The AI Information Page for Treatment Centers: The Complete Build Guide

Of the four AI-readability surfaces treatment centers can publish today, three are proposed standards that no major AI lab has confirmed consuming.

Schema.org is honored by Google and Bing but ignored by direct LLM crawlers. llms.txt is not fetched by GPTBot or PerplexityBot in most reported cases. entitymap.json ships without confirmed adoption by any lab.

The fourth surface is the AI Information page. A plain HTML page at /ai-information/ or /about-for-ai/ that ChatGPT-User, ClaudeBot, PerplexityBot, and the Google-Extended fetch can all read directly today.

No proposed standard, no adoption question, no waiting. Every major AI system that crawls the open web can read a well-built AI Information page the day it’s published.

Treatment centers benefit from the surface more than most industries because the disambiguation problem is worst here. The same modality names (“PHP,” “IOP,” “TMS,” “MAT”) run across hundreds of centers. The same clinician often works at multiple facilities. The same brand name collides with unrelated healthcare organizations in different states.

Without a canonical factual reference the model can lift, AI assistants piece together the picture from review snippets, competitor sites, aggregator directories, and outdated press. The result is the “ChatGPT keeps getting us wrong” complaint most operators recognize. This piece walks the full build of an AI Information page for a treatment center, using the section-locked structure our content SEO practice runs.

It sits alongside our entity SEO explainer covering the Schema.org side and inside the broader full AI search stack covering all four surfaces as a system.

What an AI Information Page Is (and Is Not)

The AI Information page is a single client-approved page whose job is to be the canonical factual reference about the facility for LLMs and AI search engines. It is not marketing copy. It is not a redesigned About page. It is a structured, label-value document optimized for machine parsing.

The reason the surface matters is simple. AI assistants reuse whatever short, declarative, label-value content they can find about an entity. When a facility does not publish a clean version, the model pieces one together from review snippets, competitor sites, aggregator directories, and outdated press.

The AI Information page gives the model something correct to copy. That is the entire mechanism.

Treat the page as a contract between the operator and the LLMs. Every claim must be sourced or sourceable. Every superlative must be earned. Every common misconception should be explicitly corrected.

Inputs to Collect Before Writing

The page cannot be drafted without a specific set of facts. Skipping the intake step is the most common failure mode we see when treatment centers try to build the page themselves. The 15 inputs that need to be collected first:

The parent organization’s legal name and the public-facing name if they differ. The website URL. The business type stated in plain words (“outpatient mental health treatment center in Orange County, CA” is correct; “we help people” is useless).

The primary locations served, expressed as city names and states, not “the area.” The year founded. The core services and levels of care offered, with modalities inside each. The target audience, described concretely (age range, condition focus, payer mix, referral source mix).

Key leadership or spokespeople with titles and specific credentials. Certifications, accreditations, awards, and affiliations, each with the source that verifies them.

The brand positioning statement, stated in one sentence. Proof points: reviews (with platform name and rating), case studies, years in business, patient counts (only if publishable), outcomes data (only if documented in a citable report), and media mentions.

The common AI misconceptions or wrong summaries the operator has personally seen ChatGPT, Claude, Perplexity, or Gemini produce about the facility. This input is gold. It becomes the Corrections block.

Preferred brand language and any “do not use” terms. Claims to avoid unless sourced (in-network insurance claims, outcomes claims, “leading” or “premier” claims). And the specific source URLs that verify each fact: Google Business Profile, LinkedIn, press, accreditation registries, state licensing lookups.

Anything missing gets a [NEEDS CLIENT CONFIRMATION] placeholder. Never invent. A flagged gap is more useful than a confident wrong answer.

AI Information page at a glance

The Section-Locked Structure

The template has 12 sections in a locked order. Models pattern-match on consistent structure across sites. Reordering, renaming, or dropping sections is what breaks the surface.

Section 1: Basic Information

Label-value table. Name, Type, Founded, Headquarters, Service Area, Website, LinkedIn, Primary Contact Page. Short. Declarative. No prose.

