The Query Google Refuses to Let AI Answer

Google tested AI Overviews on local healthcare queries in 2023 at 100% coverage, then withdrew them entirely. By December 2025, a search for "dokter kulit terdekat" triggers zero AI Overview presence. Not a decline, zero. Meanwhile, treatment and symptom queries in the same healthcare category now trigger an AI Overview 93 to 100% of the time. Google made an active choice about which query types AI is allowed to answer, and local, patient-acquisition queries didn't make the list, a decision with direct budget implications for anyone running healthcare marketing.

Most healthcare marketing teams read the AI Overview headlines as a reason to worry about local SEO, assuming AI is quietly eating every category of search. The data, once broken down by query type rather than read as a single headline number, says the opposite for this specific category.

A Deliberate Reversal, Not a Gap in the Data

BrightEdge's Generative Parser tracked healthcare keywords across December 2023, 2024, and 2025. Overall AI Overview coverage climbed from 59% to 89% over that period. But broken down by query type, the local category is the one clear exception, and it didn't just plateau, it reversed from full coverage to none.

The most direct reading: Google decided that when someone is trying to find and book care nearby, an AI-generated summary isn't the right answer format. The local pack, Google Business Profile, and organic listings are. That's a rare, unambiguous signal about where a specific budget line should go.

Why Two Trackers Disagree About the Overall Number

BrightEdge's 89% overall figure isn't the only published number on healthcare AI Overview coverage, and the other one doesn't match. Semrush tracked 16.04% of health keywords with AI Overview presence as of November 2025, a 9.58% decrease from March to November that same year. That's a dramatically lower figure than BrightEdge's, and the two shouldn't be blended into a single "the truth is somewhere in between" number.

The likely explanation is methodology, not disagreement about reality. BrightEdge tracks clinical healthcare keywords specifically, treatment, symptom, and coding queries concentrated at the high end of AI Overview adoption. Semrush's broader health keyword set likely includes lifestyle content and local-intent queries, which now sit at or near 0% following exactly the reversal described above. Different populations of queries, measured by different tools, produce different averages. Both figures should be cited with their own methodology attached, never as a single blended "healthcare AI Overview rate."

Ranking Well and Getting Cited Are More Connected Here Than Elsewhere

BrightEdge's data adds one more finding that changes how a hospital should think about the relationship between traditional SEO and AI visibility: roughly 80% of healthcare AI Overview citations come from sources that also rank organically in the traditional top ten. That's the highest alignment of any industry BrightEdge tracked. In most categories, ranking well and getting cited by an AI Overview are related but distinct efforts. In healthcare, they're close to the same effort.

There's a second layer to this. BrightEdge also found that YMYL verticals, healthcare included, show the smallest year-over-year change in which sources populate the AI Overview citation pool. Google has an established, trusted set of health sources and changes it slowly. The practical consequence for a hospital building from a smaller base: the compounding advantage belongs to whoever is already ranking well and already trusted, and new entrants are working against a citation pool that shifts gradually rather than resetting each cycle.

Voice Search Adds a New Front to the Same Problem

Google launched Search Live, its multimodal, voice-enabled AI Mode, in Indonesia in April 2026, extending conversational search to all supported languages and regions through the Google app on Android and iOS. Voice queries in this format read differently than typed ones. "Kenapa kepala saya sering pusing" is longer, more conversational, and more likely to include colloquial or regional phrasing than a typed "sakit kepala." FAQ schema and conversational content formatting matter more in this environment, since the query itself already arrives shaped like a question rather than a keyword fragment. Indonesian-specific voice query data for healthcare is not yet published, but the format shift itself is confirmed and already live.

What Still Wins a Local Healthcare Query

A 2025 study of self-pay patients at a hospital in Kediri found that online ratings, weighted at 0.948, and online reviews, weighted at 0.387, together explained 94.7% of the variance in hospital selection. For local, provider-intent queries specifically, that is close to the entire decision, not a contributing factor among many, and it is measured directly from Indonesian patients rather than inferred from a global benchmark.

