sources with explicit anti-abortion positions without flagging their orientation.

In Germany, the problem takes a legal twist. The country requires women seeking an abortion to undergo mandatory counseling and obtain a certificate from a recognized provider. Yet the chatbots repeatedly sent users to Caritas, a Catholic charity, which does not issue those certificates. This could lead women down a bureaucratic dead end at a stressful time.

The researchers did not name individual models or versions, but noted that the pattern held across all tested platforms. The affected responses lacked any disclaimer or label indicating the ideological nature of the linked content. Profemina, for instance, presents itself as a pregnancy support service but is actively anti-abortion.

AI companies have not yet issued formal responses to the investigation. In previous cases, firms like OpenAI and Google have acknowledged challenges with bias and factual accuracy, and have introduced measures such as source disclaimers. However, this study suggests such safeguards remain inconsistent.

AlgorithmWatch called for greater transparency from chatbot developers and for regulators to enforce disclosure rules, particularly in sensitive health contexts. “When a user is in a vulnerable situation, they deserve to know whether the information they receive is medically neutral or carrying a specific ideological agenda,” said a spokesperson for the group.

The findings add to a growing body of evidence that AI chatbots can amplify societal biases—whether political, religious, or commercial—without users being aware. In the United States, where abortion laws vary widely by state, similar concerns have been raised about chatbots steering users toward crisis pregnancy centers that may offer biased counseling.

Proponents of the anti-abortion links argue they provide valuable support to women considering alternatives. But critics counter that the lack of disclosure is deceptive, especially when users may be assuming the AI is providing objective medical guidance.

Why This Matters

Direct impact on vulnerable users: Pregnant individuals seeking unbiased information may be unknowingly funneled toward ideologically driven services, potentially delaying or altering their decisions.
Broader trust in AI: The findings erode confidence that chatbots provide neutral, factual health advice—especially when companies market them as general-purpose assistants.
Regulatory gap: No clear rules currently mandate disclosure of partisan or religious affiliations in AI-generated content, leaving users unprotected.

Background

AI chatbots have been rapidly adopted for personal advice, including health and reproductive decisions. Past investigations have revealed biases in areas like racial stereotypes, political opinions, and medical misinformation. In 2024, studies showed that chatbots sometimes invent medical citations or provide outdated treatment recommendations. The AlgorithmWatch study zeroes in on a specific, high-stakes scenario: unplanned pregnancy. Germany’s mandatory counseling system adds legal complexity, while in the U.S., variations in state laws create confusion. Regulators in the EU (under the AI Act) and elsewhere are still crafting rules for transparency and fairness, but enforcement lags behind deployment.

Key Perspectives

AlgorithmWatch: The watchdog emphasizes user autonomy and informed consent. They argue that chatbots must clearly disclose the nature of any linked organization, especially when the topic is emotionally charged and time-sensitive.
AI Companies (OpenAI, Google, xAI, Anthropic): They generally state their models are trained to be “helpful and harmless,” but admit that bias mitigation is imperfect. They may point to ongoing improvements and opt-out mechanisms, but have not yet specifically addressed this investigation.
Anti-Abortion Groups: Organizations like Profemina might claim they offer legitimate counseling and support, and that directing users to them is simply providing a resource—not an endorsement. They could argue that labeling them as “anti-abortion” is itself a value judgment.
Critics of the Investigation: Some skeptics may note that the sample size of 270 queries is small, or that the chatbot responses depend heavily on phrasing. Still, the pattern across multiple platforms is striking.

What to Watch

Company responses: Whether OpenAI, Google, xAI, and Anthropic issue statements or update their safety guidelines in the coming weeks.
Regulatory action: The German Federal Ministry of Justice or the European Commission—under the AI Act—could threaten fines or demand changes if transparency obligations are violated.
Further research: Expect follow-up studies with larger samples, possibly expanding to other languages and regions (e.g., Poland, the U.S., Brazil) where abortion access is contested.

ZOTPAPER