The twenty questions
we get asked.

We audit whether AI systems can find, read and act on your website, then generate the code and files that fix what is broken, then re-measure on the same rubric so the improvement is provable. Most tools in this space stop at the diagnosis.

AI assistants increasingly answer your customers directly, citing two or three sources. If your site cannot be read cleanly, the assistant uses a competitor's instead. You lose the consideration before anyone visits, and nothing in your analytics records it happening.

It sits on top of SEO rather than replacing it. Google states that optimising for generative AI search is still optimising for search, and AI engines draw on the same indexes. What is new is that agents want to use your page, not just list it, which is a different technical requirement.

Those measure demand: whether AI mentions your brand. We measure supply: whether an AI arriving at your site can use it. Both are useful and they answer different questions. We would rather you ran both than pretended one replaces the other.

The one-off AI Visibility Audit is $4,500 and takes five working days. Retainers are quoted per site because crawl volume, query set size and engine coverage vary enormously. You get a retainer figure on the first call.

Not yet. We launched in 2026 and would rather say so than publish invented numbers. Our own site is audited by our own engine and published in full, including the checks it fails. We also offer founding client terms in exchange for permission to publish real results.

Almost never. Most findings are configuration and structure: missing files, incorrect headers, markup that contradicts the visible page, forms without proper labels. These are targeted changes rather than redesigns.

Someone who can deploy changes. We produce the files and your developer or agency reviews and ships them. If you would rather we did that work directly, that is the remediation engagement.

The technical checks improve the day you deploy, and you can verify them yourself. How quickly that changes what AI engines say about you is slower and far less predictable, which is why we measure representation separately and never put a date on it.

No. We only request pages any AI crawler could already reach, we respect robots.txt and crawl-delay, our crawler identifies itself as Code4XBot and signs its requests. Live agent testing runs against staging by default.

We tell you that and you have a documented baseline. It has not happened yet on a site that had never been audited for this, but if it does we will not manufacture findings to justify the invoice.

Yes, and many will prefer to. The report is written to be handed over: every failure includes the URL, the test that failed, and deployable code for your framework.

Fifty-five dimensions across four layers. Forty contribute to your score deterministically. Fifteen are advisory judgement calls reported with evidence but no points, because a score an AI can move cannot prove a before and after.

It is genuinely contested. A Google Search Advocate compared llms.txt to the old keywords meta tag and said no major AI service currently requests it, while other sources report frequent crawler fetches. It costs almost nothing to add, so add it, but treat anyone selling it as the centre of an AI strategy with suspicion.

It means serving a clean text version of a page when a program asks for one, instead of a full page of markup. It can cut what an AI pays to read your page by ninety per cent or more, which makes your page cheaper to use than a competitor's. Around four per cent of sites support it.

Possibly. Roughly six per cent of sites block major AI crawlers and many never chose to, inheriting it from a template or a hosting default. It is one of the eight checks in the free scan.

It is an emerging standard letting a site describe its functions as tools an AI agent can call. Adoption is close to zero and it is still an origin trial. We check for it, but the honest advice for 2026 is to get semantic HTML forms right first and revisit WebMCP later.

Instead of a PDF, your audit is available through a server your team connects to Claude, ChatGPT or Cursor. Your CEO asks it questions in plain language; your developers point a coding agent at it and have the fixes applied. Same data, no dashboard to log into.

Yes, and we would rather you did. Every deterministic check maps to something you can reproduce with curl or your browser's developer tools, and the report tells you which command to run.

They will. We track every standard in a public index with dated change history and publish a weekly log of what moved. Your rubric version is pinned so re-audits stay comparable, and we tell you explicitly when we change it.