AEO audit for an outpatient physical therapy group

The company

Self-referral is possible because all 50 states, the District of Columbia, and the U.S. Virgin Islands allow some form of direct access to physical therapists, with limits that differ by jurisdiction (American Physical Therapy Association; data as of July 2025). Patients also ask assistants: 32% of U.S. adults had used AI for health information or advice (KFF poll of 1,343 adults, early 2026), and 47% of 992 adults surveyed by a reputation-software company had used AI to research healthcare providers (April 2026). No study found for this example measures how assistants choose physical therapy clinics.

The questions patients ask assistants

The panel follows AEO HQ's measurement design: 40 patient intents, each written three ways (120 unbranded prompts), plus 20 branded prompts. Local prompts name a suburb near a clinic, and runs use a connection inside the metro, because ChatGPT may use an approximate location based on the user's IP address. Prompts come from generic questions, never from patient records.

Each prompt runs three times a week on each of the four assistants the audit measures (ChatGPT, Perplexity, Gemini, and Google AI Overviews), in a new chat with a clean session; the ten most important prompts run eight times a week. Results are reported as rates with error ranges, not as single answers, because ChatGPT and Google's AI returned the same list of brands less than once in 100 repeated runs (vendor study; 2,961 runs). The table shows 12 of the 140 prompts.

How the assistants reach the clinics

An assistant can cite a clinic page through search only after its search system has crawled and indexed it. The diagrams show the documented routes; how AI assistants find and cite sources has the details.

• AI Overviews and AI Mode link to pages that are indexed and eligible for a snippet, and Google says Business Profiles can help a business be visible "in both AI responses and other Google Search results".
• Microsoft says Bing and Copilot "rely on the same core crawling, indexing, and ranking foundation as traditional search", so pages must be in Bing's index.
• Sites that block OAI-SearchBot "will not be shown in ChatGPT search answers"; Claude-SearchBot and PerplexityBot play the same role for Claude and Perplexity.
• Anthropic lists Brave Search as a web-search subprocessor, and Brave's crawler does not crawl a page that Googlebot may not crawl.

Directories carry weight: in 6.8 million citations studied by a listings-management company, listings were 52.6% of healthcare citations, "with WebMD and Vitals and industry-specific directories dominating visibility" (ChatGPT, Gemini, and Perplexity; mid-2025). A strong Maps position does not carry over on its own: for dentists, the domains cited by Google and by a web-enabled AI engine overlapped by 11.9% (preprint).

What the audit checks

Crawling, indexing, and page text

• Crawler rules. robots.txt is read per crawler, since a crawler follows the group that names it and uses the * group only when none does. Search crawlers are checked apart from training crawlers (GPTBot, ClaudeBot, and the Google-Extended token), because OpenAI's settings are independent.
• Firewall and CDN. ChatGPT search needs the host or content delivery network to allow OpenAI's searchbot IP addresses.
• Index coverage of the nine clinic pages and 41 bios, in Search Console and Bing Webmaster Tools.
• Facts in the HTML. In Vercel's December 2024 data, none of the major AI crawlers rendered JavaScript, so addresses, hours, and insurers must be in the HTML the server sends.
• No doorway pages for suburbs without a clinic, a pattern Google's spam policies name as doorway abuse.

Listings and the group's facts

• Business Profiles. Google requires the same name across a chain's locations unless the real-world name varies, the same category where they offer the same service, and profiles only for public-facing practitioners who can be reached at the location, never one per specialization.
• Directories. Names, addresses, phones, hours, insurers, and PT credentials are compared with one fact sheet across Business Profiles, Healthgrades, Zocdoc, and insurers' directories.
• Structured data. Each clinic page uses Google's most specific LocalBusiness subtype, which on schema.org is Physiotherapy, matching the visible text. The audit does not treat markup as a citation lever: adding it to 1,885 pages brought no reliable gain in AI citations (vendor study).

Content that answers patient questions

Health is a "YMYL Health or Safety" topic in Google's rater guidelines, and Google's systems give more weight to strong E-E-A-T (experience, expertise, authoritativeness, and trustworthiness) on such topics. Each condition page should name the PT who wrote or reviewed it, link to a bio with license and credentials, show a date, and cite sources.

Rules that limit claims, testimonials, and reviews

Measurement that stays within HIPAA

HHS's bulletin on tracking technologies says trackers behind a patient login "generally have access to PHI," that a vendor receiving PHI needs a business associate agreement (BAA), that trackers on appointment-scheduling pages "may have access to PHI," and that a cookie banner is not a valid HIPAA authorization. A June 2024 court order vacated only the part covering an IP address linked to a visit to a public page about specific conditions or providers; HHS says it is evaluating next steps. Google offers no BAA for Google Analytics and says HIPAA-regulated customers "may only use Google Analytics on pages that are not HIPAA-covered".

The audit maps every tag on every page type and each vendor that receives data. It recommends measuring AI referrals only on pages the privacy officer and counsel clear, and counting appointment requests by source inside the scheduling system. AEO HQ receives monthly totals and no PHI.

Sample findings

These are sample findings for the hypothetical group, showing the form a finding takes. They describe no real clinic, and nothing in them was measured.

The deliverable

The report has nine parts, followed by a readout call.

1. Summary. Scope, dates, assistants tested, and the ten fixes to make first.
2. Method. The full panel, session controls, run counts, matching rules, and statistics.
3. Baseline measurement. For each assistant: mention rate, citation rate, share of voice against named competing clinics, and accuracy on branded prompts, each with a 95% interval; and the cited domains, sorted into the group's pages, directories, insurers, and publishers. No pooled score and no rank.
4. Crawler access and indexing. A table by user agent (allowed, blocked in robots.txt, blocked at the CDN), index coverage in Google and Bing, and the rendering check.
5. Listings and fact sheet. One fact sheet per clinic and a table of mismatches across profiles and directories.
6. Content gaps. Each intent mapped to a page or marked as a gap.
7. Claims and privacy items. Each with the rule text and a link, for the privacy officer and counsel to decide. AEO HQ flags these items; it gives no legal or HIPAA compliance advice and does not review clinical accuracy.
8. Priority fix list and measurement plan. Every finding with impact, effort, owner, and evidence, and the plan below.
9. Run log. Every answer and citation from the baseline, as a spreadsheet.

Measurement plan

The plan follows how to measure AI visibility: rates use Wilson intervals, which suit small samples, and a change counts only when the interval for the difference excludes zero.

Engagement timeline

The pricing page gives the audit's current delivery time; this example assumes about two weeks from completed intake.

What this example does not show

• Results. No rates, citations, traffic, or patient numbers, because the group does not exist. No study has measured how long changes like these take to show up in AI answers.
• Real prompts. A real panel comes from the group's intake calls, front-desk questions, and Search Console queries.
• Legal or clinical conclusions. HIPAA, practice acts, and payer rules are summarized, not applied; the privacy officer and counsel decide, and the group's clinicians judge clinical content.
• Evidence specific to physical therapy. The surveys and citation studies cited cover healthcare in general, and several come from companies that sell related services.
• Every assistant. Runs cover four assistants. Copilot and Google AI Mode appear only through Microsoft's and Google's reports, and Claude only through logs and referrals.

Next step

The audit's current scope, price, and delivery time are on the pricing page. The healthcare industry page covers AEO for healthcare more broadly, and the methodology page gives the full measurement design.