Can AI systems understand your business website? I audited mine.

I ran an AI visibility audit on Tidegate because I wanted a harder answer than “the site looks good.” The useful questions were simpler: can retrieval systems reach it, is the real copy present in raw HTML, and can a machine tell who runs the business and what it does?

What the audit can actually prove

An automated audit cannot promise that ChatGPT, Claude, Perplexity, or Google will recommend a business. No vendor publishes a complete ranking formula. It can verify the technical conditions that make retrieval and citation possible.

• Whether important crawlers are allowed by robots.txt.
• Whether the server gives those crawlers a normal public response.
• Whether commercial copy is present in the raw HTML.
• Whether structured data identifies the business, founder, services, and service area.
• Whether names, addresses, phone numbers, and profile links agree across the site.

What Tidegate’s first pass found

Seven of seven tested retrieval and search crawlers were allowed. The audit collected raw HTML from all four discovered public pages. The homepage alone contained more than 1,500 readable words, so the main sales message was not hidden behind a browser-only application.

The weakness was entity clarity. The site identified an Organization named Tidegate, but it did not connect that organization to its founder, public identity profile, service area, or individual services. The insights index was also thin, which gave search and answer engines very few durable pages to cite for specific buyer questions.

Crawler access and recommendation authority are different problems

A site can be completely crawlable and still be absent from a buying answer. That was the important distinction here. Branded searches found Tidegate. A sample generic search for AI automation help in League City surfaced established competitors instead.

The missing ingredient was not another bot directive. It was corroboration: specific service pages, current operator proof, consistent founder identity, and third-party sources that say the same thing the website says.

The first fixes worth making

1. Connect the entity. Give the organization and founder stable identifiers, link the founder to a verified public profile, and describe the markets and services the company actually serves.
2. Create citable pages. Put each important problem, workflow, and proof story on its own permanent URL instead of asking one homepage to answer every question.
3. Publish current receipts. Show a workflow that can run today, with private information removed, rather than leaning on a result that is difficult to reconstruct.
4. Measure the market separately. Repeat the same buying and entity questions monthly across search and answer engines. Record the date, prompt, brands named, and cited sources. One answer is a snapshot, not a score.

What I would not spend time on first

I would not lead with llms.txt, hundreds of thin city pages, or a made-up visibility score. None of those substitutes for a clear business entity, useful pages, current proof, and credible outside mentions.

The operator lesson

Visibility work has the same failure mode as most automation work: owners get handed another report and still have to decide what matters. A useful audit should end with an ordered fix list, literal evidence, and a small number of actions tied to demand or trust.

The first Tidegate audit did exactly what I want these systems to do for an owner. It separated what was working from what was assumed, named what could not be assessed, and reduced the next move to a short list.