SEO and AI search checker

A full site audit, free: what stops a page from ranking in Google or being quoted by ChatGPT, Claude and Perplexity, with a fix for each problem. One page in depth, or a whole site.

A fix for every issue Over 25 AI crawlers Up to 500 pages No sign-up, nothing stored

Checked in your browser; pages come through webdave’s relayAsked: nothing yet

What does the SEO and AI search checker test?

Short answer

It reads a page the way Google and AI crawlers do and checks everything that decides whether it can be found, shown and quoted: indexing, titles and descriptions, content, links, images, structured data, social cards, speed, security headers, and which AI bots are let in. Each problem comes with why it matters and how to fix it.

  • Indexing: status and redirects, robots.txt, noindex, canonicals, hreflang, sitemaps.
  • On page and content: titles and descriptions measured in pixels as Google cuts them, headings, word count, readability, duplicates.
  • Links and images: every link and picture on the page fetched for its status, alt text, sizes and formats.
  • Structured data: each JSON-LD and microdata item against Google's rich result rules.
  • Speed: Core Web Vitals from real Chrome visitors and a Lighthouse run, through Google's PageSpeed Insights.
  • Whole site: up to 500 pages crawled for broken links, duplicate titles, orphan pages, redirect chains, thin pages and click depth.

What is GEO?

Generative engine optimisation is making pages that AI answer engines, such as ChatGPT search, Perplexity, Claude and Google's AI Overviews, can reach, read and quote. Much of it is ordinary SEO. The differences: most AI crawlers do not run JavaScript, so the text must be in the HTML; they favour a plain answer near the top, question headings, lists and tables; and they weigh who wrote a page and when.

The AI search tab shows the page as an AI crawler gets it: the main text as Markdown, its length in tokens, and whether any of it needs JavaScript.

Why AI bots matter

Each AI company runs several bots. Some collect pages for training, some build the index their answers search, and some open a page because a person asked about it. Blocking a training bot keeps your pages out of future models but not out of answers; blocking a search bot hides you from that company's answers.

Firewalls often block AI bots without anyone deciding to: Cloudflare's "Block AI bots" switch, challenge pages and rate limits. With "Check AI bot access" on, the page is fetched with each bot's user agent to see what it gets. Only the user agent is the bot's: real bots also come from the company's own addresses, which some firewalls check.

How the score works

Each category scores 0 to 100 from its checks: an error costs 10, a warning 4, a notice 1, and on a whole site more for problems on more pages. The overall score is the average of the categories. It shows how much there is to fix, not where a page will rank or whether an AI will cite it.

What is sent where

Nothing leaves your browser until you press Check. Then:

  • webdave's relay on Cloudflare fetches the page, robots.txt, llms.txt, sitemaps and the links' status, because a browser may not read other sites. It keeps answers 10 minutes so a second check is quick, and keeps no record of who asked. Cloudflare's Turnstile checks that a person is asking.
  • Google's PageSpeed Insights gets the address, to measure speed. Switch PageSpeed off to skip it.

Every check runs in your browser. Paste HTML sends nothing at all. Pictures from the page are not loaded unless you ask.

Questions

For auditing a site, yes: it runs the same kind of technical checks as their site audits, up to 500 pages, with no account or limit on reports. It does not have their backlink or keyword databases.

That is your choice. GPTBot and ClaudeBot collect training data; blocking them does not hide you from ChatGPT or Claude answers, which use OAI-SearchBot and Claude-SearchBot. Block the search bots only if you do not want to appear in those answers at all.

A Markdown file at the root of a site, proposed at llmstxt.org, that gives language models a short summary and a list of the pages that matter. Few AI companies say they read it yet; it costs little to add. The checker validates it against the proposal.

Google cuts titles by width, not by characters: about 600 pixels of 20-pixel Arial on a computer. The preview measures each letter the same way, so a title of wide capitals is cut sooner than one of narrow letters.

Not by address: the relay sees what a logged-out visitor sees. Save the page from your browser or copy its source, and use Paste HTML. Every check that needs only the HTML runs, and nothing is sent.

No. The score counts technical problems. Rankings and AI citations depend mostly on whether the page answers the question better than others, and on who links to it.