Here is the most surprising number in local AI search, and it belongs to healthcare. Between December 2023 and December 2025, BrightEdge measured AI Overview coverage on health provider-finding queries, the "near me" searches, falling from 100% to 0%.
Google built the ability to answer "find me a therapist", deployed it, looked at it, and took it back out.
If you run a private practice, the temptation is to read that as a reprieve. It is not. The decision about which therapist a client contacts did not disappear when Google stepped back. It moved somewhere Google does not control. This post is about where it went, and what to do about each place.
Why Google retreated
Google classifies mental health content as YMYL, "Your Money or Your Life", the category for material that can directly affect someone's wellbeing. Its quality rater guidelines put mental health crisis information in the most sensitive tier that exists, next to medical diagnosis and pharmaceuticals.
You can see that caution in the coverage data. Across the same period, healthcare AI Overview presence overall grew from 59% to 89%. But the growth was not evenly distributed:
| Health category | AI Overview coverage |
|---|---|
| Genetic / genomic | 97% |
| Cardiology | 96% |
| Gastroenterology | 94% |
| Primary care | 74% |
| Mental health / behavioral | 63% |
| Provider-finding "near me" | 0% |
Mental health sits at the bottom of the answered categories, and the referral query is not answered at all. Google is willing to explain what EMDR is. It is not willing to tell a person in distress which specific human to call.
A caveat, in the spirit of the thing: this is one vendor's parser measuring snapshots on a fixed December date each year. Treat the direction as solid and the decimal points as indicative.
Where the decision actually went
1. Back to the directories
When Google declines to answer, it falls back to ranking pages, and for "therapist near me", the pages that rank are directories. Psychology Today remains the first stop for most people looking for a therapist, alongside insurer directories and the newer platform networks.
This is worth sitting with, because it inverts the usual advice. For a dentist, the website is a genuine lever. For a therapist in 2026, your directory profiles are doing more work than your website, and they are the thing an answer engine reads when it does describe you.
Which makes the most common failure mode expensive: insurer directories are notoriously out of date. Old panels, closed practices, therapists listed as accepting plans they dropped years ago. Every stale entry is both a client who bounces and a signal that you are not a stable, well-described entity.
Do this first: audit every directory you appear in, Psychology Today, each insurance panel, any platform network, your licensing board listing, and make the name, credentials, address, modalities, and accepted plans identical across all of them. This is unglamorous and it is the highest-yield hour you will spend.
2. To the engines that did not step back
Google's caution binds Google. It does not bind ChatGPT or Perplexity, and both will happily name individual therapists when asked.
That asymmetry is the practical heart of AEO for a therapy practice. The referral question your clients ask has effectively split in two: on Google it returns a directory list, and on ChatGPT it returns a short recommendation naming two or three people. You need to know what the second one says about you, and almost no practice does.
3. Up one level, to the education layer
63% of mental health queries still get an AI answer, just not the referral ones. That is the entire band of questions people ask before they are ready to contact anyone:
- "what is the difference between a therapist and a psychiatrist"
- "how much does therapy cost without insurance"
- "is EMDR effective for trauma"
- "how do I know if I need therapy or coaching"
- "what happens in the first therapy session"
These are answered by AI, they are the questions your future clients are actually asking, and they are winnable. A therapist who writes the clearest available answer to "what happens in a first session" is building the kind of page an engine cites, and being cited in the education layer is what makes you a known entity when someone later asks the referral question elsewhere.
What makes an engine willing to describe you
In YMYL categories, models are conservative about who they will vouch for. Three things move that needle.
Credentials, stated as text. Your licence type and number, the state, your degree, your years in practice, your supervision and certifications. Not in a header image, not in a PDF, as readable text on your site and identically across your directory profiles. An engine that cannot verify what you are will not recommend you to someone in distress.
Claims that are sourced. If you cite a prevalence figure or an outcome for a modality, link the primary source, APA, NIMH, a journal. Pages that reference primary literature get read as reference material rather than marketing, and that distinction is doing a lot of work in a YMYL category.
Specificity over warmth. "A safe, non-judgemental space" appears on ten thousand therapist websites and tells a model nothing. "EMDR and IFS for adults with complex trauma, evenings available, Aetna and Cigna in network" tells it exactly who to hand you.
