Google’s AI Overviews answer 82% of health searches. Here is who they cite.

In December 2025 the SEO platform SE Ranking ran 50,807 German-language health searches from Berlin and recorded what Google showed. AI Overviews — the AI-written summaries above the search results — appeared on more than 82% of them. For a patient searching in German, an AI summary is now the default first answer.

Who gets cited

The study counted 465,823 citations inside those summaries. The single most-cited domain was YouTube, with 20,621 citations (4.43%), ahead of the public broadcaster NDR (3.04%) and MSD Manuals (2.08%). Academic journals accounted for 0.48% of citations, German government health institutions for 0.39%, and international government institutions for 0.35%. By SE Ranking’s classification, about 65% of citations came from sources not designed to ensure medical accuracy.

The figure that matters most for anyone publishing health content: only 36% of the links cited in AI Overviews also appeared in Google’s top ten organic results. Being cited and ranking are no longer the same thing.

When the summary is wrong

In January 2026 a Guardian investigation documented AI Overviews giving misleading health information — among the examples, ‘normal’ ranges for liver blood tests presented without accounting for factors such as age or sex, and dietary advice for pancreatic cancer that experts called wrong. Google disputed the scale of the problem, saying its clinicians found that in many instances the information was not inaccurate — but by 11 January it had removed AI Overviews for the queries ‘what is the normal range for liver blood tests’ and ‘what is the normal range for liver function tests’. The British Liver Trust welcomed the removal and noted that the wider issue remained.

What this means for a medical practice

A practice cannot control what Google summarises. It can control whether its own content is a good candidate to be cited — and the data suggests that door is more open than rankings alone would imply.

• Specific beats general. AI systems lift clear, self-contained answers. A page that explains exactly how a hearing test works at your practice, what to bring and how long it takes is more citable than a generic ‘services’ list.
• Video counts. YouTube being the top cited domain is a signal, not an accident. A short, factual explainer recorded by the doctor — how a procedure runs, how to prepare — is a legitimate, citable asset.
• Structure the facts. Healthcare structured data, clear headings and consistent practice details make content easier for machines to read and attribute.
• Stay within the rules. Everything above remains medical advertising. No promises of cure, no superlatives, no comparisons — factual information only. That is also the kind of content least likely to end up in the next investigation.

How FIB Consulting approaches it

We build practice websites and patient-facing content that is factual, structured and easy for both patients and AI systems to read — and we keep it within the HWG and the professional codes. See our medical IT work, the structured data that matters, or get in touch.

Sources

SE Ranking study of 50,807 German-language health searches (December 2025), as reported by Search Engine Journal, January 2026 · The Guardian, January 2026 · eWeek, 5 January 2026 · TechCrunch, 11 January 2026 · Storyboard18, 12 January 2026 (British Liver Trust statement).