Presence on search engines and AI assistants. They cite you when someone asks.
Someone types your question into Google, or asks an assistant. They get a written answer, with three or four names in it. This piece works so that one of those names is yours, and so that you know whether it is happening.
The goal is to appear in the answer. Not to rank first on a generic word, but to be the name that comes up when someone describes the problem you solve.
The two games are measured with two different numbers. On search engines position and clicks count. On assistants citation counts, because the click hardly comes: 8% against 15%, measured by the Pew Research Center.
No special files are needed. Google writes in its guide that no file, no markup and no schema “for AI” is needed. What matters is a real answer to a narrow question.
This is the page for a single piece of the system. The other pieces, and the criterion for choosing which to start from, are on the services page.
People searching today get an answer, not a list
The click is coming away from search, and it has been measured. The Pew Research Center followed the real browsing of 900 US adults across 68,879 searches in March 2025: where the generated summary appears, people click a result 8% of the time, against 15% where it doesn't.
The number that matters most is the third: only 1% click the sources cited inside the answer. The reader takes the information and moves on, and the name cited is the only thing they carry away. That is why the citation is the result, and the traffic that follows is a side effect.
The same thing shows from the publisher's side. Ahrefs compared 300,000 keywords, half with the generated summary and half without: the page in first position gets 34.5% fewer clicks when the summary appears above it.
Staying first and staying cited are two different jobs. The second doesn't produce visits; it produces the fact that your name is already in the head of whoever calls you.
What changes in practice
Who calls you changes. Requests that come from a page answering an operational question are from people who already have that problem in hand, and the conversation starts further along than with a generic request that came from an ad.
The second change is that you stop depending on a channel you pay for. An indexed page keeps answering when the campaign is switched off, and the cost it has already had doesn't repeat every month.
The third is the most uncomfortable to accept: you become measurable. You know how many questions bring you an answer, on which you are absent and for how long, instead of believing you are visible because the website exists.
You choose the questions, and they are narrow
The work is on the operational questions a buyer actually writes, never on the category. A page aimed at the name of a whole sector competes with anyone who has a budget and answers nobody, while a page that answers a precise question has far fewer people in front of it.
Good questions can be recognised by one thing only: someone has already asked you them on the phone. The starting point is the requests you already receive, written in the words of those who made them and not in those of your price list.
| Type of question | What happens | What you get |
|---|---|---|
| The categoryone or two words | Whoever searches like this is still working out what it is about, and in front of them are those who have been investing in advertising for years. |
No useful result, and months spent on a position that won't come. |
| The operational questiona whole sentence | Whoever searches like this has a concrete problem in hand and is looking for how to solve it, now. |
Fewer people in front, and whoever arrives is already considering buying something. |
| The company namewho you are | Whoever searches like this has already heard someone mention you and is checking whether you really exist. |
The check ends well, and the negotiation doesn't stall on doubt. |
What the system does, step by step
The work lies in five points, and each one leaves a trace that can be looked at afterwards. None of these steps promises a position: they promise that the question you answer is written somewhere in a form a machine can cite.
| Step | What happens | What you get |
|---|---|---|
| The questionswhat you are asked | The real requests you receive are gathered and turned into a list of questions written in the words of those who ask them. |
A list of topics someone actually searches for, instead of a list of words chosen at a desk. |
| The pagesone answer per question | Every question has its own page, with the complete answer in the first sentences and the rest below, in passages that stand on their own. |
A page that answers even those who read only three lines of it, which is how machines read it. |
| Structured datawho you are, for machines | The name of the business, what it does, where it works and the linked public pages are declared in a form a program can read. |
Your name stops being ambiguous, and whoever looks for you finds you and not a business with the same name. |
| Consistencythe same name everywhere | Name, address and contact details are aligned across website, public listing and profiles, because a difference between them counts as two different businesses. |
Local searches stop splitting between two versions of the same business. |
| Measurementtwo dashboards | On one side indexed pages, impressions and average position. On the other a fixed list of questions asked again every month to an assistant. |
You know whether you are appearing, where and since when, instead of inferring it from revenue. |
Measurement on assistants, which almost nobody builds
For assistants there is no Search Console, so the dashboard is built by hand, and this is the piece that holds up the rest. A list of questions is set, the same question is asked again every month from a clean session, and you note whether the business appears, at what point and with what sentence.
The clean session isn't a detail. An assistant that already knows you because you talked about it ten minutes earlier will name you anyway, and that answer measures nothing. The same question, with no history behind it, is the only one that says something.
The dashboard also serves to stop. A front with no exit condition drags on by inertia, so you set beforehand how many questions must change status by which month for the work not to be considered a failure.
Mentions count, and nobody buys them
Ahrefs compared 75,000 brands and their appearances in generated answers. The highest correlation of all isn't a website factor: it is mentions on YouTube, at 0.737. Next come mentions of the name on the web, between 0.656 and 0.709, and anchors with the name, between 0.511 and 0.628.
Domain authority, which is what almost all the consultancy in the sector is sold on, stops between 0.266 and 0.326. These numbers should be read for what they are, and the authors write it themselves: «correlation isn’t causation», meaning they could be the sign of an already strong brand rather than a lever.
The practical consequence is only one, and it moves the work off the website. Appearing where your trade is discussed, with your name written in full, counts at least as much as the page. Chasing fake mentions doesn't count, and Google says so in the same guide.
What Google says not to do, word for word
In its official guide to optimising for generative features Google takes apart almost everything sold today as optimisation for AI. Word for word: «You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search».
On structured data the position is just as clear: it isn't required by generative features and there is no special schema to add. It remains useful for what it has always done, that is telling a machine what the page contains, and the system uses it for that.
