How ChatGPT, Perplexity, and Gemini Find Therapists Differently — And What to Do About Each One
The five major AI engines use completely different data sources to recommend therapists. ChatGPT now runs on Yelp's 330M reviews. Perplexity crawls the live web. Gemini reads your Google Business Profile. One strategy won't cover all of them.
Not One Engine, Five — and They Don't Agree
Most guidance on AI discoverability treats AI search as a monolith: "optimize for AI" and you're done. That framing broke down in 2026. The five major AI engines a therapy client might use — ChatGPT, Perplexity, Gemini, Apple Intelligence, and Microsoft Copilot — are built on completely different data plumbing. A therapy practice that's prominent on one can be invisible on another, for reasons that have nothing to do with the quality of their clinical work.
In July 2026 we published the first data we're aware of on this: 96 live AI answers to real local therapy queries, every cited URL classified. The top-line finding — 90% of answers cited a practice's own website — held across all three engines we probed. But the engine-level fingerprints were sharply different: ChatGPT leaned on Google Maps data, Perplexity drew mostly from practice websites directly, and Gemini tracked Google's local pack almost exactly.
Since that probe, one major shift happened: on July 23, 2026, OpenAI announced a data-licensing deal with Yelp — 330 million reviews and 8 million-plus business listings now directly accessible to ChatGPT (SearchEngineLand, July 2026). That changed ChatGPT's local data fingerprint more than anything in the prior year.
This post breaks down each engine's data sources, what changed, and one specific action per platform.
ChatGPT: Now Running on Yelp
ChatGPT is the engine most therapists think of first — and its local data story has changed the most in 2026.
The July 2026 shift: Before mid-2026, ChatGPT's local business results drew heavily from Foursquare's Places API (which shut down its consumer-facing platform in 2025) and from Bing's web index. On July 23, 2026, OpenAI announced a non-exclusive licensing deal with Yelp: 330 million reviews and over 8 million business listings now feed directly into ChatGPT's local recommendations (SearchEngineLand, July 2026). Yelp data — ratings, review text, hours, services, photos — is now a primary ingredient in what ChatGPT tells a client when they ask for a therapist recommendation.
Our July 3 probe (conducted before the deal) found 27% of ChatGPT's raw citations were Google Maps links and 57% were practice websites. That ratio has shifted since Yelp became a direct input — Yelp reviews now appear prominently in ChatGPT's local answers. The underlying principle hasn't changed: ChatGPT combines structured business data (now primarily Yelp) with its own web-indexed content (via GPTBot) and Bing's search index (when ChatGPT Search is enabled).
Three sources ChatGPT uses for therapist queries:
- Yelp business data — ratings, review text, business hours, services, photos (post-July 2026 deal)
- Bing's web index — your website content as indexed by Bing, accessed via ChatGPT Search
- GPTBot crawl — OpenAI's own crawler reading your practice website directly (when allowed in robots.txt)
The one action for ChatGPT: Claim and complete your Yelp listing. This is now the highest-impact action for ChatGPT visibility that most therapists haven't done. Your Yelp name, address, phone, specialty categories, and recent reviews are feeding directly into ChatGPT's local answers. If your listing is missing, incorrect, or has no reviews, ChatGPT is working around a gap in your data. Also: ensure your site appears in Bing's index by submitting your sitemap at bing.com/webmasters, and allow GPTBot in your robots.txt.
Perplexity: The Live Web Reader
Perplexity operates differently from every other engine on this list. Instead of pulling from a licensed directory or a static index, it runs retrieval-augmented generation (RAG): every query triggers a live web search, and Perplexity's model synthesizes the results on the spot, citing each source inline.
This has a specific implication for therapists: Perplexity can read your website directly. Its crawler (PerplexityBot) visits practice sites and indexes their content. When someone asks Perplexity for a trauma therapist in their city, it may pull directly from your trauma therapy service page, excerpt a relevant passage, and cite it with a link.