Modality pages are how clients get matched to you
The single most useful structural change most practices can make is one page per modality you actually practise (EMDR, CBT, IFS, DBT, EFT for couples) each explaining what it is, who it suits, what a course of it looks like, and what it costs.
Two things happen at once. The page answers an informational query that is already AI-answered ("is EMDR effective for trauma"), and it attaches you to that modality as an entity. When someone later asks for "an EMDR therapist in Portland", you are a candidate rather than a generic listing.
The same logic applies to fees. "Contact for rates" is invisible. A page that states your session fee, whether you hold sliding-scale slots, what out-of-network reimbursement typically looks like, and which plans you take is one of the few pages in this field that an engine can quote directly, and it screens for fit before anyone books.
A word on ethics, because this field has them
Everything above is marketing, applied to a clinical practice, and that deserves care that dental marketing does not require.
Two lines worth holding. Do not let optimization turn into outcome claims, a model rewards specificity, but specificity about what you offer and to whom, never about results you can guarantee. And do not solicit client testimonials; most licensing boards and codes of ethics restrict or prohibit them, and no visibility gain is worth a board complaint. Your corroboration comes from professional directories, licensing bodies, association memberships, referring clinicians, and the informational content you publish, not from reviews.
The good news is that the ethical route and the effective route point the same way here. Clear credentials, honest scope, sourced claims, plain pricing. That is both what a board wants and what a model needs.
How to check where you stand, in twenty minutes
- Write down the ten questions someone would ask before contacting you. Include at least three informational ones, one about cost, one about insurance, and one referral question ("EMDR therapist in [your city]").
- Ask each in ChatGPT, Google, and Perplexity.
- Record what comes back: for each engine, whether you are named, which directories appear, and which other practices are recommended.
You will find the pattern this post describes, Google handing you a directory, ChatGPT naming a handful of people, and you will find out whether you are in the handful.
FAQ
Does ChatGPT actually recommend individual therapists?
Yes. Unlike Google's AI Overviews, which stepped back from health provider-finding queries entirely, ChatGPT and Perplexity will name specific practitioners when asked for a therapist in a given city, usually two or three, often drawn from directory profiles and practice websites rather than reviews.
Is Psychology Today still worth paying for?
For most practices, yes, but for a reason that has changed. Its value is no longer only the referral traffic it sends you directly. It is that Google now falls back to ranking directories on "therapist near me", and answer engines read those profiles when describing you. Treat the profile as structured data about your practice rather than as an advert, and keep it identical to everywhere else you appear.
Can I ask clients for reviews to improve my AI visibility?
No, and you should not try to route around it. Most licensing boards and ethics codes restrict or prohibit soliciting client testimonials. Build corroboration instead through directory and licensing-board listings, professional association memberships, referring clinicians, and published informational content. This is the one place where standard local-SEO advice does not transfer to therapy.
Why does mental health get fewer AI answers than other health topics?
Google classifies mental health as YMYL, material that can directly affect someone's wellbeing, and holds it to a higher credibility bar. Mental health and behavioral queries sit at roughly 63% AI Overview coverage against 94–97% for clinical specialties like cardiology and gastroenterology, and provider-finding queries have been pulled back to zero.
What should a solo practitioner do first?
Make every directory profile agree with every other one, then write one page per modality you practise and one honest page about fees and insurance. That sequence costs nothing but time and addresses the three things engines actually use: entity consistency, topical specificity, and quotable facts.
Will this change again?
Almost certainly. Google has already expanded, evaluated, and retracted AI coverage in health on a roughly monthly cycle, and the provider-finding reversal shows it will undo a capability it has shipped. That instability is the argument for tracking answers over time rather than auditing them once.
The short version
Google stepped back from telling people which therapist to call. The directories picked that up, ChatGPT never stepped back at all, and the informational layer above the referral is wide open. Your work is to be consistently described everywhere you already appear, specific about what you practise and for whom, plain about money, and present in the questions people ask before they are ready to book.
Kaymak tracks what ChatGPT, Google AI Overviews, and Perplexity actually say when someone asks for a therapist like you, including which directories and which competing practices get named, and turns the gaps into a ranked to-do list. The live demo is free to explore, no signup.
Sources: BrightEdge, Healthcare and AI Overviews, 2023–2025 · Search Engine Land, What gets cited in YMYL AI Overviews · Psychology Today, How to find a therapist