It is ready if a machine can read it. That is the whole condition, and it is more demanding than it sounds: in the most complete evaluation published to date, an AI assistant sent out to find a product finishes the task two times out of three. The rest of the time it finds nothing to grab onto in the pages it visits, and it walks away.
That last part is what matters and what almost nobody says: the assistant does not persist. It does not call you, it does not come back later, it does not tell you. It moves on to another page. Nobody finds out that you were about to appear and did not — not you, and not the person who asked.
So the question in this article is not how to optimize for one engine or another. It is an earlier one, and it holds equally for a store, an accounting firm, a clinic or a site that explains something: is what you published in a format a machine can understand?
There are two, and it is worth telling them apart because they do not work the same way.
The first is direct. A person asks an AI a question and expects an answer. "Which one suits this?", "who does this near me?", "how is this solved?". The AI answers and, along the way, names sources.
The second is delegated. The person does not ask: they assign. "Find me the three best options and compare them." Now the AI goes out to visit sites on its own, gathers, discards, and comes back with a short list.
In both cases what reaches your site is not a person: it is a program reading. The difference is what it does afterwards with what it read.
Being cited means appearing by name inside an answer. Someone asked something, the AI replied, and your site shows up as the source part of that reply came from.
Being recommended means appearing on a short list. Nobody asked about you: someone asked for options, and you survived the filter.
They are two different acts and they are worth different things. A citation gives you authority. A recommendation puts you on the decision table. But both depend on the same thing: before citing you or recommending you, something had to be able to read you. That is the single door, and it comes before everything else.
Nothing visible happens, and that is the problem.
The Mannheim researchers described the moment precisely: when the assistant can only go by what it sees on the screen, without the structure of the page underneath, it keeps attempting the same click over and over against something that is not where it appeared to be, until its allowance of attempts runs out and it leaves.
There was no error. There was no message. No record was left anywhere. That page simply stopped being a candidate, and whoever asked received an answer that no longer included it.
Compare that with what used to happen: if someone came to your site from a search engine and left, at least the visit was logged. Here not even that is left.
No, and this is the most expensive confusion.
A page can look impeccable and be illegible to a machine. It happens all the time: the important information tucked inside an image, the data that only appears after the visitor clicks something, the price or the opening hours loaded by an external tool that draws itself only once the page has finished opening.
A person takes all of that in effortlessly. A program that reads does not. To it, that part of the site is blank.
And there is an irony here: what tends to be hidden is exactly what you most want to show. The promotion lives in a banner. The specialty, in a graphic. The differentiator, in a video. The best thing you have, stored in the format the new reader cannot open.
No. It applies to anyone who wants to be found.
A store wants its product to enter the comparison. A professional firm wants to be named when someone asks who solves a given problem. A clinic wants to come up when people ask about a treatment. A site that explains something wants to be the source the AI cites.
The goal changes, the condition does not. In all four cases, first you have to be readable.
Badly, and that is the opportunity for whoever moves first.
23 out of every 100 sites analyzed have absolutely nothing a machine can read in an orderly way. Not little: nothing.
62% scored below 30 out of 100 on how they organize their content, and the overall average was 35 out of 100. The best of them reached 62.
These are not abandoned or improvised sites: they are active companies with money invested in their web presence. They are ready for the visitor who looks. Not for the one who reads.
By finding out where you stand. It sounds obvious, but almost nobody has done it: most decisions about a site are made by looking at how it appears, not at how it reads.
Three questions to ask yourself today, with nobody's help:
After that comes the technical part, and there you need to measure it. But the order matters: first find out whether you can be read, and only then discuss how to improve it.
Get your GEO score across eight academically-backed criteria and find out where your site stands with AI engines.
Run the analysis at nostosgeo.app