A new question has started showing up in therapist marketing conversations: how do I get ChatGPT to recommend me. It's a reasonable thing to want to know, and it's currently attracting a lot of overconfident answers.
Some of what circulates under the label GEO, generative engine optimization, or AEO, answer engine optimization, is genuinely documented by the companies building these systems. A larger share is speculation dressed up as expertise, sold with the same confidence regardless of which claim it's attached to. This article tries to keep those two categories separate, because conflating them is where most of the bad advice comes from.
Four categories, used consistently throughout: documented, meaning a company's own current technical documentation says so directly. Reasonable but unproven, meaning it's a sensible inference with no direct confirmation either way. Unsupported, meaning it's circulating as advice with no documented basis at all. And contradicted by official documentation, meaning a company has said the opposite of what's being claimed.
What AI search actually is
Conventional search returns a list of links. AI-assisted search can instead generate a synthesized answer, sometimes citing or linking to the sources it drew on. Google AI Overviews, Google AI Mode, and ChatGPT Search all work roughly this way, though each implements it differently. The interface changes. The underlying need, for the system to find reliable, accurate source information somewhere, does not go away, it just gets one more layer between the source and the reader.
Can ChatGPT actually recommend a specific therapist?
ChatGPT can use web search to surface and cite public information when search is invoked for a query. Whether any particular therapist's page gets surfaced for any particular question depends on factors OpenAI hasn't published a formula for, and there is no legitimate way to guarantee a specific outcome. Anyone selling a "rank #1 in ChatGPT" service is selling something that doesn't correspond to a documented mechanism. Unsupported, and worth being direct about with clients who ask.
A more realistic goal, and the one the rest of this article works toward, is making a practice easy for these systems to discover, understand, and corroborate. That's a lower bar than guaranteed recommendation, and it's also the only bar these companies have actually described a path toward.
Does conventional SEO still matter for AI search
According to Google, yes, and unambiguously. Its generative AI optimization guide states that its AI-powered search features are rooted in the same core ranking and quality systems as conventional Search, and that foundational SEO practice continues to apply (Google's guide to optimizing for generative AI features). Documented. Crawlable pages, useful and original content, clear site structure, accurate business information, sound internal linking, legitimate topical authority: all of it still matters, because none of it stopped mattering when the interface changed. For Google specifically, AEO and GEO are not a separate discipline layered on top of SEO. They’re SEO, evaluated by systems that happen to generate a synthesized answer instead of only a list.
What Google explicitly says you do not need
This is the most immediately useful part of Google's current documentation, because it directly contradicts several claims already circulating in therapist marketing. The same guide states plainly that no special markup is required to appear in its generative Search experiences, and specifically addresses llms.txt files, unnecessary content chunking into short segments, invented "AEO schema," and manufactured web mentions as unnecessary or counterproductive. Documented, and it's worth reading that guide directly if you want the primary source rather than a summary of it.
So is structured data pointless
No, and this is a distinction worth holding carefully rather than collapsing into either "you must have it" or "it does nothing." Structured data remains legitimate, ordinary SEO infrastructure. Google uses it to better understand information on a page and to potentially make content eligible for certain rich result features, but valid structured data never guarantees a specific search appearance (intro to structured data). A therapist site can appropriately use Article, BreadcrumbList, Person, and relevant organization markup where accurate. There is no invented "AEO schema" type to add on top of that. Documented that standard schema helps in the ordinary way it always has; contradicted by official documentation that a special AI-specific schema exists or does anything.
What OpenAI documents for ChatGPT specifically
OpenAI's technical documentation is more concrete here than Google's, and gives an actual action to take rather than a description of a philosophy. OpenAI operates a crawler called OAI-SearchBot specifically for ChatGPT search, and states that sites wanting their pages available in ChatGPT search results should allow it; sites that block it will not appear in ChatGPT search answers, though they may still surface in some non-search contexts (OpenAI crawler documentation). Documented, and checking your site's robots.txt for this specifically is a genuinely useful five-minute task, not busywork.
OAI-SearchBot and GPTBot are not the same crawler, and this trips people up constantly
OAI-SearchBot governs ChatGPT search visibility. GPTBot is a separate crawler associated with gathering data that may be used to improve OpenAI's underlying models. OpenAI's documentation is explicit that these are controlled independently through robots.txt: a site can allow OAI-SearchBot for search visibility while disallowing GPTBot if the owner doesn't want crawled content used for model training (OpenAI crawler documentation). So the claim "you must allow AI training if you want ChatGPT to show your site in search" is contradicted by official documentation. The two settings are genuinely separate, and a therapist can decline the training use without losing search visibility.
Building this out: five layers, in the order they actually matter
Accessibility comes first, because nothing else works without it. For Google, pages generally need to be crawlable and indexable to participate in Search at all, including its AI features. For ChatGPT, that means checking OAI-SearchBot isn't blocked. If the whole site is inaccessible to a given crawler, no amount of polish further down the list changes anything.
Clarity is second, and it's less about writing "for machines" than it sounds. Compare two openings. One reads: Find the courage to become who you were meant to be. No location, no designation, no service stated anywhere near it. The other reads: Jane Smith, Registered Psychotherapist. Anxiety and burnout therapy for adults. In-person in Toronto, virtual across Ontario. The second gives any retrieval system, human or otherwise, a set of plain facts to work with. This is the same clarity argument made elsewhere for prospective clients, and it isn't a coincidence that it applies here too.
Useful service pages are third. A real page for an actual service, covering the therapist's designation, who it's for, the approach, location and format, fee, and how to get in touch, does double duty: it's a normal good SEO page, and it's explicit source material a retrieval system can actually draw from. There's no separate version of this page you'd write differently "for AI."
