What a buyer's AI sees when it reads your homepage.

The cold read

Take the visible text of a homepage. Hand it to a model with one instruction: based only on this, say what the company does and who it is for, and say plainly if something important is missing. No brand context, no prior knowledge, no charity. We call the result the cold read, and we run it as the final step of every AEO audit because it answers the question every other check only circles: can a machine describe this company the way the company would describe itself?

The answers fall into three groups, and the group tells you what to fix.

Three kinds of answer

The clean read. "Provides revenue intelligence software for sales teams; serves B2B companies from mid-market to enterprise; used by over five thousand customers." Specific, bounded, quotable. When a buyer asks an assistant about this company, this is roughly what they hear. The homepage did its job.

The hedge. "Appears to offer a platform that helps organizations with their data." The model is being polite about the fact that it could not find a noun. Homepages that lead with a manifesto, a category the company invented, or three rotating taglines produce this. A person can squint past it. A model reports it faithfully, and the buyer reads "helps organizations with their data" next to a competitor's clean read.

The gap. "The text does not provide clear information on pricing" or "it is unclear which industries the company focuses on." This is the useful one. The model is naming the sentence a buyer would want and the homepage does not contain. Every one of these is a line of copy you can write this afternoon.

Running it in one click

We put the cold read, and the three checks that sit in front of it, into a Chrome extension: zRev AEO Lens. Click the icon on any site and you get an AEO readiness score out of 100, the llms.txt grade, the schema types the homepage declares, whether robots.txt lets the AI crawlers in, and the cold read itself. No account, no sign-up, nothing tracked. Twelve reads an hour, cached for ten minutes so re-opening the popup is instant.

The score is deliberately simple. Half of it is the llms.txt grade, a quarter is structured data, a quarter is open crawler access. Three things you control, weighted by how often they are the reason a company is illegible. It is not a ranking of anything. It is a checklist with a number on it, so you can screenshot it and send it to whoever owns the website.

What to run it on

• Your own homepage, first. Read the cold read as the buyer. If you wince, that is the finding.
• Three competitors. The one with the clean read is the one the assistants will name. Notice what their homepage states that yours implies.
• A prospect's site, before the call. If the read says "unclear who this is for", you have your opening question, and it is a question they will not have heard from anyone else.

Closing the gap

When the llms.txt is missing or weak, the extension's Improve button opens our free generator with the domain filled in. It grades the file on seven checks, names the specific fix for each miss, including the pages the file forgot and the links that leave the domain, and hands back an improved draft that keeps everything already written. When the gap is in the homepage copy itself, the fix is the sentence the model asked for. Write it, in plain words, near the top.

Then run the read again. That loop, read and rewrite until a model with no context describes you the way you would, is the whole of AEO at the page level. Everything above it, the profiles and the corroboration and the content, only works once this does.

One honest caveat

The cold read uses one model on one page of text. Assistants in the wild read more than your homepage and weigh what others say about you. A clean read does not guarantee a mention. A hedge or a gap, though, is close to a guarantee that when you are mentioned, it will be vaguely, and vaguely is how you lose to the competitor with the clean read.