The best in the sector, according to AI
A prospective client no longer types three words into Google and scrolls ten results. They open a conversational assistant and ask a whole question: which are the best companies in this sector for doing this, in this area, with this feature. And they get back an answer with names in it.
Some companies appear in that answer. Others do not. For anyone who sells, the difference is enormous — and it slips by, because that conversation happens elsewhere: you do not see the client who asked and heard other names in reply. The single lost lead stays invisible; but whether or not you appear is something anyone can check, with the same question and a few minutes. It is understandable that this is becoming a widespread anxiety among people running a business. Less understandable is the answer starting to circulate: that adding the right file to your site is enough to get you into those answers.
It is not. And the explanation of why is more useful than the shortcut.
Appearing is not the same as Being good
There is a claim going round, elegant and misleading. Take two sites in the same sector, ask four different assistants who the best are, and observe that one appears four times out of four and the other never. Conclusion: the second is badly built, the first well. Fix the structure of your data and you will appear too.
The conclusion skips a step. Whether a company appears when someone asks who is the best in the sector does not measure how readable its data is: it measures whether those systems consider it among the best. These are two different things. A company may fail to appear for the simplest reason of all: it is not held to be among the best by the sources those systems read. Its site can be flawless, first on Google, technically in order — and still not appear, because the question was not about its site.
Turn the claim the other way and it collapses on its own. If building your data well were genuinely enough to be cited among the best, then an assistant would have to name as excellent a mediocre company whose information is tidy and up to date, in place of an excellent one whose information is old or missing — on its site and elsewhere. No one would accept an answer like that: serious systems do not work that way.
Readability puts existing reputation in order; it does not invent it.
What actually decides who Appears
An assistant asked who is the best in a sector does not read a file and recite what it finds written there. It weighs what the independent sources say: reviews, articles, mentions, category listings, recognitions, the consistency with which a name recurs attached to a particular competence. It is authority gathered, not declared.
Hence the distinction that changes everything. Building a company’s information well helps it be understood correctly when reasons to cite it already exist — to avoid being filed under the wrong category, to avoid being credited with old data, to be recognised for what it actually does. It does not manufacture an authority that is not there. Readability puts existing reputation in order; it does not invent it.
Anyone selling the file as a visibility switch is promising the second thing. It is the same promise the keywords meta tag made twenty years ago: a field sites filled in diligently, convinced it made them rank, and that no engine ever used for that.
Why the question Weighs more than before
There is a reason all this matters more today than five years ago, and it is a fact about behaviour, not about technology.
In the first four months of 2026, 68% of Google searches ended without a single click to any website. It is a share that has been climbing for a decade: around 34% in 2016, 50% in 2019, 60% in 2024. The answer increasingly arrives inside the results page — a summary, a box, a card — and the user does not leave.
The consequence is that a company’s content increasingly works away from its own site: inside a synthesised answer, inside a card compiled by third parties, inside a list generated on the fly. Value shifts from being clicked to being cited correctly. And the few who still click are, paradoxically, the best visitors — they arrive having already read a summary, with an intention already formed, not out of curiosity. Traffic volume falls; the quality of those who remain rises.
The difference is not between little traffic and a lot. It is between the traffic that browses and the traffic that buys.
Not who searches, but what they ask
A distinction most analyses overlook is needed here, and it is the one that makes the whole question make sense. It is not about the kind of person searching, but the kind of question they ask — and the same person, on the same day, asks both kinds.
There is the generic question: “best men’s fragrances”, “communication agency in Milan”. High volume, broad intent. The answer is wide, the engine or the assistant returns it directly on the page, and the searcher settles for it: they stay at zero clicks whatever a company does with its site. On that kind of question there is no lever, and promising one would be dishonest.
