More buyers now start with a question typed into an assistant rather than a search box. The answer they get is built from what your site makes machine-readable: the structured data behind your pages, what your opening words say plainly, whether your contact details exist as text, and whether the crawlers that feed those systems are allowed in at all. When those are thin, the answer is thin. Sometimes it is wrong, and nobody at the company knows.
This check fetches your public pages once, scores twenty-eight signals, and returns the entity readback: your business as a machine can state it, with a blank wherever your site never says. Nothing is gated, and the tool collects no contact information. For the reasoning behind what it measures, see the reference on how AI describes your business.
How this check works
It requests your homepage the way an AI crawler would, then your robots.txt, sitemap.xml, and llms.txt, and it requests the homepage a second time declaring itself as an AI agent. That second request is the one most checks skip: a firewall can block AI crawlers your robots.txt welcomes, and the site owner never sees it happen. Eight requests, once, no crawl of the rest of the site.
Twenty-eight signals are scored across four sections, out of one hundred points. Every point is accounted for here, because a score whose math is hidden is a number you cannot argue with.
| Section | Points | The question |
|---|---|---|
| Access | 40 | Can AI systems reach the site at all? Fetchability, certificate, redirects, robots.txt health, and which classes of AI agent are permitted: search-index crawlers, live assistant fetchers, and Google’s ordinary index, which is what AI Overviews actually read. |
| Legibility | 20 | Is the content there without JavaScript? AI crawlers do not run it, so a page assembled in the browser is a blank page to them. Plus canonical, sitemap, and whether the site answers inside the few seconds a live assistant fetch will wait. |
| Identity | 32 | Can a machine state who the business is? Structured data completeness, how specific the business type is, whether the title and opening words say what you do and where, contact facts as text, and whether your structured data agrees with your visible page. |
| Hygiene | 8 | How a citation looks when it happens: the link preview, the favicon on the citation chip, the language declaration, and whether anything on the page suggests the business is still operating. |
The four bands: Citable at 80 and above, Legible from 60, Reachable from 35, Invisible below that or whenever the homepage cannot be fetched at all. Two deliberate weightings are worth naming. Blocking model-training crawlers costs nothing here: it is a legitimate policy choice, and a tool that scolds you for it is selling something. And llms.txt, which some checkers weight heavily, is worth a single point, because no major assistant vendor documents consuming it yet.
What this check refuses to pretend
It does not know what any assistant said about you. That requires querying them, which is a different instrument; several tools that only crawl a site nonetheless advertise “see how ChatGPT sees you,” and it is not true. This one reads the inputs those answers get built from, and says so.
It cannot see your Google Business Profile, your reviews, whether your details match across directories, or how often the web mentions you by name. All of those weigh heavily in what an assistant says about a local business. The result page lists those limits beside your score rather than below the fold. It reads one page of your site, not all of them. And it produces a starting point for thinking, not a diagnosis. The diagnosis is the audit.
Binary Glyph is a brand and marketing practice in Toledo, Ohio. The gaps this check surfaces on your own pages are one part of what a fixed-fee marketing audit documents in full, across the surfaces a crawler cannot reach. The audit credits toward a six-month Foundation Engagement. If the readback was missing something it should not have been, begin a conversation →
This check fetches your public pages once, at your request, using an identified agent (BinaryGlyphAICheck/1.0). It keeps none of them: the pages are read, scored, and discarded. It stores no contact information, and the address you enter is never recorded in analytics. Results for a given site are reused for a day so repeat checks cost nobody anything.