You bring your CEO a detailed analysis of AI visibility and they nod, say it's interesting, then the budget doesn't get unlocked. The problem isn't the strategy: it's that the person making the decision wants three numbers in two minutes, not fifteen slides. Presenting the same information in the right format completely changes the outcome of the conversation — and the difference between an approved budget and one postponed indefinitely.
Every business owner who pays me to work on visibility in AI answers sooner or later asks the same question: “so, how are we doing?”. They don’t want 15 slides, they want 3 numbers in 2 minutes. Here’s the format I use.
If it has ever happened to you — even just once — to look at a 20-page report on your brand’s presence in ChatGPT, Claude and Perplexity and set it aside after thirty seconds, you know exactly what I’m talking about. The problem isn’t the data. The problem is the format.
Over the past few months, working with companies trying to break into AI answers, I’ve understood something that seems obvious but almost nobody applies: decision-makers don’t read technical reports, they read a single page with the decision already prepared. The monthly summary I’ll explain here is the document that turns 40 hours of analysis into five minutes of reading and — if it’s written well — into the decision to keep spending on it.
Let me explain how I build it and why it works.
What I mean by an AI visibility summary
One page. Three numbers. Three actions. Stop.
In my practice I call this format 3+3 and I’ve used it ever since I stopped sending long PDFs. The logic is simple: whoever pays for this work doesn’t want to understand how RAG (Retrieval Augmented Generation) works, they want to know whether their brand is gaining or losing ground in the answers of AI assistants, and what to decide in order to improve.
There’s no academic paper telling me “do it like this”. This is my own deduction, built on three documented principles that I’ve covered in the previous articles in this series:
- Presence in AI answers is measured in citation share of voice, not in ranking (I talked about it in the series on measurement)
- A brand’s authority is recognized by the models through patterns of consistent citation — see the piece on backlinks as a citation proxy
- Entities recognized by the Knowledge Graph have a higher probability of being cited — I wrote about it in Google Knowledge Graph entry
From these three principles it follows that whoever leads the company must see, every month, the AI visibility delta, one thing that worked and one thing that is broken along with the estimated cost to fix it. Nothing else.
The Loreto case: how the page with 3 numbers got the spend approved
I work with a hospitality business in Loreto, in the province of Ancona, specialized in religious and Marian tourism. For those unfamiliar with the context: Loreto is one of the most visited Marian pilgrimage destinations in Europe, and the ecosystem of hotels, pilgrim houses and agencies that organize groups lives on a very specific seasonality — Easter, May (the Marian month), August, October.
The owner had asked me to figure out why, when a parish priest or a pastoral coordinator searched ChatGPT for “where to stay in Loreto for a diocesan pilgrimage”, their brand didn’t show up. Two competitors in the area showed up and — systematically — the Shrine’s portal.
I summed up the first two months of work in an 18-page report. Beautiful, detailed, with share-of-voice charts query by query. Result: the decision to continue the work was postponed twice. The owner didn’t have time to read it.
In the third month I sent a single page. Three numbers at the top, three actions at the bottom. Green light in 48 hours.
Proposing more than one action at a time.
The three numbers I always put at the top
They aren’t numbers made up on the spot, they are the only three a business owner understands without explanations.
Number 1 — AI Share of Voice vs the previous month. How many times the brand was cited by the main AI assistants (ChatGPT, Claude, Gemini, Perplexity) across a fixed set of relevant queries, compared with the month before. For Loreto the queries were things like “best hotel in Loreto for pilgrimages”, “where to sleep near the Shrine”, “accommodation for parish groups Loreto”. A set of 30 queries, repeated identically every month. The number is a percentage: “the brand is cited in 23% of the answers, it was 17% last month, +6 points”.
Number 2 — A win to celebrate. A concrete citation, copy-pasted from an AI assistant’s answer. For Loreto, in the third month, it was this: “On Perplexity, for the query ‘organizing a diocesan pilgrimage to Loreto’, the brand is cited as one of the three recommended properties for parish groups, with an explicit mention of the in-house chapel”. One sentence. No charts.
