toolkit

Scenario · Content and search

Are we showing up in ChatGPT?

“I asked ChatGPT for a surgeon in our city and we weren't in the answer.”

Not showing up when someone asks an AI assistant for a recommendation is the newer version of not ranking on a search results page, and it comes with much weaker levers to pull, because the same question asked twice can genuinely return a different answer. This measures it properly — a fixed set of real prospect questions, asked repeatedly over time rather than once anxiously — and checks the underlying signals AI engines actually read, like whether the practice's own machine-readable details are correct and consistent, while being honest that none of this is a guarantee any specific answer changes.

What to ask for

AI search visibilityAI Search Visibility report — measures how visible a business is inside AI search answers (ChatGPT, Gemini and other engines) and builds a client-ready report.On-page SEOOn-page SEO scan — reads a page's delivered HTML and checks the head, meta and heading signals search engines use: title presence and length, meta description, rel=canonical, meta robots (noindex), Open Graph tags, viewport, html lang, and exactly one H1, plus a structured-data summary.Structured data (schema)Structured-data audit — checks a page's JSON-LD against a per-type rubric: valid JSON, a schema.org context, and the required and recommended properties that gate rich results for Organization, WebSite, Article, Product, LocalBusiness, BreadcrumbList, FAQPage, Event, Recipe and more.Local SEO (NAP)Local SEO / NAP audit — reads a page's delivered HTML and checks the local-search signals a business page lives or dies by: LocalBusiness structured data with a complete Name-Address-Phone, a phone and street address on the page, NAP consistency between the schema and the visible page (a mismatched phone fails the page), and Google Business Profile readiness — opening hours, geo coordinates, an embedded map.

See it work

A real run of Structured data (schema):

FAIL — 1 page(s) · 0 type(s)
      [error] invalid JSON-LD: Expected double-quoted property name in JSON at position 59 (line 1 column 60)

  playbook: local fallback

  report: ./out/structured-data-schema-structured-data-schema.html
  text:   ./out/structured-data-schema-structured-data-schema.txt

The captured report, exactly as a run hands it to a client —open the full report ↗

A newer version of “why aren’t we ranking”, with the same emotional weight and much weaker levers. What makes it tractable is measuring it properly: real prospect questions, repeated, recorded — not one anxious query someone ran once.

  1. Write the questions a real prospect would ask. Procedure plus city, “best X near me”, cost questions, “who should I see for”. Not the practice’s name — a query containing the name tests nothing.
  2. Measure it. The AI visibility check puts those questions to the engines with live search, and scores whether the practice is named, cited and where it sits in a list, plus which competitors take the share of voice. Run it as a baseline before changing anything.
  3. Check the things the engines actually read. On-page structure, and the structured data — a LocalBusiness or MedicalBusiness block with correct name, address, phone and services is how a machine knows what this practice is and where. Missing or invalid schema is the most fixable input.
  4. Check name, address and phone consistency across the site. An engine reconciling three different phone numbers has no reason to trust any of them.
  5. [manual] Fix the inputs, and set expectations honestly in the same breath. Schema, consistency, clear service pages that answer the question directly. Then say plainly that these are inputs and not controls — a practice told otherwise will ask why the answer has not changed next week.
  6. Re-measure on a schedule, not once. Monthly against the same question set. The deliverable is a trend, because a single run of a non-deterministic system is an anecdote.

The same question twice gives different answers

Which is why step 1 and step 6 matter more than any fix in step 5. Without a fixed question set measured repeatedly, there is no way to distinguish work that helped from a model that shifted, and the whole conversation becomes untestable.

What this does not cover

Any guarantee of movement. AI answers are non-deterministic, personalised, and change when the model changes — the same question twice can return different practices, and nothing here can promise that fixing the site changes an answer. It measures what the engines say now, against real prospect questions, so the change over time is observable rather than asserted.

← All scenarios