How Often AI Assistants Name a Therapy Practice: 155 Local Answers, Counted (October 2026)

About one answer in four. In WebsiteTherapy's October 2026 check, we asked three AI assistants 155 local questions on behalf of 11 US therapy practices, and the practice was named in 39 of the 155 answers (25%).

The questions were the kind a person types when looking for help: "Who are the best therapists in [city]?", "Can you recommend a few counseling practices in [city] that are accepting new clients?" and "Where can I find help for [issue] in [city]?" Each question went to each assistant once, between October 2 and October 5, 2026.

The rate alone hides most of what we found. What the question asked changed the odds more than which assistant answered it, and the practices split into those named often and those never named at all. The method is at the end, with what it does and does not show.

Questions about a specific issue in a city named the practice about twice as often as general "best" or "accepting new clients" questions. Across the 11 practices, we counted:

• "Where can I find help for [issue] in [city]?" We counted the practice named in 21 of 69 answers (30%).
• "Who are the best therapists in [city]?" and "Can you recommend a few practices in [city] accepting new clients?" We counted the practice named in 8 of 56 answers (14%).
• Questions with no place in them, such as "Who are the best therapists?", were asked separately for 7 practices. We counted 0 of 51 answers that named the practice.

The other 30 local answers came from questions written around a single practice's specialty, so they count in the total but are not broken out here.

A question with no city cannot be about a local practice, so the zero is expected. It still matters, because it is how many people start: if a person asks an assistant for "a good therapist" without saying where, the answer points them to directories and national services, not to a practice like yours. The practical lesson is the other half. When the question names an issue and a place, a practice whose own pages name the same issue and place is in the running.

Across the 11 practices, we counted similar rates from all three assistants:

With about 50 answers each, these differences are within what chance alone would produce. Do not choose one assistant to "optimize for" on numbers like these. The same pages that get a practice named by one tended to get it named by the others.

Psychology Today was cited as a source in 89 of the 103 answers whose sources we could read (86%), across the 11 practices: 39 of 52 from OpenAI's model and 50 of 51 from Perplexity. Gemini's 52 answers linked only through Google redirect addresses, so we could not see which sites they drew on.

That fits what we found in our earlier probe of directories in AI answers: AI assistants read Psychology Today and cite it often. Being in the directory is not what separated the practices that were named from those that were not. When an answer did name the practice, it linked the practice's own website in 32 of those 39 answers, across 7 practices.

The directory gets cited either way. What a practice controls is whether its own site is the page the answer can quote: what you treat, where you practice, who you see, written in plain words on your own pages. Our guide to writing a service page an AI answer can quote covers how.

The named answers were concentrated. Of 11 practices, 4 were never named in any local answer, and three practices accounted for 28 of the 39 answers that named a practice.

We did not test why, and with 11 practices we cannot. The pattern is still worth knowing before you read any single number about AI visibility, including ours: our average of 25% describes almost none of the practices in it. For one practice, the useful question is not the average but its own count, asked in its own city about its own specialties.

1. Ask the questions yourself. Ask ChatGPT, Perplexity and Google, in your city: "Where can I find help for [what you treat] in [your city]?" Note whether you are named and which pages each answer cites.
2. Write the issue and the place on your own pages. The questions that named practices most often were about a specific issue in a specific place. A page for each thing you treat, saying where you see clients, gives an answer something to quote.
3. Do not expect a general question to find you. A question with no place in it named none of the practices we checked. That is not a fault in your site; it is what the question asks for.
4. Check again next month. Answers vary from one run to the next, so a single answer proves little either way. A count over time tells you more.

Our free AI search audit checks the parts of your site that AI crawlers read. For the practices we host, WebsiteTherapy also checks what ChatGPT, Perplexity and Google AI say about the practice and drafts a fix for anything that is wrong or out of date; the features page sets out what that covers. We do not promise that any assistant will name you.

Population. 11 US therapy practices that work with WebsiteTherapy or were being onboarded in October 2026, We left out our five fictitious template practices (made-up names in made-up towns, which no assistant can name), non-therapy businesses and test sites. These practices are not a random sample: many already have pages written to be found, so the rate for all therapy practices may well be lower.

Questions. 155 local answers to questions that named a city or state, and 51 answers to questions with no place, from the discovery questions set up for each practice. Each question went once to each of three assistants through their APIs: OpenAI gpt-4.1 with its web search tool, Perplexity Sonar, and Google Gemini 3.5 Flash with Google Search grounding. These are the models behind the consumer products, not the consumer apps themselves, and answers in the apps can differ.

Dates. October 2 to October 5, 2026.

Named means the practice's name appeared in the answer text. Linked means the practice's own website was among the answer's cited sources. A source counted for Psychology Today when any cited link pointed to psychologytoday.com.

What it does not show. One answer per question per assistant, so run-to-run variation is not measured. It shows how often practices were named, not why. No practice is identified, and no figure below 50 answers is broken out except the totals above, each with its practice count.

Cite this: WebsiteTherapy, "How Often AI Assistants Name a Therapy Practice: 155 Local Answers, Counted," October 5, 2026, https://websitetherapy.com/blog/how-often-ai-names-therapy-practices-2026

• OpenAI API documentation: web search tool
• Perplexity API documentation
• Gemini API documentation: grounding with Google Search
• WebsiteTherapy: The Therapist Directories AI Engines Actually Ignore