A parent asks ChatGPT which ABA clinic near them takes their insurance. A clinic owner asks Perplexity to compare practice-management platforms. Both get a written answer with a short list of sources under it. This page explains, mechanically and without hype, what determines who is in that list - and what ABA Rank publishes so that your listing can be.
The short version: AI assistants answer questions about ABA providers the same way they answer anything else: they retrieve pages, read them, and cite the ones they used. Being retrievable is therefore a publishing problem, not an advertising one. ABA Rank publishes every claimed listing as schema.org structured data, as plaintext markdown at a stable URL, through a public JSON API and an MCP server, and in an llms.txt index - and explicitly allows the retrieval agents behind ChatGPT, Claude, Perplexity, and Google AI Overviews to read all of it. No directory can promise that a model will name you. What ABA Rank can do is make sure that when one goes looking for an answer about ABA, there is a current, structured, attributable page about your organization for it to find.
None of this is a prediction. It is a description of how a retrieval-augmented assistant assembles an answer, and what that mechanically rewards.
A parent typing "best ABA clinic near me that takes Aetna" into a search box used to get ten links and pick one. Increasingly they get a written answer with three or four sources named underneath it. The click, if it happens at all, happens after the recommendation. Whoever is in that answer got the referral; whoever ranked eleventh never entered the conversation.
A retrieval system picks sources on mechanics: can it fetch the page, is the content in the HTML rather than assembled by client-side JavaScript, are the facts marked up in a schema it recognizes, is there a date on it, does anything corroborate the claim. A beautiful site that hides its facts in an image carousel is, to a retrieval agent, a blank page.
Assistants lean on sources that state many comparable facts in one predictable shape: name, location, services, insurance accepted, credential, rating, last updated. That is what a directory is. It is also why a listing you control on a structured, ABA-specific index does more for AI visibility than another paragraph of prose on your own homepage.
Retrieval systems weight corroboration: an independent page that says the same thing your homepage says is worth more than your homepage saying it twice. A profile on a third-party index that publishes its ranking formula, dates its data, and links back to you is exactly that corroborating source.
Most directories publish a web page and hope. These are the formats retrieval systems actually consume - and each link below opens the real thing, right now, for a listing that costs nothing to claim.
Clinic profiles emit LocalBusiness + MedicalBusiness, software emits Product with SoftwareApplication, services emit Organization + Service - each with AggregateRating, Review, address, geo coordinates, areaServed, and audience. This is the vocabulary Google AI Overviews and every major assistant already parse. It is on your profile whether or not you ever write a word of marketing copy.
Every public profile is also served as clean markdown at /api/vendors/{slug}/md - facts, the AI-generated review summary, and sample reviews, with no navigation, no scripts, and no markup to strip. An agent that follows this URL gets the whole profile in one cheap, unambiguous fetch.
The emerging convention for telling an assistant what a site contains and where the citable material lives. /llms.txt is the index; /llms-full.txt is a single plaintext bundle of top vendors, the ranked methodology, every FAQ block on the site, and recent editorial - built for one fetch rather than a crawl.
Anonymous reads work against /api/v1 - vendors, clinics with radius geo-search, categories, articles, unified search, and the ranking weights themselves at /api/v1/methodology. An agent building an answer can query the index directly instead of scraping a page and guessing.
A Model Context Protocol endpoint exposing nine read tools - search_vendors, get_vendor, search_clinics, get_clinic, list_categories, list_subcategories, get_articles, search_all, get_methodology. Any MCP-capable assistant can connect and query the ABA directory as a first-class tool, with no scraping in the loop.
Most sites silently inherit whatever a wildcard rule implies. ABA Rank names 15 AI user-agents individually and grants each the same access - including the 6 live-retrieval agents that actually generate citations. The read-only API prefixes are carved back out of the /api disallow so agents can reach them.
Edits are pushed to participating search engines on change rather than waiting for an organic crawl, and the Index recomputes nightly. Freshness is a retrieval signal: an assistant asked a question today prefers a source dated this month over one dated three years ago.
Two different jobs hide behind the phrase "AI crawler". Training crawlers build the corpus a model learns from. Retrieval agents run while the answer is being written - those 6 are the ones that produce a visible citation, and they are the ones worth caring about.
Fetch during an answer. These produce citations.
OAI-SearchBotChatGPT-UserClaude-SearchBotClaude-UserPerplexityBotPerplexity-UserBuild the corpus. No link, but they shape what a model knows.
