See what AI reads from your site, and what it misses

Torumata crawls your site the way a language model reads it: headings, descriptions, structured data (schema.org), and files like llms.txt. From that it calculates a score and builds a list of concrete things to fix.

We crawl most sites within 30 minutes and email you the result. Larger sites with dozens of pages can take longer, because the crawler reads each page individually.

Not always. Smaller findings, like missing descriptions or alt text, you can handle yourself using the guide attached to each finding. For more complex fixes you get ready-made files to upload to your server or hand to a developer.

Yes, no credit card and no commitment. We crawl twelve pages once a month, and every finding we hit there you see in full, with fix instructions. We hide nothing. From that same sample we also estimate what an audit of your whole site would cost in credits, so you know the price before you pay for anything.

Where the content comes from. A free generator asks you what is on your site and assembles a file from your answers, so the result is only as good as your description of it. Torumata does not ask: it crawls as many pages of your site as you choose, reads the headings, the structured data, and the real structure of the site, and builds the file from that. You will notice the difference most on a site you have not touched in years: we find things you would never have thought to mention.

Technical checks tell you whether AI can reach your site and read it. That's only half the job. You still don't know whether it can answer your customer's question from what it read. That's what the language model determines: it reads your pages the way something looking for an answer would, then lists which questions they answer, what context they lack to make sense away from your site, and which technical terms you use without ever explaining. That is what decides whether a chatbot cites you or reaches for a competitor instead. It doesn't run on the free tier; it's the most expensive part of the audit.

We sort claims by strength of evidence and say for every finding in the report which class it falls into. Necessary condition: Google's own documentation says a page cannot appear as a supporting link in AI Overviews or AI Mode unless it is indexed and eligible to be shown in ordinary Search with a snippet; it lists no further requirements (Google Search Central). Documented effect: the only peer-reviewed experiment in the field (Princeton, KDD 2024, 10,000 queries, arXiv 2311.09735) measured that quotations in the text raised visibility in generated answers by 41 %, concrete statistics by 33 % and source links by 28 %. No support: domain authority, backlinks and domain age; we have no measurement for them, so we keep them out of the score. The full ranking with sources is in the article What documentably increases your chance of being cited by AI, and terms like GEO or AEO are explained in the glossary.

It's a signpost for language models, the equivalent of sitemap.xml for search engines: one file telling them what your site covers and where to find it. We'll say plainly where this stands today, because elsewhere it gets glossed over. In June 2026 Google added to its documentation that Google Search itself doesn't use the file (AI optimization guide), and neither OpenAI nor Anthropic states in their crawler documentation that they read it. How much of that is left in practice was measured by Ahrefs in June 2026 across 137,210 domains: 28 % of sites publish the file and 97 % of published files never get a single request (Ahrefs study). What demonstrably fetches it are coding tools, which is why Stripe, Vercel, and Cloudflare publish one. If you run documentation or an API, it pays off immediately. For an ordinary site it's a bet on the future: adoption is growing, the field is moving fast, and having the file does no harm, just don't expect more citations in AI answers from it yet. That's why we generate it, but what we actually sell you is what demonstrably works: the robots.txt check, whether AI crawlers are allowed in at all, and if they are not, a ready-made change that fixes it. We generate structured data too, but the same applies to it as to llms.txt: no documented effect on AI.

Neither, in the usual sense; you buy credits. After the free audit you see the exact price of the next step in credits; you pay for exactly what the audit actually processes, and unused credits stay on your account for next time. You choose and pay for a credit pack directly in the app.

We only store what we need for the report: page text and technical metadata, not your visitors' personal data. You can delete your project's data anytime from the dashboard.

Because 100 is not a target a site should be aiming for. The score is a weighted average of five categories and a quarter of it is a language model's assessment of your content: only a page that answers with concrete figures, prices or a procedure gets the top mark. A contact page or an about page cannot do that by their nature, so 97 to 98 points is the practical ceiling. Part of the assessment is done by a language model, which does not score identically every time, so the result can move by a few points between two audits. Watch high severity findings and the trend between audits rather than a single number. How the score is calculated