We ran Citrion on ourselves. Here is the whole report.
On September 3, 2026 we audited ezweb.dev with the same instrument, the same ten question protocol, and the same report every client gets. Nothing was edited afterward. The score is 50 out of 100, and the report says why.
Half the answers named us. The other half named someone with our name.
Our AI visibility score: the share of question and assistant pairs in which EZ Web was named or linked. Computed from 107 recorded answers, never estimated.
ChatGPT, Perplexity, Claude and Google AI, on our own accounts with memory off. Grok hit a capacity wall that day and is recorded as paused, not as a miss.
Both technical, both fixable in under a day: our phone number lived only in page markup, and nothing on the homepage quoted a named source.
The headline problem is an entity collision. Eleven unrelated businesses trade as some form of "EZ Web", and the assistants blended their reviews, complaints and locations into ours. Perplexity answered "is EZ Web a legit SEO agency" with a warning about churn and burn operators that share the name. ChatGPT could not find ezweb.dev reliably and recommended EZ Rankings instead. Google AI and Perplexity did cite us on the branded questions, and no assistant named us on a single category question such as "generative engine optimization agency".
That is what a real audit looks like when the client has a naming problem rather than a content problem. The roadmap in the report reflects it: disambiguation pages first, directory listings second, category pages third. You can see the pages it produced at Which EZ Web is this, About and the comparison hub.
Every page a client receives, unedited.
- The score and what it means. One number, the answers behind it, and the split between cited, named only, and absent.
- Who owns the answers. The domains the assistants sent buyers to instead, ranked by share. For us that list is directories and look-alike sites, not rival agencies.
- What the assistants say about you. Every answer that named us, read for favorability, with the sentence quoted. Thirteen raised a concern, almost all of them about businesses that are not us.
- Technical findings with evidence. Crawler access for every AI fetcher, Lighthouse scores, and the two problems above with the fix for each.
- Google, labeled. Where Google AI was measured it says measured. Where it was modeled it says modeled, and modeled numbers stay out of the score.
- A six month roadmap. What to fix, publish and earn, in order, with the first month's work stated plainly enough that our own team could do it.
Recorded September 3, 2026. Scores move as answers change, so a report is dated and a client on a plan is re-measured monthly against the same questions.