Test whether a public page can be retrieved, extracted and clearly understood by systems such as ChatGPT, Claude and Perplexity. You will receive an evidence-backed report with practical improvements.
No email address required. The scanner assesses one public page and its supporting files; it does not crawl your whole website.
Whether AI retrieval agents are permitted to reach the page, and whether they actually can. Permission in robots.txt counts for nothing if bot protection refuses the request, and that refusal is usually nobody’s decision. We found this pattern in roughly one in five of the UK’s largest listed companies when we tested the scanner: bot protection refusing agents that the site’s own robots.txt explicitly permits.
Whether the content is in the HTML an agent receives. Many agents do not run JavaScript, so a page assembled in the browser can be close to empty from where they stand. This is the most common serious failure we see, and the most expensive to fix late. It is worth fixing before a website or platform build rather than after.
Whether the page’s shape says what is a heading, what is the main content and what the page is called.
Whether the page states what it is in structured data, rather than leaving an agent to infer it from the layout.
Whether the page carries the provenance an agent looks for before repeating what it says: dates, attribution, a reachable about page.
Every completed scan contains an overall band, a five-pillar breakdown and the evidence behind each applicable check. Failures and partial results include a suggested fix; anything the scanner could not verify is labelled rather than guessed.
Scan a homepage, service, product, article or campaign page. Readiness can vary between pages on the same website.
The scanner tests retrieval, server-delivered content, document structure, structured data and trust signals.
Your report shows what was observed, how each check was scored and what to improve next.
A webpage can be clear and convincing to a person while being difficult for an automated client to process. Important copy may only appear once JavaScript has run. A firewall may treat non-browser requests as suspicious. Headings, canonical addresses and structured data may be absent. Ownership, authorship and dates may be obvious to a reader and invisible in the markup.
None of this makes a website invisible to AI on its own, and fixing it guarantees nothing. But one thing is certain: if a retrieval agent cannot fetch your page, it cannot cite it. Whether it would have cited you is the uncertain part. The retrieval is not.
It helps to think of it in three stages.
Retrieval – Can the agent reach the page and its supporting discovery files at all? This is where stated policy and actual behaviour most often diverge. A robots.txt that welcomes every crawler counts for nothing if a CDN rule written years ago refuses anything that isn’t a browser.
Extraction – Does the response contain enough useful content to work with? Not every retrieval client executes page JavaScript fully, so important content that only appears in the browser may not be recovered. A page assembled in the browser can look immaculate to you and arrive as an empty shell.
Interpretation – Does the page identify its subject, its structure, the organisation behind it and the signals that let a reader judge the source?
Each stage depends on the one before it. A beautifully structured page nobody can fetch scores nothing that matters, which is why our scoring rubric weights access most heavily.
The scan tests whether a page exposes signals that can help automated systems retrieve and interpret it. It does not measure whether an AI assistant currently cites, recommends or ranks the page.
The result applies to the submitted URL when it was tested. Other pages, temporary bot protection and later website changes may produce different results.
Requests present published crawler identities but do not originate from those services’ verified networks. They are practical access tests, not proof of how every genuine AI service is treated.
If enough evidence cannot be recovered, the report shows what was and was not verified. It does not turn unavailable checks into failures or invent an overall score.
It is the extent to which a website makes its public content retrievable, extractable, structurally clear, machine-readable and attributable. Those conditions can help AI systems use a page, but they do not guarantee inclusion in an answer.
No. Some good practices overlap, but this scan does not assess rankings, keywords, backlinks, Core Web Vitals or the quality of an entire search strategy.
No. Citation and recommendation depend on many factors outside this scan, including relevance, authority, the assistant’s sources and the question being asked.
Bot management, network policy, consent walls, redirects or transport failures can prevent individual checks. The report separates verified evidence from anything the scanner could not establish.
It requests the public URL you submit and supporting public files such as robots.txt, sitemaps and llms.txt. No account or email address is required. The submitted URL, report and supporting evidence are retained for 90 days; anonymous aggregate statistics may be retained for longer-term analysis.
Yes, within the tool’s fair-use rate limit. Scan representative page types separately because a homepage and a product or article page may expose very different signals.
Run a free scan, no account or email address required.