Then there is the precise question: “alternative to Tom Ford Oud Wood”, “supplier with certification X in Emilia-Romagna for this process”. Here the intent is narrow — the searcher already almost knows what they want — and to answer, the system has to cross structured sources rather than stop at the first generic results. That is where being described well, in the right category, with the exact data, in the places those systems read, makes one name appear instead of another.
The difference is not between little traffic and a lot. It is between the traffic that browses and the traffic that buys. The precise question is almost always the one closest to purchase, and it is the same one whose outcome depends on how orderly a company’s information is out there. It holds in B2B as in B2C: the product changes, not the mechanism.
And the weight of this second question is set to grow. Today’s generic searching reflects in good part how people who did not grow up with these tools search. Someone who is twenty now and already frames articulate, conversational questions will be the average client in ten years. The tools that read structured data and the people who ask the right questions to trigger them move in the same direction.
What Google says, and what it does not
At this point it is only fair to report the most-cited objection. There is a file, called llms.txt, proposed in 2024 to give automated systems a concise, orderly version of what matters on a site. Google has stated publicly, and repeatedly, that its engine does not use it and has no plans to; one of its engineers compared it precisely to the old keywords meta tag. The data prove it right about the present: an analysis of more than a hundred thousand domains found that, in spring 2026, almost none of the published files had received even a single request.
It should be read for what it is. Google speaks about its own engine, and about its own engine it is right. But its own engine is no longer the only place companies are searched for, and it has an obvious interest in keeping people inside its own perimeter rather than sending them towards systems that read content elsewhere. When the source of a judgement has an interest in the judgement itself, that judgement should be weighed, not taken as a verdict.
The point, in any case, was never the file. A file produces no authority. At most it makes readable the authority that exists, on the channel that will matter most. It is a piece downstream of a reputation, not the lever that creates it.
A company’s profile may exist—and be incorrect—without the company even knowing it.
The blind spot: how you are described Elsewhere
There is an experience many business owners have already had, almost by chance: searching for their own company and finding wrong information. A mistaken category, an old figure, a description that no longer matches what they do. It is not an isolated case, and it is set to weigh more and more — because, as we have seen, that wrong information is increasingly the only thing the user sees, without ever reaching the site where the data is correct.
The phenomenon has taken a new and little-known form: platforms have emerged that catalogue tens of thousands of companies — by category, country, product, certification — and publish their profiles in a format designed precisely for automated systems to read. A company’s profile can exist, and be wrong, without the company knowing anything about it. Correction, where it is provided for, is manual and left to the party concerned; the spread of the wrong data, by contrast, is immediate and frictionless. The two speeds are not comparable.
Checking how you are described in these listings — and claiming your own profile to correct it — is the work almost no one does today, and it is more concrete than any file added to the root of a site. It is the equivalent, for the age of assistants, of the care once taken to make sure you appeared with the correct name and address in the category listings.
What gets Published and what gets Kept
One decision precedes every technical choice, and it belongs to communication, not to configuration. Making your data readable to one system makes it readable to all of them, competitors on the same market included. The shrewder platforms know this: they expose a generous, easy-to-consume public layer — the part needed to be catalogued, cited, recognised as a source — and keep behind the login the value they actually sell.
It is the distinction between the shop window and the warehouse. It holds for a platform with tens of thousands of profiles as much as for a company with a product catalogue: which part of your information holdings is worth more distributed than protected is a judgement to be made before publishing anything, not after.
The Right question
For a company that wants to appear when a client asks an assistant who is the best in the sector, the issue is not which file to add to the site. It is whether a real reputation exists — made of work, recognitions, mentions, consistency — and whether that reputation is told clearly and uniformly wherever a system might meet it.
Building information in an orderly way serves that reputation: it makes it readable, protects it from errors, puts it in a position to be recognised. It does not replace it. Anyone promising visibility as the effect of a file is selling the shortcut; anyone with something worth reading is doing a different, and longer, job.
Before asking how to be read by machines, it is worth asking whether you have a reputation worth reading — and whether the market tells it the way you would tell it yourself.
