How Therapists and Mental Health Practices Get Recommended by AI
By the alphaa team — we run AI-visibility scans across thousands of businesses, including licensed clinical practices. This is marketing guidance, not legal, clinical or ethics advice; check anything below against your licensing board's advertising rules and your professional code before publishing. Last updated 8 August 2026.
Short answer: Therapists get recommended by AI when a specific and verifiabledescription of who they treat, how, and under what payment terms exists across the sources assistants actually retrieve — directory profiles, licence registries, insurance provider lists, specialist associations — and matches what is on their own site. The counterintuitive part is that the practice website is usually the least influential of those sources. And unlike most industries, the highest-leverage work here is almost entirely ethics-safe: it is factual precision, not persuasion.
What people actually ask
Nobody types "therapist" into ChatGPT. Every real query we see carries three or four constraints stacked together, because that is how someone in distress narrows a decision:
- "EMDR therapist in Portland who takes Aetna and does evening sessions"
- "affirming couples counsellor near me, sliding scale"
- "child psychologist who assesses ADHD, Spanish-speaking, accepting new patients"
- "online therapy licensed in both Texas and New Mexico"
Read those again as a machine would. Each one is a filter set: modality, population, location or licence jurisdiction, payment, language, availability. An assistant can only name you if it can verify enough of those filters from retrievable text. A page that says "compassionate, client-centred care in a warm, non-judgemental space" matches none of them. That sentence appears on tens of thousands of practice sites, and it is why so many perfectly good clinicians are invisible to AI search.
Where the answer actually comes from
When an assistant answers a therapist query, it retrieves live documents and synthesizes from what they say. In this category the retrievable set is unusually directory-heavy, roughly in this order of influence:
- Clinical directories. Psychology Today, Inclusive Therapists, TherapyDen, Zencare, Open Path and their regional equivalents. These are structured, densely filterable, heavily crawled, and they publish exactly the fields the question asks about. This is where most named recommendations come from.
- Licence and registry records. State board look-ups and national registries establish that you exist, in what discipline, in which jurisdictions, in good standing. This is the entity backbone — it is what lets a model treat "Dr. Rivera, LCSW" as a real, resolvable professional rather than a name on a website.
- Insurance provider directories. Payer-side listings resolve the single most decisive filter in the query. If your panel participation is not listed accurately, you lose every insurance-qualified question by default.
- Association and training rosters. Modality-specific bodies — EMDR certification registries, gottman referral lists, ABCT or BACP-type directories depending on your country. These are how a model verifies a modality claim rather than just repeating it.
- Your own site. Real, but mostly as the confirming source. It corroborates the directory picture and supplies detail; it rarely originates the recommendation.
This is the same multi-source consensus mechanism that governs every category — assistants trust facts that several independent sources state the same way. Our entity guide covers how that resolution works in general.
The five fixes, in order of leverage
1. Make one canonical fact set, then propagate it
Write down, once: legal practice name, clinician name with credentials exactly as licensed, licence type and number per state, address (or "telehealth only, licensed in X and Y"), phone, modalities, populations, languages, insurance panels, session fee or range, and current intake status. Then make every directory profile, your Google Business Profile, and your website agree word-for-word on the checkable fields. Fragmentation — three name spellings, two phone numbers, one stale address — is the most common reason a model hedges instead of naming you. Fixing it is unglamorous and it is the single highest-return hour in this list.
2. Fill every structured directory field, especially the unloved ones
Most clinicians complete the narrative bio and skip the checkbox fields — session length, sliding-scale availability, telehealth platform, accessibility, evening or weekend hours, accepting-new-clients status. Those checkboxes are precisely the structured data a retrieval step can match against a constrained query. A complete profile with a mediocre bio outperforms a beautiful bio with half the fields blank, every time.
3. Write one specific page per modality and population
Replace the single "Services" page with pages that answer a real question: what EMDR sessions involve and who they suit, what a perinatal-mood intake looks like, how ADHD assessment works for adults and what it costs. Lead each with a short definitional paragraph a model can lift verbatim — this is the core of writing content AI engines actually quote. Say plainly who the approach is not right for; scope-limiting language reads as expertise to a human and as precision to a machine.
4. Publish fees and access terms as plain text
"$180 per 50-minute individual session; sliding scale from $95 for six concurrent slots, currently two available; in-network with Aetna and Cigna; superbills provided for out-of-network" answers several filters at once and is the kind of sentence assistants quote. Buried, image-based or omitted fee information hands the answer to whoever published theirs — the same dynamic we cover in how pricing pages shape AI recommendations.
5. Keep intake status current
"Accepting new clients" is a freshness signal with an unusually short half-life, and it is weighted heavily because the query almost always implies availability. Stale "accepting" flags waste referrals and erode the accuracy that earns you the next one. Update directory status monthly, and date-stamp the waitlist line on your site.
What not to do
This category has real ethical and legal constraints, and the AEO advice that ignores them is dangerous. Do not solicit or incentivise client reviews — most codes of ethics restrict testimonials from current or recent clients precisely because of the power imbalance, and no visibility gain is worth a board complaint. Do not claim a modality you are not trained in; the certification registries that verify modality claims are the same ones a model checks. Do not imply guaranteed outcomes, and do not publish anything that could identify a client, including composite case studies detailed enough to be recognised. Do not let a marketing vendor put condition-treatment claims on your site that your licence or local advertising rules would not support.
Common questions
Does my practice website matter at all?
Yes — as corroboration and depth. It is where a model verifies that the directory claims hold up and finds the specifics no directory field captures. It is simply not usually the source that produces the first recommendation.
I am telehealth-only across several states. How should that be handled?
State it explicitly and per-jurisdiction: "licensed in Texas (LPC #XXXXX) and New Mexico (LPCC #XXXXX); telehealth only." Cross-state licence questions are frequent and models answer them conservatively — vague geography usually means exclusion rather than a guess in your favour.
How long before this shows up in answers?
Directory and profile corrections can surface within days because those pages are re-retrieved live; content and third-party signals build over weeks to months. Results vary by engine, by phrasing and over time, and no one can guarantee you a place in any assistant's answer — see is AEO real for the honest limits.
The bottom line
Therapists lose AI recommendations to vagueness, not to competition. Say which modalities you practise, which populations you serve, where you are licensed, what you charge, whether you are open — then make every public record agree. That is the same information a person in a difficult week needs in order to pick up the phone. Writing it down clearly serves both readers at once.