From our July probe, 72% of Perplexity's classifiable cited URLs were individual practice websites — the highest of any engine we tested. Psychology Today was the only directory Perplexity cited with any frequency; Yelp appeared occasionally; everything else was close to zero.
What Perplexity is extracting from your website is specific:
- Direct-answer opening paragraphs — Perplexity extracts passages that answer the implicit query in the first ~150 words of a page. A service page that opens with "I help adults in Austin recover from trauma using EMDR therapy" is more extractable than one that opens with "Welcome to my practice."
- Question-format headings — H2s like "What does EMDR therapy involve?" give Perplexity a passage it can match to a query and cite.
- Structured data — FAQPage and Service schema markup tell Perplexity's model what your pages are about with higher confidence than unstructured prose.
The one action for Perplexity: Make sure PerplexityBot is allowed in your robots.txt, and structure your top service pages with direct-answer opening paragraphs and question-format H2s. A 5-minute robots.txt edit is the difference between Perplexity being able to read your site and being unable to. For deeper Perplexity optimization, see our Perplexity guide for therapists.
Gemini and Google AI Overviews: Two Different Surfaces
Google's AI surfaces are where most therapists get confused — because there are two of them, and they behave differently.
Google AI Overviews (the SERP feature) are the AI-generated summaries that appear above traditional search results on google.com. As of late 2025, Google made a deliberate decision to exclude sensitive healthcare provider queries — including "therapist near me" and similar local mental health searches — from AI Overviews. For those queries, the traditional local pack (the map + three listings) still appears without an AI summary above it. This means Google AI Overviews are less relevant to therapists' core client-acquisition queries than to most other businesses. Traditional local SEO still governs the results a client actually sees when they search Google for a therapist.
Gemini (gemini.google.com) is a different story. The Gemini AI chat tool does answer therapist recommendation queries — and it draws entirely from Google's local stack: Google Business Profile data, Google Maps, and Google Reviews. In our July probe, Gemini's recommendations matched Google's local results almost exactly. It masked its source URLs behind redirect links, but brand-mention tracking confirmed the pattern: the practices Gemini recommended were the same ones ranking well in Google Maps for that query.
This makes Gemini the most Google-native of the engines. The same signals that rank you in the local pack — GBP completeness, review velocity, category accuracy, NAP consistency — determine whether Gemini includes you in its recommendations.
The one action for Gemini: Treat your Google Business Profile as a real-time document, not a setup-and-forget listing. Categories, services, photos, and hours should match your current reality. Review recency matters: a practice with 12 reviews, the newest from last month, consistently outperforms one with 20 reviews all from two years ago in both Google Maps and Gemini recommendations. For the full GBP optimization guide, see our GBP guide for therapists.
Apple Intelligence and Siri: The Yelp-Maps Stack
When an iPhone user asks Siri for a therapist, the recommendation comes from two sources: Apple Maps and Yelp. Apple has a direct integration with Yelp for business reviews and ratings, which surfaces through Siri, Apple Maps, Spotlight, and Apple Wallet. This makes the July 2026 Yelp/ChatGPT deal even more significant for therapists: a complete, accurate Yelp listing now feeds both ChatGPT and Apple Intelligence simultaneously.
Apple's coverage of the therapist-finding use case has grown with Apple Intelligence — the broader AI layer Apple launched across iPhone, iPad, and Mac. When a client asks Siri "find me a trauma therapist near me," the result is driven by Apple Maps data first, with Yelp reviews and ratings enriching the listing.
The one action for Apple: Claim your business on Apple Business Connect (businessconnect.apple.com). It's free, takes about 10 minutes, and gives you direct control over your Apple Maps listing — name, address, phone, hours, and category. Since Yelp feeds both Siri and ChatGPT, a single Yelp cleanup benefits three AI surfaces. For the full Apple setup guide, see our Apple Maps and Siri guide for therapists.