There is also a field test. Across 1,885 pages that added structured data, compared with four thousand control pages, citations didn't rise: +2.4% on one feature, +2.2% on another, both indistinguishable from zero.
What goes out on its own and what waits for a person
Everything that doesn't commit the business goes out on its own: gathering the questions, the monthly measurement, technical checks on the pages and flagging whatever has broken. These are repetitive jobs nobody looks at until they cause damage.
Nothing that gets published goes out on its own. Every page, every listing and every line bearing your name goes through a person who reads and approves it before it exists in public. The full reasoning about the human constraint is on the page about the principles we build with.
The same thing changes name with the trade
The mechanism is identical; the question isn't. It is worth looking at your own case, because that is where you see which question is worth a page.
| Sector | The question that brings a buyer | Where you see it |
|---|---|---|
| Food and agricultureand export | Which documents are needed to ship a sample to a specific country, and who signs them. Whoever writes it is about to make that shipment. |
|
| Restaurantsand bars | How to book, what is on the menu for someone with an intolerance, whether you can bring a dog. Questions asked while deciding where to go. |
The service of a dining room where the phone rings unanswered |
| Hospitalityaccommodation and events | How many people fit in a hall, which dates are still free in season, what is included. Questions that come before the quote request. |
What this piece doesn't do
It doesn't promise a position and doesn't promise a citation. Nobody controls a search engine's result, and whoever guarantees first place is guaranteeing something they don't own. What is guaranteed is the work and the measurement that says whether it is paying off.
It doesn't buy mentions and doesn't sign the business up to directories that promise a link. Many free profiles openly state that they give no followable link, and the item that matters has to be checked one by one before spending time on it.
It doesn't write the texts on its own. Producing the content, with declared human review, is another piece: it is handled by sales copy and materials. And the page where visitors land is a different piece again, the integrated digital storefront.
Questions and answers
Do you need a special file to be read by AI?
No, and whoever sells one is selling something Google says it ignores. In its official guide to optimising for generative features, Google writes that you don't need new machine-readable files, AI text files, markup or Markdown to appear in Search, and that there is no special schema to add.
What counts is that the page really answers a question someone asks, and that the answer sits in a passage that stands on its own.
If the assistant cites the business but nobody clicks, what's the point?
The point is that the citation reaches the person who is deciding, and at that moment the click isn't the gesture that matters. The Pew Research Center followed the traffic of 900 US adults across 68,879 searches in March 2025: when the generated answer appears, people click a result 8% of the time against 15% without it, and only 1% click the sources cited inside the answer.
Whoever finds you this way arrives later, and arrives already knowing what you do. That is why the measure isn't traffic but presence in the answer, checked on fixed questions.
How long does a new page take to appear?
On a young domain it takes weeks, not days, and no lever buys that time. The things that shorten it are known and all of them apply: requesting indexing one page at a time instead of in bulk, linking the new page from pages already indexed, and publishing gradually.
The signal to watch in Search Console is the ratio between indexed pages and submitted pages. If pages are added and that ratio drops, the engine is telling you it isn't keeping them.
Which questions is it worth working on?
On the narrow, operational questions a buyer actually writes, never on the category. A page aimed at a whole sector competes with anyone who has a budget, and answers nobody.
A page that answers a precise question, the kind you write when you have a concrete problem in hand, has far fewer people in front of it and reaches someone who is already looking for how to solve it. The questions are chosen together, starting from those that already reach you by phone.
How do you measure whether it's working?
With two separate dashboards, because they measure two different things. On search engines you look at indexed pages, impressions and average position in Search Console, split between mobile and desktop, because they often tell opposite stories.
On assistants you keep a fixed list of questions, ask the same question again every month from a clean session and note whether the business appears and in what form. The second dashboard is built by hand, and without it you know nothing.
Notes on sources
- The 8% against 15% of clicks and the 1% on cited sources come from the Pew Research Center, July 2025, on recorded real browsing. The sample is 900 US adults, observed in March 2025 across 68,879 searches: we write this because whoever reads this page isn't that sample, and Italian search habits could move differently.
- The 34.5% drop in click-through rate in first position comes from Ahrefs, on 300,000 keywords split half and half between those with and those without a generated summary, March 2024 against March 2025. It is a company that sells tools for organic search, so it has an interest in the result: we cite it because the method is declared and the comparison is between groups.
- The correlations between mentions and appearance in generated answers come from Ahrefs, on 75,000 brands with a domain above an authority threshold. They are correlations, not causes, and the authors state it in the text: they could be measuring already strong brands rather than usable levers.
- What Google considers useless is set out in its guide to optimising for generative features. The evidence on structured data is the experiment on 1,885 pages against four thousand controls. Those pages were already receiving more than a hundred citations each, so the result says what happens to those who are already visible, and might not hold for those who aren't yet.
- We don't publish any percentage increase in traffic or citations. The writing techniques used here have a measurement behind them, taken from studies on large samples, and they remain a way of choosing how to work. They aren't a forecast about your case.
- This page doesn't report results obtained for a client, because this piece hasn't yet been delivered to a client. The tests cited are functional checks carried out in testing.
The other pieces in this group
Customers who are out there, and don't reach youFifteen minutes, with your case in front of us.
Try typing into an assistant the question a customer asks you on the phone, and see which three names come out. If yours isn't there, that is already the first line of the dashboard. In fifteen minutes on the phone we look at it together and tell you where it makes sense to start, even if we don't end up working together.
You get Mattia Esposito, who then builds the system: there's no salesperson in between. If you'd rather measure on your own before talking, the Diagnostico (in Italian) is twenty questions and five minutes.