How to Get Your Therapy Practice Recognized by AI (ChatGPT, Gemini, Perplexity)

The problem: AI can't recommend a practice it can't recognize

When someone lands on your homepage, they're scanning for one sentence: is this person for someone like me? In our Therapist Recognition Index — a rule-based scan where we attempted 181 private-practice homepages and could score 109 — only 29 of those 109 scored pages (27%) clearly named who they help. Most pages told visitors what they are ("a therapy practice") and where they are, then stopped before the sentence a prospective client was actually looking for.

An AI assistant is, in one narrow sense, just a very fast first-time visitor. It reads what's publicly written about you across your site and the places that mention you, and it tries to answer who is this practice for? If your public writing never names who you help — or names it in vague, credential-first language — then a person struggles to recognize you, and so, it appears, does the model. We call that the recognition gap: the distance between the good work you do and what anyone (human or AI) can tell you do from the outside. AI didn't create the gap. It just makes an old gap more expensive.

How AI assistants appear to choose what to recommend

Here's the honest part most guides skip: we do not know exactly how ChatGPT, Gemini, or Perplexity decide who to recommend. Those systems are closed, they change often, and the same question can return different answers tomorrow. So treat everything here as how these systems generally seem to work, drawn from how they're built and from small public tests — not as rules you can game.

• They read your public footprint, not your inner quality. These systems tend to judge how clearly your practice communicates who you help, what you treat, where you are, and why you're credible — not how skilled a clinician you are. A machine can't see your outcomes; it can only read your words.
• They look for the same signal in more than one place. When your name, specialty, and location line up consistently across your own site, your directory listings, and other pages that mention you, it appears more likely you'll be surfaced. Consistency reads as confidence.
• Specialty pages seem to matter more than specialty bullets. In a small February 2026 test of ChatGPT recommendations across 10 US cities, most recommended practices had a dedicated page about their specialty — not one line in a list of fifteen services. That's a small, unstable snapshot, not a law; the useful signal is the pattern across runs and cities, not a permanent ranking.
• Directories the assistant already trusts get pulled in. In that same test, practices listed on directories like GoodTherapy showed up more often than practices with only a single profile elsewhere. (More on where AI tends to look, and why Psychology Today alone may not be enough, in Before you cancel Psychology Today.)
• They don't search your exact words — they "fan out" into several sub-questions. When someone asks an AI to find a therapist, it appears to break that one prompt into multiple smaller searches ("fan-out") and pull pages for each. One 2026 analysis found that roughly a third of the pages AI cited showed up only in those fan-out searches — and most of those sub-questions had little or no traditional search traffic, so keyword tools can't even see them. The practical read: a page whose headings answer the specific sub-questions a client's request breaks into (e.g. "anxiety therapist in [city] taking new clients," "does CBT help with anxiety," "online therapy in [state]") appears to get surfaced more than one built around a single phrase — and a focused page tends to do better than a sprawling one. Directional, not a guarantee.
• Plain, structured, machine-readable pages help. Broader (non-therapist) research on optimizing content for AI answers found that adding clear citations, plain statistics, and quotable, well-sourced statements raised how often content appeared in AI-generated answers. Important caveat: that study tested general web content with an older model pipeline, not therapist websites and not client-facing recommendations — so read it as directional, not as a promise for your practice.

None of this is a dial you turn for guaranteed placement. It's the same discipline that helps a nervous human recognize you fast — written down clearly enough that a machine can read it too.

What a therapist can actually do (a plain checklist)

You can't control the model. You can control whether your practice is easy to recognize. In rough priority order:

1. Name who you help, in plain words, near the top of your homepage. The single highest-leverage move for humans and, it appears, for machines. "Therapy for adults whose anxiety looks like overthinking and a mind that won't shut off — in-person in [city], online across [state]" beats "compassionate, holistic care for your journey." (Worked example: Anxiety therapist website.)
2. Give each specialty its own real page. If you treat anxiety, trauma, and couples, that's three pages that each describe who it's for and what treatment looks like — not one bullet list. Small tests suggest dedicated specialty pages correlate with being recommended; they also just serve the reader better.
3. Keep your name, address, and phone identical everywhere. Your site, your Google Business Profile, and every directory should agree, character-for-character. Mismatches make you harder to verify — for a person and, apparently, for an AI drawing on multiple sources.
4. Be present on directories the assistants tend to read. A single profile is thin. Consistent listings across the directories AI is observed to pull from build the cross-source agreement these systems seem to favor.
5. Say your approach plainly — if it's true. If you genuinely practice CBT for anxiety, name it. Don't manufacture modality claims to look AI-friendly; a made-up specialty is both an ethics problem and, eventually, a recognition problem.
6. Add basic structured data (schema) so machines can parse the facts. Marking up who you are, what you treat, and where lets a machine read your practice details cleanly. It's a modest technical step, not magic.
7. Earn honest mentions on other credible pages. When reputable sites reference you, the cross-source signal strengthens. This is slow, real, and unbuyable — which is exactly why it holds up.

Notice what's not on this list: tricks, hidden text, or "prompt-stuffing" your site. Those tend to age badly and can undermine the trust these systems are built to reward. The durable move is clarity.

How to check whether AI recognizes your practice

Two ways — a quick DIY test and a structured one.

The DIY test (10 minutes). Open ChatGPT, Perplexity, and Google's AI mode, and ask each the way a client would: "Find me a therapist for [your specialty] in [your city] who's taking new clients." Run it a few times. Note whether your practice appears, and — just as telling — whether what the assistant says about you is accurate. Then flip it: ask "What does [your practice name] specialize in?" and see if the answer matches reality. One honest caveat: these results are unstable by design. You're looking for a pattern across several runs, not a verdict from one.

The structured check. The free Recognition Score is a deterministic, rule-based scan (no AI in the scan path) that reads your homepage the way a stranger would and reports what's clear and what's missing — who you help, what you treat, where you are, what to do next. It also reports an AI-Readiness signal: whether your page exposes clear, extractable specialty, population, and location details. One precise limit, so we don't overclaim: AI-Readiness measures whether those signals are present and machine-readable on your page — it does not test whether any particular assistant will actually retrieve or recommend you. No one can test that reliably from the outside. What it can do is tell you whether you've given the machines something clear to work with.

Want the guided version? The free First Hour guide walks you through your first hour of recognition fixes, step by step.