Getting Cited by AI: How to Measure, and Build, Visibility in Generative Answers

There is a scene that keeps repeating in marketing departments between Milan and the industrial belt around it. The sales director walks into the meeting and says: “I asked an AI assistant who makes this component in Lombardy, and three of our competitors came back. We did not.” Silence follows, then the hard question: how do we fix it?

The short answer is that it is fixed the way SEO has always been fixed — with content that deserves to be cited and a structure that makes it citable — with one substantial difference: before acting, you have to be able to measure. Without measurement every action is a bet. This guide covers how to measure visibility in generative answers, what the dedicated module actually does, and which changes to a page genuinely move the result.

Why classic rankings are no longer a sufficient measure

Organic ranking measures one precise thing: where your link appears when somebody types a query into a search engine. It is still a fundamental metric, but it describes only part of the journey.

In generative answers the mechanism is different. The model builds a synthetic answer and cites a handful of sources. There is no “position three”: you either exist inside that answer or you do not. And the selection does not necessarily reward whoever ranks first — it rewards whoever has pages the model can read, understand and reuse as a reliable fragment.

This produces counter-intuitive situations we see often: companies with excellent organic rankings completely absent from AI answers, and small manufacturers with modest traffic cited regularly because they have one well-written technical page with concrete data and a clean structure.

What the AI visibility module actually measures

The module works on a live index of how the models answer: more than 2.8 million sites classified, over 1.2 billion AI queries monitored and more than 52 million topic categories. On that basis it returns, for any domain, a handful of operational facts.

That last row is the one that changes the tone of a meeting. Knowing you are present 30% of the time says nothing on its own; knowing that for “who are the best suppliers of X in Lombardy” the model names three competitors and cites their technical sheets is information you can build a plan on.

The test we suggest running before anything else

Before any investment, one hour of work. Open the panel, enter your own domain, then one at a time the three competitors your sales team mentions most. Note four numbers for each and look at the cited pages.

In ninety per cent of the cases we have followed, the same pattern emerges. The competitors’ cited pages are not their home pages, and not their “about us” pages: they are product sheets with complete specifications, comparison pages, technical guides, question-and-answer sections. They are pages that answer, not pages that introduce.

The cross-check against search data

The second step is to place the same picture as the classic engines see it alongside. In the Search Console module, look at which pages receive real impressions; in the rank tracker, check where the domain sits on the same queries, and where the competitors sit. The full picture comes from the comparison: some are strong in the SERP and absent from AI answers, some the reverse, and some have an indexing problem upstream of both.

What makes a page citable

Analysing hundreds of cited pages, the same characteristics recur. None is a revolutionary discovery, but together they decide the outcome.

The content is in the HTML. Text that appears only after JavaScript runs, or inside components loaded later, is far less likely to be collected. It is the first technical check to make, and on Milanese fashion sites built with modern frameworks it is a more frequent problem than people expect.

The answer comes before the preamble. A section that opens with “the short answer is six to ten working days” and then explains why is far more citable than one that builds the reasoning over three paragraphs before reaching the point.

The numbers are explicit. Measurements, lead times, tolerances, prices, production capacity. A model happily cites a verifiable figure and rarely cites an adjective. “We deliver quickly” is worth nothing; “standard delivery in 6–10 working days, 48 hours with a surcharge” is citable material.

The structure is predictable. Headings that are questions, paragraphs that answer them, tables for comparisons, lists for procedures. The same structure that makes a page readable for a hurried customer makes it usable for a model.

The identity is clear. Who wrote it, which company, where it is based, when the content was updated. For companies rooted in a territory, stating the geographic context explicitly — the premises, the area served — helps attribution in answers to local questions.

A realistic quarterly plan

You do not need to rewrite the site. You need to work on a few pages, in the right order.

Three Milanese sectors, three different ways of being cited

The mechanism is the same, but what the model is looking for changes considerably by sector. Here is what we observe in the three best represented in the city.

Fashion and retail

The questions are comparative and about fit: which label suits a given build, where to buy a particular garment in Milan, whether a brand runs large or small. The cited pages are size guides with real measurements in centimetres, comparisons between lines, and hands-on write-ups. Product pages carrying only emotive description never appear. A brand with a serious size guide — a table, notes on fit — gets cited even when it is not the biggest name.

Design and furniture

Here the weight sits on materials, lead times and compatibility: how long a made-to-measure sofa takes, the difference between veneer and solid wood. The period around the city’s big trade events amplifies everything, but the pages cited in those weeks were published months earlier. That is why the editorial calendar in this sector has to be built at least two quarters ahead of the events.

Professional services and B2B

The questions read like a request for a quotation: what service X costs for a company of N employees, which requirements Y involves. What gets cited are pages with price lists or ranges, listed requirements, timings. This is the sector where reticence about pricing costs most: whoever publishes not even an indicative range is simply excluded from the answer in favour of whoever does.

How to take this into a board meeting without losing the room

A visibility score on its own does not survive ten minutes in a management meeting. What survives is one slide with three elements.

The first is the direct comparison: our score next to the three competitors the sales team names, with the change on the previous quarter. The second is the plain list of questions where the competitors appear and we do not — read aloud, those questions land harder than any chart, because they are exactly the questions customers ask on the telephone. The third is the cost of the proposed work, expressed in working days rather than in concepts.

Three mistakes we see repeatedly

Treating AI visibility as a volume contest

Publishing twenty generic articles does not move the score. What gets cited are the pages that answer precise questions best, not the ones covering the most ground.

Postponing because “it does not bring traffic yet”

Citations build months after publication. Whoever starts measuring once the traffic is obvious is already behind whoever started now.

Watching only your own score

The absolute number says little. What counts is the distance from direct competitors and how it moves over time: it is the only figure that justifies a budget in a board meeting.

A note on how often to measure

Visibility in generative answers swings more than organic positions, because it also depends on model updates. Measuring it weekly produces nothing but noise and nervous meetings. The right rhythm is quarterly for the score and monthly for the query list, which moves more steadily and signals sector trends earlier.

When a managed campaign makes sense

If the work described above is sustainable in house, the free side of the platform is enough. When volume and continuity are what is needed — wide catalogues, several languages, foreign markets — the campaigns come into play: AutoSEO at $149 a month with AI-assisted keyword selection, daily link building from the network of more than 230,000 partner sites, on-site suggestions and live reporting; FullSEO from $500 a month with a dedicated team, manual keyword review, premium backlinks targeted by DR and content written by editors.

Frequently asked questions