The Journey That Skips the Hospital Entirely

The local-query reversal matters more once it's placed inside the actual patient journey, because the local search is often the last chance a hospital gets to intercept a patient before they convert somewhere else entirely. A typical path: a patient searches a symptom, lands on an Alodokter article, reads the educational content, and then opens Halodoc directly for a teleconsultation, never touching a hospital's website or local listing at all. Only 8.72% of Indonesian internet users reported using an online health app in a 2025 APJII survey, which sounds small until it's read alongside how concentrated that usage is, Halodoc and Alodokter alone account for the large majority of it.

This is precisely why the local pack and Google Business Profile carry more weight than their query volume alone suggests. For a hospital, winning the local, provider-intent search is often the only remaining checkpoint before a patient defaults to a teleconsultation app that has no institutional connection to that hospital whatsoever. Losing that checkpoint doesn't just cost a click, it can cost the entire patient relationship to a platform built to keep the interaction inside its own app.

None of this is an argument against symptom and condition content. It's an argument for treating the local, provider-intent stage as the checkpoint it actually is, fully resourced with Google Business Profile management and review generation, rather than the leftover budget line after informational content is funded first.

---

Frequently Asked Questions

Does this mean local SEO doesn't need to account for AI search at all?

For the local query itself, correct, Google has removed AI as an intermediary there. But the patient's journey usually starts earlier, with a symptom search that does trigger an AI Overview, before narrowing to a local, provider-intent search. Both stages still matter, they just play by different rules.

Why did Google remove AI Overviews from local healthcare queries specifically?

Google hasn't published a stated reason. The most consistent inference across the data: local queries carry strong transactional intent tied to a physical decision, booking, visiting, calling, which the local pack and Google Business Profile already serve well without a generated summary in between.

Should we still add FAQ schema to our location pages?

Yes. FAQPage schema remains a supported signal for Google's own indexing and rich results, independent of whether an AI Overview appears. It also plays into eligibility for the local knowledge panel entity signals.

Could this reverse again?

Possible. Google removed AI Overviews for other specific medical query types after documented accuracy problems on lab-value queries. Treat the current 0% as the present state, verified as of December 2025, not a permanent guarantee.

Does this pattern hold for Indonesian-language local queries the same way?

Unconfirmed directly. The BrightEdge trigger-rate data is English-language, US-market measurement. Indonesian-language AI Overview behavior for local queries specifically has not been separately published, though the underlying local-pack mechanics are not language-dependent.

Why do BrightEdge and Semrush report such different overall AI Overview coverage numbers?

They're most likely measuring different populations of queries. BrightEdge tracks clinical healthcare keywords specifically, which sit at the high end of AI Overview adoption. Semrush's broader health keyword set likely includes local and lifestyle queries, which now trend toward zero following the reversal described above. Cite each figure with its own methodology rather than averaging them.

Should we invest in Search Live optimization now, or wait for more Indonesian-specific data?

Start structuring for it now. FAQ-formatted content and conversational, question-shaped headings help regardless of whether a query arrives typed or spoken, and Search Live is already live in Indonesia as of April 2026. Waiting for Indonesian-specific voice data means waiting for a dataset that may not exist for some time.

What's the single biggest risk of ignoring this local-query reversal?

Losing the last checkpoint before a patient defaults to a teleconsultation app entirely. A hospital that isn't winning the local, provider-intent search isn't just missing traffic, it's often losing the patient relationship to a platform with no institutional tie to that hospital at all.

The voice-search and AEO implications of this same reversal, and how it interacts with Indonesia's growing conversational search behavior, get a fuller treatment in Found Before They Search. The free chapter is at arfadia.com/resources/ebook-found-before-they-search, also on Amazon, Google Play Books, and Apple Books.

Multi-location hospital groups rebuilding their local SEO around this finding can start with the audit inside our Healthcare SEO service.

---