A GEO checker that scores any page for AI search

GEO stands for generative engine optimization. It asks whether ChatGPT, Perplexity or an AI Overview can read your page, trust it and quote it, where SEO asks whether Google will rank it. The two overlap, which is why both tools ship in Hoverify and sit next to each other: the SEO panel covers Core Web Vitals, meta tags, headings and links, and GEO covers structured data, machine readability, citability, entity signals and the crawler files AI systems look for.

Each check lands as a pass, a warning or a fail, worth full credit, half or none. Each also carries a weight of its own, from 3 for a check whose failure hides the page from AI systems down to 0.5 for a nicety. A category's score is the weighted share of credit it earned. Tips that should not count either way stay out of the arithmetic. The 16 categories carry weights too: machine readability 1.5x, citability 1.3x, semantic HTML 1.2x, structured data 1.0x, agent interactivity 0.2x. The 0 to 100 score is the weighted average of the ten technical categories. A category with nothing to score drops out of it, and a page is never marked down for a standard it doesn't use. A few findings act as gates on top of that. A page that isn't server-rendered, carries noindex or noai, or blocks every core AI retrieval crawler is capped at 49 whatever else it gets right. Blocking some of those crawlers caps it at 69. The grade follows from the score: A at 90, B at 80, C at 70, D at 60 and F below that.

llms.txt is a plain text file at the root of a site that points LLM crawlers at the pages worth reading, in the way robots.txt gives search crawlers their rules. It is a proposed convention rather than a settled standard, and no search engine reads it, so nothing breaks without it. Hoverify checks whether you have one and whether anything on the page links to it, along with llms-full.txt, your sitemap, an MCP server card, an A2A agent card and a handful of other emerging files. On most pages a missing llms.txt is an unscored tip. On a documentation page, where coding agents such as Claude Code do use the file, it is a warning, and the row drafts an llms.txt from your own sitemap that you can copy straight out of the panel.

Open the panel on a page and it audits that page there and then. You get the score, the grade, the page type it detected and a list of the top issues ranked by how many points fixing each one would add, with that estimate shown beside it, so the answer arrives as a number and a to-do list. Every flagged row also carries a tag for who usually fixes it: Developer, Content or Infrastructure. Scan again after a change and the History tab puts the new score next to the old one.

Yes, for everything read from the page: structured data, headings, links, images, text ratios and the rest run on whatever you have open, including a members area or a local build. The off-page probes are the exception. Checks for robots.txt, llms.txt, your sitemap and the agent card files need the origin to answer a request. Those requests are made from the page itself and carry the same cookies your browser already sent it, which is what lets the server-rendering check work behind a login. Where the origin refuses a request, with a 401, 403, 429 or 503 or a Cloudflare challenge, the row says it couldn't verify the file rather than counting against you. A plain 404 is a different matter: a sitemap that isn't there, and isn't named in robots.txt either, is a failure.

Nowhere until you switch it on. The feature stays off until you add your own OpenRouter key and choose a model in settings. After that, an audit sends the model you picked, by way of OpenRouter, the page's address with any query string and fragment removed, its title and description, and up to 40,000 characters of the main text. It also sends the page type it detected, today's date, the language attribute, a word count and up to 60 headings. The image data is a count of images, how many lack alt text, how many are marked decorative and up to 10 alt text samples. The Schema.org types in the markup go along with it. Compare and Batch send the same for every page they scan, which means other people's pages go to the model too. OpenRouter is told to route only to providers that don't retain prompts. Your key is stored on that device and never synced, and none of it reaches us. With the feature off, the audit runs in your browser apart from the requests to your own origin: robots.txt and the files beside it, plus the page itself fetched twice, once as a browser would and once asking for Markdown.

Yes. Every tool runs in the Firefox build, and the same license covers it. Chrome, Edge, Brave and other Chromium browsers are supported too.

It comes with Hoverify, a paid extension with a 14-day refund behind it. Audit a few of your own pages and you'll know whether it stays, and you can get your money back if it doesn't. That stands in for a trial. One license runs in 3 browsers at a time, and nothing auto-renews. See the pricing.