GPTBotClaudeBotanthropic-aiGoogle-ExtendedApplebot-ExtendedAmazonbotmeta-externalagentCCBotBytespiderA bot that matches its own user-agent group ignores the wildcard rule entirely, so the full policy is repeated for each. Check it yourself at /robots.txt - the page you are reading renders the same list the policy is generated from.
Ordered by leverage, not by price. The single highest-return action is the one that costs nothing: claiming the listing so the facts an assistant repeats are facts you chose.
An unclaimed profile is still indexed, but it carries only what our editorial team could verify from public sources. Claiming is what lets you correct the facts an assistant will repeat - and a wrong service line or a closed location gets quoted back to a parent exactly as confidently as a right one.
Completeness is worth real Index points, but the bigger effect is on retrieval. Each field you fill - insurance plans accepted, age bands, languages, modalities, service areas, credentials - is one more question an assistant can answer about you specifically. Empty fields are queries you silently lose.
Reviews are the largest single component of the Index, and they are also the part assistants quote most directly - an AI summary of what actual clients said is far more citable than a self-description. Reviews are moderated and, where possible, tied to a completed verification workflow.
ABA Rank Verified is a human review of identity, ownership, credentials, profile evidence, and continuity, re-checked annually with the date shown publicly on your profile. Retrieval systems weight corroboration and recency; a dated, independently checked fact is the strongest form both of those signals take. This is the one step on the list that costs money, and the only one that changes what a crawler can independently confirm about you.
Every tier ships the same ingestion layer. What the paid tiers move are the two things a retrieval system actually weights once it can read you: whether your facts carry an independent, dated check, and where you sit in the ranked lists it reads.
Everything in the machine-readable layer above ships on the free tier. Claiming is what makes the facts an assistant repeats facts you chose, and completeness is worth 25 Index points on its own. If you do nothing else on this page, do this one.
Retrieval systems weight corroboration and recency. Verification is a human review of identity, ownership, credentials, profile evidence, and continuity, and it publishes the result as a date on your profile - "VERIFIED MARCH 2026", visible to any crawler that reads the page. It also unlocks the 15-point verification-recency component of the Index, which is an operational signal, not a purchased one.
When an assistant answers "best ABA billing companies," it retrieves a ranked page and paraphrases the top of it. Position decides who is in that paraphrase. Sponsor carries the full 15-point tier signal - the maximum rank influence money can have here, published and capped, against 85 points it cannot touch.
None of this buys a citation, and none of it buys readability - the ingestion layer is identical on all three tiers. Upgrading changes two things a retrieval system genuinely weights: whether your facts carry an independent, dated check, and where you sit in the ranked lists an assistant reads. 85 of the 100 Index points remain unpurchasable at any tier. See the full formula or compare every tier benefit.
An honest version of this page has to include the part that does not sell. Here it is.
No one can guarantee a citation. Model outputs are non-deterministic, retrieval sets change without notice, and every assistant weights sources differently. Any vendor selling you guaranteed placement in an AI answer is selling something they do not control.
Paying us does not buy a different ingestion path. Every surface on this page - structured data, the markdown profile, the API, the MCP server - covers free Claimed listings identically to Sponsor listings. What the paid tiers change is the strength of two signals inside that layer: verification puts an independently checked, publicly dated fact on your profile, and the tier signal moves rank position within the lists assistants retrieve. The tier signal is capped at 15 of 100 Index points and disclosed wherever it applies. Neither is a citation.
Sponsored placement is labeled, and being labeled is the point. Sponsored slots render above the ranked list with a visible chip and never reorder it. An assistant reading that page reads the chip too. Buy the Sponsor tier for reach and for the capped rank signal, not in the expectation that a model will mistake a paid slot for a ranked one.
This is not a substitute for your own site. The strongest pattern is corroboration: consistent facts on your site, on your Google Business Profile, and on an independent index that dates its data and publishes its method. We are one leg of that, not all three.
We do not attempt to manipulate model output. There is no prompt injection in our pages, no hidden text, no cloaking for crawlers. What an assistant reads is what a person reads.
AEO, GEO, RAG, grounding. Consultants charge for these words. They describe fairly simple mechanics, and you should know them well enough to tell a real proposal from a repackaged SEO retainer.
Short, self-contained answers - written so an assistant can quote one with attribution and a link back here.
Claiming puts your facts into every surface on this page at no cost. Verified adds the two things a retrieval system weights on top of readability - an independent check, and a public date on it - and moves you up the ranked lists assistants read.