What do you want to know about AI search?

There is no form to fill in and no slot to buy: answering systems build their answer from sources they consider clear, consistent and verifiable. Appearing means becoming one of those sources — having pages that state explicitly what you do, for whom, where and with what results, in a form that still holds when lifted out of its context. It is not about keywords: it is about statements that can be quoted without risk of getting them wrong.

A search engine produces a list and leaves the judgement to whoever is searching: ten links, and the choice is yours. An answering system has already made that judgement and hands you the result: two or three names, with a reason. The change is not in the interface, it is in the role — whoever does not make those two or three names is not discarded after a comparison, they are never compared at all.

It depends on the system, and it changes often. Some lean on an existing search index, some have their own crawler, some combine both with what they learned in training. For a site owner the practical consequence is one: the technical foundations serve all of them — reachable pages, readable text, consistent information — while chasing the behaviour of a single system is a bet on something that changes by itself.

The three most common reasons have nothing to do with the quality of what you sell. First: the site never states explicitly things that are obvious to you — sector, cities served, type of client. Second: information contradicts itself across the site, profiles and reviews, and a system that has to choose would rather not take the risk. Third: the content is written to convince a human who is already reading, not to be understood by something that reads it once, in one go.

It is useful, but not for the reason usually repeated. It is not a magic ranking signal: it is a way to state without ambiguity something the prose leaves implicit — that this company supplies window frames, that it operates in Milan, that this price is in euros. A system that has to answer prefers sources it does not have to guess about. Structured data removes the occasions for guessing.

Almost never. In most cases the problem sits in a few places: the page that should say what you do and instead tells a story, the service pages with no price and no boundary, the contact details that differ in three places. That is where you act. Rewriting everything is the typical reaction of someone who did not measure first, and it usually moves less than it costs.

It is a text file at the root of the site summarising what it contains and where the pages that matter are, meant for whoever reads the site with a machine. It is nobody’s imposed standard and no system is obliged to read it. It costs ten minutes, does no harm, and in exchange it forces you to write on a single page what the company does: often that is the first time anyone asks the question explicitly.

Yes, and it will stay that way: a site Google cannot read is not read by an answering system either. But SEO optimises for a list of ten results the person chooses from, while an answering system makes the choice itself and shows two or three. They are two different jobs sharing the same foundations. Whoever did SEO well starts ahead; whoever stops there stays out of the part where the decision happens.

You start from the real questions of the sector — the ones where somebody asks who to choose — and look at what the system answers. Three numbers come out of that: whether the brand appears, in what order it appears against competitors, and in what words it is described. The third is the most overlooked and often the most useful: being cited badly is a different problem from not being cited, and it is solved differently.

Volumes are lower than those of a first position on Google, and that is normal: a generated answer intercepts a different moment. The person is not exploring, they are choosing. Counting those visits the way organic ones are counted leads to wrong conclusions — fewer visits, but they arrive with a recommendation already received, and you can tell from the questions they ask.

You trace it back to the source. A wrong description almost always comes from something that really exists: an old page never updated, a listing on a directory with stale details, a press release from three years ago. The system is not inventing, it is reporting the most consistent information it found. You fix the source and give the system a clearer and more recent version to prefer.

The first measurement arrives in about ten minutes: it is a snapshot of how the site is read today. Changes in the answers, though, follow the pace at which systems re-read their sources, and that pace is not the same for everyone — some update in days, some in weeks. Anyone promising a precise date is promising something they do not control. What can be said is that the comparison is continuous: measure, act, measure again.

Because the job consists of getting one name chosen among the two or three the system shows. Taking on two competitors in the same sector and the same area would mean working against ourselves, and promising it to both would be dishonest. It is a choice that reduces the number of possible clients: that is the price of being able to say we are really working for one.

A report that says three things: where you stand today in the answers of your sector, which points of the site are blocking you, and what to change first. Not a list of generic advice, but indications about your own content — this page does not say what you sell, this claim does not hold once extracted, these details contradict each other. Plus the comparison with competitors on the same questions.

There is no public price list, and the reason is not secrecy: we work with a single client per sector and geographic area, so the scope varies a lot from case to case — how many questions to track, how many competitors, how many areas. The figure is set after the first measurement, when it is clear what is actually needed. The initial measurement of the site is free and commits you to nothing: that is where it starts. For an answer about your case, write to [email protected] or fill in the access request.

No, and be wary of anyone who promises them. The answers of artificial intelligences are produced by third-party systems that change without notice: no vendor controls them, so nobody can guarantee a position or a date. What can be guaranteed is the work on the conditions that make a citation likely — clarity, consistency, quotable statements — and the honest measurement of what happens, before and after, on the same questions.

Yes, and it is the first step for everyone: the initial measurement of the site takes about ten minutes and costs nothing. It produces a report with where you stand today in the answers of your sector, what is blocking you and what to change first. It is useful material even for those who then decide not to continue. No access to the site is needed: the address is enough.

In writing at [email protected], or through the access request form, which is the quicker route because it already collects what is needed to assess the sector: company name, website, sector and a contact. We answer both. If your sector in your area is already taken we say so straight away, without wasting your time.

For anyone looking for an immediate, guaranteed result: that does not exist, and promising it would be dishonest. For anyone who cannot touch the content of their own site, directly or through whoever runs it: the measurement can still be done, but without being able to apply the fixes it stays a snapshot. And for anyone in a sector and area where we already work with somebody else: in that case the request is not accepted.