Number 3 — A critical gap with action and cost. The most urgent problem, framed as a decision: “On German-language queries (Loreto has strong traffic from the Bavarian and Austrian area), the brand never appears. Proposal: professional translation of 12 key pages + a Wikidata entry in German. Cost: X euros. Time: 6 weeks. Expected effect: appearance in AI answers in German within 90 days”.
Three numbers. One page. Decision made.
Build the frozen query set — don’t change it month by month, otherwise the delta means nothing.
Why this format works where others fail
Put in practice: the business owner isn’t evaluating AI visibility in the abstract. They’re evaluating whether over the next three months it’s worth spending more money on it.
A 20-page report requires you to build the summary yourself. No SME owner does that, they don’t have time. The 3+3 page hands them the summary already made — and leaves them only the decision to say yes or no to the single proposed action.
There’s a second reason, more subtle. When you propose a single action with its cost, the answer is yes or no. When you propose five, the answer is zero because “let’s talk about it again next month”. One action per month, twelve actions in a year, is much more than “twenty actions proposed across five reports that were never decided on”.
The test you can run in 30 minutes next week
You don’t need a professional tool to build the first version of this page. You need discipline on the format.
- Choose a fixed set of 20-30 queries representative of your sector. For a notary firm in Verona they could be “best notary Verona for corporate deeds”, “notary Verona inheritance”, “notary firm Verona online”. For a winery in Friuli “best Ribolla Gialla producer”, “Friulian white wine to pair with fish”, “wineries to visit in the Collio”.
- Open ChatGPT, Claude, Gemini and Perplexity. Run each query on each one. Count in how many answers your brand appears. Save the numbers.
- Repeat the next month, same queries, same assistants. Calculate the delta.
- Write one page. Number 1 at the top: share-of-voice delta. Number 2: the proudest citation you found. Number 3: the biggest gap and what you propose to do.
An indicative test, not a scientific study. It’s not the same thing as continuous monitoring with professional tools — that requires a serious setup. But it’s enough to get the decision on the first spend going.
The mistakes I see most often when writing to decision-makers
Putting bare percentages with no time reference. “AI Share of Voice 18%” says nothing. “AI Share of Voice 18%, it was 11% in January, +7 points” says everything.
Proposing more than one action at a time. Three actions on the page = zero decided. One action = a binary decision, yes/no.
Using agency jargon. “We optimized the entity fingerprint on the Knowledge Graph” isn’t a win for the business owner. “When a customer asks ChatGPT who we are, the right answer now comes up” is.
Missing the number for the cost of inaction. Not FUD, facts. “Over the last 90 days the two competitors cited in our place gained +12 points of share of voice in German. Every month that passes without action on the language content, the gap widens”. That works.
What to do Monday morning?
- Decide the fixed day of the month on which you send the page. For me it’s the first Monday. Never skip it.
- Build the frozen query set — don’t change it month by month, otherwise the delta means nothing.
- Choose the four AI assistants you measure on: ChatGPT, Claude, Gemini, Perplexity. They’re the four your customers use.
- Limit the page to one A4 side. If it doesn’t fit, you’ve written too much.
- Always benchmark against the 3-5 competitors that the AI cites in your sector, not against an abstract universe.
A page built like this isn’t the most revolutionary format in the world — it’s simply the format that works when you need to break into AI answers and you need someone in the company to say yes to the spend. It’s not a magic factor, and it isn’t enough on its own: behind it there has to be real work on entities, content, authorship — the kind I wrote about in Author Entity Recognition and in Event Entity and Speaking Authority.
But it’s the piece that closes the loop: without a format that decision-makers can read, even the best AI visibility work stays invisible inside the company. And work that’s invisible inside the company is work that, sooner or later, loses the spend that sustains it.
In the next articles in this series I cover the operational document for those who work on marketing every day (more detail, weekly frequency) and the technical memo for whoever manages the site (parsing, schema, robots.txt). They’re three different documents for three different readers — confusing them is the fastest way to make sure none of the three gets read.
