AI visibility audit example: a worked walkthrough

The business we’re auditing

The owner’s complaint is the one we hear most: “I asked ChatGPT for the best drain company in town and it never mentions us.” The audit’s job is to turn that frustration into a specific list of reasons and fixes — not vague advice about “doing more SEO.”

The five checks

Every audit on this site runs the same five checks, in the order an answer engine itself works: it has to find and read a page before consistent details matter, and details have to agree before corroboration counts. That order is the Crawl-Read-Cite Ladder — these checks sort onto its Crawl, Read and Cite rungs. None of this needs paid tools — a browser, the AI engines themselves, Google’s free Rich Results Test and a spreadsheet are enough.

1. Read the page the way a crawler does (View Source)

   Open the business’s key pages, right-click and choose “View Source,” and search the raw HTML (Ctrl/Cmd+F) for the business name, full address and phone number. If they aren’t there — because the page only fills in after JavaScript runs — a retrieval layer can’t read them either.

2. Run the prompts a customer would actually type

   Ask ChatGPT, Perplexity and Google's AI Overviews the real recommendation questions — "best [service] in [town]", "who does [service] near [neighborhood]" — in a fresh/logged-out session. Record whether the business is named, named correctly, and with the right details.

3. Check the structured data against what's on the page

   Paste each page into Google’s Rich Results Test (or read the application/ld+json block in View Source) and confirm there is LocalBusinessschema whose name, address and phone match the visible text exactly — and that any FAQ schema mirrors questions the page genuinely answers.

4. Compare the listings and citations side by side

   Put the website, Google Business Profile, Foursquare, Apple Maps and the top directory listings in one table. Note every disagreement in name, address, phone, hours and category — those conflicts are what make an engine treat the business as ambiguous.

5. Write down findings and a fix list, in priority order

   For each gap, record what an engine sees today and the single concrete change that closes it. Order the list the way an engine works — crawlable and readable first, consistent details next, corroboration last.

The audit, step by step

1. View Source — what a crawler actually sees

We opened the River City Drains homepage, right-clicked, chose “View Source,” and searched the raw HTML for the phone number. It wasn’t there. The source was essentially an empty <div id="root">— the name, address and phone only appeared once the React app hydrated in the browser. To a retrieval layer that reads the initial HTML, this business had no contact details at all. That single finding explains a lot: if an engine can’t read the NAP, it can’t confidently name the business.

If your name, address and phone aren’t in View Source, the AI layer reading the page can’t see them either — no amount of “AI optimization” fixes that.

2. Prompt tests — ask what a customer asks

In fresh, logged-out sessions we ran the same three prompts a real customer might type, in each engine:

• “Best drain cleaning company in [town]?” — ChatGPT named three competitors and not River City Drains; Perplexitynamed two, both with citations to directory pages; Google’s AI Overviewshowed a map pack the business wasn’t in.
• “Who fixes a blocked drain near [neighborhood]?” — no engine surfaced the business, even though its real service area covers that neighborhood.
• “Is River City Drains any good?” (a direct, name-it-yourself prompt) — ChatGPT hedged that it couldn’t find much information and offered the old phone number from a stale listing. That last detail is the tell: the engine had a record, just an out-of-date one.

Recording the exact wording matters. “Not recommended” and “recommended with the wrong phone number” are different problems with different fixes — and only the name-it-yourself prompt revealed that an old listing was leaking bad data.

3. Schema check — are the facts machine-readable?

We pasted each page into Google’s Rich Results Test and also scanned View Source for an application/ld+jsonblock. There wasn’t one anywhere on the site — “no items detected.” Schema doesn’t manufacture authority, but its absence means the few facts the page did state were left for the engine to infer rather than read directly. Adding LocalBusiness schema whose name, address and phone match the visible text exactly is one of the cheapest wins on the list.

4. Listing & citation consistency

Finally we put the website beside the Google Business Profile, Foursquare, Apple Maps and the top two directory listings in a single table. The conflicts jumped out: two different business names (“River City Drains” vs. “River City Drain Co.”), an old suite number on the profile, and a previous phone number still live on Foursquare — the very source ChatGPT leans on for local answers. To an engine cross-checking sources, these disagreements read as risk, and risky candidates get left out.

Findings & the fix list

Here’s the full audit summary — what an engine sees today, and the single concrete change that closes each gap. We work the list top-down, because a page has to be readable before consistent details or corroboration can help it.

Notice the pattern: none of these are exotic, and the most damaging one — NAP missing from View Source — is also the most fixable. Run the same five checks on your own business and you’ll get a list that looks a lot like this. The walkthrough for testing your AI-search visibility covers the prompt-test step in more detail if you want to start there.

Where Google Business Profile and Foursquare fit

Two of the six findings above were listing problems, which is typical. Here’s the honest, non-salesy version of why those listings carry so much weight in a local audit.

Frequently asked questions

Keep going