By the iSuggest.ai Team · Updated for 2026
Every audit on iSuggest.ai ends with a number: a page health score, and eventually a domain-wide GEO score on /scoring. Numbers like that are easy to distrust if you can't see what's inside them. So here is exactly what we look at on a single page before that score exists, with nothing left out and nothing dressed up.
Page identity: the tags that tell a machine what a page is
Before anything else, we read the basics a browser and a crawler both rely on: the title tag, the meta description, the canonical URL, the declared language, the viewport tag, and any robots directives (noindex, nofollow, noarchive) that might quietly keep a page out of an index or an AI crawler's answer set. A page can look perfect to a visitor and still be invisible to a bot because one of these is missing or wrong — that gap is one of the more common findings in how an audit actually works.
Headings, links, and images: how the page is structured
We count h1 through h6 and read the actual text of the h1, because heading structure is how both search engines and language models figure out what a page is actually arguing. We also count internal and external links, flag nofollow links, and sample a handful of each so you can see real examples rather than just a total. Images get checked for missing alt text, since an unlabelled image is a small, cumulative signal that a page wasn't built with machine readers in mind.
Structured data and content signals
We count JSON-LD blocks and sample the schema.org types present on the page — to be precise, we detect which types are declared (Article, Product, FAQPage, and so on), not whether every property inside them is filled in correctly. That's a real limit worth stating plainly rather than glossing over, and it's the same reason our domain-level GEO criteria never scored a property like sameAs directly: we only measure what we can actually observe. Alongside structured data we measure word count, the text-to-HTML ratio, and estimated reading time, because a page that's mostly markup and thin on real content gives an answer engine very little to quote. For the deeper mechanics of why any of this matters to a model rather than a human reader, see structured data for AI search engines.
Crawl basics: can a bot even get the page reliably
Every audited page records its HTTP status code, response time, and content length, and we flag pages that answer slowly or that aren't HTML at all. This is a check on baseline reachability, not a full performance profile — a standard audit does not run the heavier page-speed analysis that a dedicated performance tool would. If a page is slow or a status code is wrong, an AI crawler may simply give up before it ever sees your carefully written content, which is the whole argument behind why page speed still matters for AI search.
From one page's score to a domain grade
A single page's score becomes an overall_score with severity-tagged issues (critical, high, medium, low), each with the specific evidence we found and a concrete fix, plus a list of what already passed — because a report that only lists faults reads as if nothing works. When you've audited enough pages on a verified domain, those per-page scores roll up into the domain's page-health pillar as 70% the mean score and 30% the tenth-percentile score, with the home page counted twice. That weighting is deliberate: it means nine strong pages can't hide one badly broken one, and it's fully documented, not a black box, at /scoring. If you want a walk-through of reading an actual report end to end, we covered that in reading your first page health report.
What this is, and what it isn't
We want to be precise about what these signals add up to. This is a fast, specific, GEO-first read of the technical and structural things that decide whether a page is easy for a bot to parse and easy for a model to quote — not a replacement for a dedicated enterprise SEO suite. We don't track keyword rankings over time, we don't crawl a backlink index, and we don't estimate search volume; tools like that exist and do that job well. What we do is check the specific, observable signals above, on your own pages, for free, every time, and let you publish a frozen version of a page straight to an AI directory once it's genuinely ready. You can see the account side of that credit-based publishing step at /account.
None of this is exotic. It's the same handful of technical facts a careful human reviewer would check by hand — we just check all of them, on every page, every time, and show you exactly which ones failed and why.