Microsoft Copilot: Healthcare Directories and Bing Places
Microsoft launched Copilot Health in preview in May 2026 — a dedicated health-focused AI layer that connects to US provider directories, FHIR-connected healthcare networks (52,000+ healthcare organizations via FHIR endpoints), wearable data platforms, and major health information sources. For clinical provider searches, Copilot Health draws from formal healthcare directories that most therapists have never thought to update: NPPES (the NPI registry), CAQH ProView, and state licensure databases that feed commercial payer directories.
For local therapist queries that aren't explicitly routed through Copilot Health, Microsoft Copilot uses Bing's web index and Bing Places — the same underlying data as ChatGPT Search. Submitting your sitemap to Bing Webmaster Tools ensures you appear in both.
The one action for Copilot: Verify that your NPI registry entry (nppes.cms.hhs.gov) reflects your current practice name, address, phone, and specialty. It's the most-queried credentialing database in US healthcare and feeds multiple AI health surfaces you can't directly claim. For broader Copilot guidance, see our Microsoft Copilot guide for therapists.
The Cross-Engine Priority Matrix
Here's what this looks like as an action table — which signals matter for which engines, and what to prioritize if you're starting from scratch:
| Action | ChatGPT | Perplexity | Gemini | Siri | Copilot |
|---|---|---|---|---|---|
| Claim + complete Yelp listing | High (primary data source) | Medium (cited occasionally) | — | High (primary data source) | — |
| Complete Google Business Profile | Medium (Gemini drives; some Maps citations) | Low | High (foundational) | — | — |
| Allow AI crawlers in robots.txt | High (GPTBot) | High (PerplexityBot) | High (Googlebot-Extended) | — | High (BingBot) |
| Submit sitemap to Bing Webmaster Tools | High (ChatGPT Search) | Medium | — | — | High |
| Service pages with direct-answer structure | High | High (primary extraction source) | Medium | — | — |
| FAQPage and Service schema markup | High | High | High | — | — |
| Claim Apple Business Connect | — | — | — | High (direct) | — |
| Verify NPI registry entry | — | — | — | — | High (direct) |
| Google Review velocity (monthly) | Medium (via Maps) | — | High (foundational) | — | — |
| Psychology Today profile | Low (post-Yelp deal) | High (most-cited directory) | Low | — | — |
The top three rows are where most therapists should start: Yelp (feeds ChatGPT and Siri), Google Business Profile (feeds Gemini), and robots.txt + Bing Webmaster Tools (feeds the crawl-based engines). Those four actions cover all five engines at a basic level before you touch anything else.
What Holds Across Every Engine
Despite the different data sources, one signal appears in every engine's ranking logic: the practice website itself. Our probe found 90% of AI answers cited at least one practice website, and that held whether the answer came from ChatGPT, Perplexity, or Gemini. The engines differ in how they read it (direct GPTBot crawl, Bing index, Google index) but they all weight it heavily.
The pages that get cited consistently share the same structure: dedicated per-specialty pages (not one umbrella Services page), direct-answer opening paragraphs, question-format H2s, and schema markup that tells the engine what the page is about without requiring interpretation. Building that foundation benefits all five engines simultaneously — and it's the one investment that doesn't need re-done as the engine landscape shifts.
That's the practical case for treating your website as the core of your AI visibility strategy, rather than chasing each engine's specific data sources in isolation. The engines compete with each other for data advantage; the therapist who owns the canonical source of information about their practice — their own well-structured website — is positioned for wherever the data landscape moves next.
WebsiteTherapy builds this in from the start: structured service pages, 25+ schema types, AI crawler configuration, Bing indexing, and Yelp monitoring — so the engines that matter most for finding clients can read your practice clearly. Here's how it works.
Sources: SearchEngineLand (Yelp/OpenAI licensing deal, July 23, 2026); WebsiteTherapy first-party probe data (96 AI answers, July 2026; see The Anatomy of an AI Citation); Microsoft Copilot Health launch announcement (May 2026); Perplexity AI crawler documentation (2026); Google AI Overviews guidance for sensitive health content.