Third-party corroboration is fourth, and it's where restraint matters most. A regulatory directory listing, a professional association membership, a legitimate clinic profile, real media coverage, these all give outside confirmation of who a therapist is beyond what the therapist's own site claims. What doesn't belong here is a campaign to manufacture "AI mentions" through low-quality directories or paid placements. Google's guidance explicitly advises against pursuing inauthentic web mentions as a generative-search tactic (Google's generative AI optimization guide). Documented as something to avoid, which makes the "buy hundreds of mentions" advice contradicted by official documentation, not merely unproven.
Source-worthy content is fifth, and it's genuinely the hardest one, because it's the one place a template can't do the work. "5 Tips for Managing Stress" is competing with tens of thousands of interchangeable pages saying roughly the same thing. Something like "Virtual vs In-Person Anxiety Therapy in Ontario: How to Decide" is answering a specific question a specific way, which is closer to what "worth citing" actually requires. Google's current guidance specifically emphasizes unique, valuable content over commodity content in this context (Google's generative AI optimization guide). Documented.
The specific claims circulating in therapist marketing, checked
"You need an llms.txt file." Google's documentation states directly that it does not use llms.txt as a mechanism for generative Search visibility. Contradicted by official documentation, for Google specifically. Other AI products may make different decisions over time, and this is worth re-checking rather than assuming it holds everywhere.
"You need special AEO schema." No such schema type is documented by Google, and standard structured data's role hasn't changed for AI features. Contradicted by official documentation.
"Every answer needs to be 40 to 60 words." No company has published this as a citation requirement. It shows up frequently in GEO advice with no source attached. Unsupported.
"FAQ sections rank better in ChatGPT." A clear question-and-answer format is a legitimate writing tool because it's easy for a human to scan and gives an answer that stands on its own. That's a real readability benefit. It is not documented evidence of a ChatGPT ranking mechanism, and creating a hundred FAQ entries specifically to chase this is solving for something nobody has confirmed exists. Unsupported as a ranking claim, reasonable as ordinary good writing.
"You need hundreds of web mentions." Beyond being unsupported, this one is actively contradicted by official documentation, since Google specifically warns against pursuing inauthentic mentions as a tactic.
"Traditional SEO is dead, you need GEO instead." Google's own position is the opposite: its generative features are built on the same core systems as conventional Search. Contradicted by official documentation, at least for Google. This doesn't mean nothing has changed, the interface and some mechanics genuinely differ, but "dead and replaced" overstates what's actually documented.
"GPTBot must be allowed to appear in ChatGPT search." Addressed above: OAI-SearchBot and GPTBot are independently controlled. Contradicted by official documentation.
Does a Google Business Profile help with AI visibility
Google's generative features can draw on local business information, and Google continues to point businesses toward Business Profile where relevant (Google's generative AI optimization guide). But this doesn't change the underlying eligibility question. A profile still requires genuine in-person contact with customers during stated hours, and manufacturing one in pursuit of AI visibility carries the same suspension risk covered in the companion guide to Google Business Profile eligibility for therapists. Chasing AI visibility isn't a reason to bend a rule that exists independently of it.
Make geography and identity boringly explicit
Because therapy practice is provincially regulated and availability genuinely differs by jurisdiction, vague geographic language costs more here than it would for most businesses. "Online therapy available" leaves a retrieval system, and a prospective client, to guess. "Virtual psychotherapy for adults located in Ontario" states the actual fact, assuming it's accurate for the practice in question.
The same logic applies to professional identity. If one directory lists a therapist as "Jane Smith, RP," another lists "Jane Smith, RCC," and the website itself says "mental wellness expert," that's not variety, it's ambiguity that serves nobody. Accurate, consistent designation across every place a practice appears is a small, unglamorous task that does more for both AI systems and human readers than most of what gets marketed as GEO strategy.
Can any of this actually be measured
More than a year ago, and getting better. Google launched a dedicated Search Generative AI performance report inside Search Console in June 2026, giving impressions and visibility data specifically for AI Overviews and AI Mode appearances, currently rolling out to a subset of sites rather than universally available yet (Search Generative AI performance reports). Documented, though worth noting the rollout isn't complete, so a practice checking for it and not finding it yet isn't necessarily doing anything wrong.
For ChatGPT, OpenAI states that referral links from ChatGPT search include the parameter utm_source=chatgpt.com, which a practice can filter for in ordinary analytics (OpenAI Publishers and Developers FAQ). Documented. Neither of these gives full visibility into every time a practice might come up in conversation inside these tools, and that gap is exactly where third-party "AI visibility tracking" tools have proliferated. Treat claims from those tools with real skepticism; none of the underlying companies has documented an API or mechanism that would let an outside tool see the full picture these products claim to show.
What this actually adds up to
The genuinely useful part of all this is that very little of it asks a therapist to build something separate for AI systems. An accessible site, clear and specific language, real service pages, a legitimate professional footprint, content worth reading rather than worth scanning: all of it serves an actual prospective client first, and happens to be legible to a retrieval system as a byproduct rather than the other way around. Distorting a website to chase an unconfirmed AI ranking signal is very likely to make it worse for the person it actually needs to persuade, which is still a human being deciding whether to reach out.
Treat every specific AI-visibility claim, including ones not covered here, the way this article has treated the claims above: find the primary source, or don't repeat it as fact.
Last reviewed August 2026. AI search products and documentation change quickly; re-check the linked sources before repeating any claim here.