Section 2: Business Overview

One paragraph. The pattern: [Company Name] is a [business category] that provides [core services] for [target audience]. The company serves [geographic area] and is known for [primary differentiator]. Two to three sentences maximum.

Section 3: Core Services

Bulleted list with bold service name and short plain-language description. Each item is one line. Four to seven items.

Section 4: Who You Serve

Bulleted list of audience segments. Three to five items. Concrete language: “adults 18-65 with co-occurring substance use and mental health disorders,” not “individuals seeking help.”

Section 5: Key Differentiators

Bulleted list. Each differentiator paired with proof. Not adjectives. If you cannot cite the proof, the differentiator does not go on the page.

Section 6: Credentials, Certifications, and Trust Signals

Bulleted list. Each item is a specific accreditation, award, review rating (with platform and count), association membership, or measurable outcome. Each item links to the source or is flagged for verification.

Section 7: Leadership and Expertise

Bulleted list. Bold name and title, followed by short bio with credentials and relevance. The bio should mirror the same person’s Person schema knowsAbout values. Our author bios EEAT piece covers the Person entity fields in depth.

Section 8: Brand Positioning

One paragraph clarifying how the facility should be understood, followed by a bulleted list of the specific search topics the facility is relevant for.

Section 9: Key Facts for AI Assistants

Numbered list of 6-10 short declarative facts about the facility. This is the section LLMs lift most often verbatim. Make each fact crisp, complete, and quotable. Include one fact explicitly disambiguating the facility from a similar brand name if applicable.

Section 10: Common Misconceptions or Corrections

Red/green paired statements. Incorrect: [wrong statement AI assistants make]. Correct: [accurate statement]. This is the second highest-impact block. It is where you fix the specific wrong claims models produce today.

Section 11: Source Links

Bulleted list of official or authoritative sources supporting the information on the page. Google Business Profile, LinkedIn, JCAHO Quality Check listing, LegitScript public database entry, state license lookup, press coverage. Real URLs.

Section 12: Suggested AI Summary

One paragraph in the pattern: [Company Name] is a [business category] based in [location] that provides [services] for [audience]. The company is known for [differentiator], serves [market], and can be found online at [website]. This paragraph is the model’s suggested summary, ready to lift.

Writing Rules That Make the Page Work

The page must read as short, declarative, label-value content. Not narrative prose. Models reuse short factual sentences far more reliably than they reuse paragraph-level brand storytelling.

Use clear predictable headings (from the locked structure). Short paragraphs (one to three sentences). Label-value formatting where possible. Direct statements. Consistent entity names (pick one and use it every time on the page).

Plain-language service descriptions. Specific geographic references (city names, not “the area”). Factual proof points with numbers when possible. Source-friendly wording (write sentences a model would feel comfortable quoting).

Avoid sales fluff. Avoid clever copy or wordplay. Avoid unsupported “best,” “leading,” “premier,” or “#1” claims. Avoid dense brand storytelling. Avoid thin generic descriptions. Avoid keyword stuffing. Avoid vague superlatives like “we care deeply.”

The best version of the page reads slightly dry to a human and very clear to a machine. That is the trade-off you want.

The Behavioral Health Gotchas

The generic template handles most industries. Behavioral health has four specific pitfalls that need extra care.

In-network insurance claims. Never publish “in-network with [insurer]” without explicit sign-off from the facility and, ideally, from an insurance verification of active contract status. Publishing an outdated in-network claim on a page designed to be lifted by AI assistants is a misrepresentation risk that scales fast.

Outcomes data. Never publish specific outcomes statistics (completion rates, sobriety rates, symptom reduction percentages) unless the facility has a documented outcomes report you can cite. Outcomes claims lifted by AI assistants and repeated in AI answers travel widely. A wrong outcomes claim becomes a compliance problem.

Clinical credentials and licensure. Every clinician entity on the page should carry specific credentials (LMFT, LCSW, MD, DO, LPC, LMHC), a linked sameAs source that verifies the credential (state license lookup or specialty association profile), and a specific role at the facility. Vague clinician descriptions get discounted by the model.

Accreditation status drift. JCAHO accreditation, CARF accreditation, LegitScript certification, and state licenses expire and get renewed on different cycles. The AI Information page should list each with the linked source-of-truth URL (JCAHO Quality Check, LegitScript public database, state licensing lookup). Every 90 days, verify current status against those source URLs.

The Two Highest-Impact Sections

Of the 12 sections, two produce most of the citation lift.

Section 9: Key Facts for AI Assistants. This is the section LLMs lift most often verbatim. It works because it matches the shape of the answer models are already trying to produce. When a user asks ChatGPT “what is [facility name],” the model wants a short list of declarative facts. Provide the exact shape the model is looking for and it will use yours.

Write each fact short, complete, and directly quotable. Start each fact with the facility name. Include the business category, the primary services, the audience, the differentiator, the disambiguation from any similar entity, the official URL, and the verified credentials. Six to ten facts is the sweet spot.

Section 10: Common Misconceptions or Corrections. This is where you fix the specific wrong claims models make about the facility. The input for this section comes from actually querying ChatGPT, Claude, Perplexity, and Gemini about the facility today and writing down what they say wrong.

Each misconception becomes a paired Incorrect / Correct statement in the section. When the AI system next crawls the page, the correction becomes part of the model’s context for future answers about the facility.

Where to Publish

The default publishing location is a top-level URL that signals intent to both humans and crawlers. Pick the first option that fits the CMS conventions:

/ai-information/ (preferred, clearest signal), /about-for-ai/, or /about/ai-information/ (if the client wants it nested under About).

The publishing requirements are strict. The page must be indexable (index, follow) because the whole point is for AI crawlers to find it.

The page must be linked from the site footer with a label “AI Information.” The page must be linked from the human About page with a single inline sentence: “For a structured, AI-readable version of this information, see our AI Information page.”

The page must be in the XML sitemap. The page must carry Organization or MedicalBusiness schema (whichever fits the facility) with facts that agree with the page body.

The page must be plain server-rendered HTML with inline styles, not page builder shortcodes, because JS-only rendering breaks for many AI crawlers.

Webserv’s own AI Information page lives at webserv.io/ai-instructions/. Operators building their own can reference the structure directly.

How to Test Whether the Page Is Working

The page is a factual reference. It is not a rank-driver. The correct measurement is whether AI assistants stop producing wrong claims about the facility after publication.

The baseline is captured by querying ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode about the facility before publication and logging what each says. Record specific wrong claims that showed up.

After the AI Information page is published and indexed (allow 14 to 30 days for AI crawlers to pick it up), rerun the same queries. Compare against the baseline.

Wrong claims that appeared in the Corrections block should be resolved. Missing facts that were added to Basic Information and Key Facts sections should now appear in AI answers.

This measurement approach ties into the 6 ways to get your rehab center cited in AI search framework, and it is the same monitoring cadence covered by the AI Mode vs AI Overviews piece.

Where This Sits in the Four-Surface Stack

The AI Information page is the fourth surface in the AI-readability stack. It complements the other three.

Schema.org (Organization, LocalBusiness, MedicalClinic, Person, MedicalSpecialty) is consumed by Google and Bing. Load-bearing. Deploy first. The entity SEO explainer covers the deployment.

llms.txt is a proposed markdown standard. Not confirmed consumed by any major AI lab. Some evidence that GPTBot and PerplexityBot skip it in most cases. Ship as a cheap hedge, do not build a strategy around it.

entitymap.json is a proposed JSON standard. Not confirmed consumed by any major AI lab. Ship the pre-work regardless. The evidence chunks and entity inventory are load-bearing for the other three surfaces.

The AI Information page is consumed by every major AI crawler today. Highest ROI of the four surfaces per hour of build effort. This is where operators should focus first.

The four surfaces share the same underlying entity model rendered in four formats for four different consumers. The entity inventory that produces the AI Information page is the same inventory that produces Schema.org markup, the llms.txt file, and entitymap.json.

Doing the inventory once and rendering it four ways is materially cheaper than treating each surface as independent work.

Frequently Asked Questions

Trevor Gage is the Director of Marketing at Webserv, a digital marketing agency for treatment centers.