Checking your AI visibility every now and then isn't enough: a single data point tells you nothing, and you can vanish from the answers for months without noticing. A restaurant that loses visibility in February and finds out in July has already burned the season. Building a systematic quarterly reading is the only way to tell a real problem from a normal fluctuation — and to step in while it still matters.
In 2015, SEO agencies produced quarterly reports. In 2026, AI visibility demands that same quarterly discipline: the trend matters more than the single data point.
Let me explain it right away with an example I ran into in Valeggio sul Mincio, in the province of Verona (the Mincio hinterland, about ten kilometers from the lower Garda). A Michelin-starred restaurant in the area, one of those names that live on international bookings. In January it appeared on three important AI queries (“best Michelin-starred restaurants Lake Garda”, “gourmet restaurants near Sirmione”, “where to dine well at Garda”). In February, overnight, it had disappeared. Not from Google: from ChatGPT and Perplexity. Without a structured quarterly report, we would have discovered it in July, reading the booking numbers from Northern Europe.
Let me walk you through how quarterly trend analysis applied to visibility in AI answers works, why it really matters, and how you can do it too without a big-company budget.
Why the quarterly window, and not monthly or annual
AI models are updated every 2-4 months. It’s not an opinion: it’s the cadence observed over the past two years across ChatGPT, Claude, Gemini and Perplexity. Each release brings a new training set, new weights on sources, new criteria for selecting citations.
In the world of information retrieval and ranking research, it has long been documented that every model change introduces systematic shifts in source selection. The practical consequence for your business is that a monthly analysis is too noisy (it picks up variations with no statistical significance), while an annual one is too slow (you notice it once you’ve already lost two quarters of bookings).
It follows that the quarter is the right window: long enough to absorb the noise, short enough to intercept the effect of a major update before it does you harm.
What to actually measure every 90 days
The temptation is to collect hundreds of metrics. Resist it. A quarterly report that works focuses on four families of data, and each one answers a business question.
AI citations on strategic queries. How many of your 20-30 priority queries cite you today, on which engines, in what position within the answer flow. For the Valeggio restaurant the queries were things like “Michelin-starred restaurant romantic dinner Lake Garda” or “where to eat lake fish near Peschiera”.
Share of voice vs direct competitors. If in your sector the AI cites 8 brands, are you one of the 8 or are you out? For the starred restaurant the competitors were another 6-7 starred venues between Bardolino, Salò, Sirmione and Desenzano: the question was whether ChatGPT listed them all, or only the top 3.
Health of authority assets. Your Google Business Profile, mentions in industry guides, your Wikidata entry if you have one. These are the signals that keep you inside the citation window, and you must check them every quarter with Google Business Profile and Wikidata.
Quarter events that can explain the swings. New model announcements, citation policy changes, modifications to your site, important new reviews, press articles. Without this column the report is a collection of numbers with no cause.
Without a baseline snapshot, the first quarter has nothing to compare against and you lose 90 days.
The report template I use (4 sections)
Here I’ll give you the concrete structure. You can keep it in a Google Sheet, in Notion, in a PDF: the form matters less than the discipline of filling it in every 90 days.
Section 1 — Quarter snapshot. A table with the 20-30 priority queries on the rows and the AI engines (ChatGPT, Claude, Gemini, Perplexity) on the columns. In each cell a binary value: cites you / doesn’t cite you. Below it, the same table from the previous quarter, and the difference.
Section 2 — Gains and losses. Two side-by-side lists. On the left “what worked”: queries where you entered for the first time, engines where your presence consolidated. On the right “what to investigate”: queries where you dropped out, competitor brands that appeared in your place.
Section 3 — Relevant events. A timeline of the quarter with two columns: your events (site changes, new content, PR earned) and market events (releases of GPT, Claude, Gemini, known changes in citation criteria). It serves to read cause and effect instead of reacting in panic.
Section 4 — Three actions for the next quarter. Not twenty. Three. The ones that, if done, have the highest probability of moving the needle. For the Valeggio restaurant in Q1 they were: obtaining a stable Wikidata entry for the chef, fixing the Restaurant schema with awards/menu fields, earning 3 mentions in internationally recognized gastronomic guides.
Measure the queries where the customer doesn’t know you yet: that’s where AI visibility is won or lost.
The test to run now, in 30 minutes
Before building the full report, take your baseline snapshot. Without a baseline snapshot, the first quarter has nothing to compare against and you lose 90 days.
- Open a sheet. Write 15-20 queries a customer would use to search for what you sell (for the restaurant: local and national gastronomic queries). Not brand queries: need queries.
- Open ChatGPT, Perplexity, Gemini and Claude. Run each query on each one. Mark in the grid who cites you and in what position.
- For each query also note the 3-5 brands that get cited in your place or alongside you. They are your real competitors in the AI space, and they often don’t match the ones you have in mind.
- Save everything with the date. In 90 days repeat it identically.
This is an entry-level check: it gives you direction, not the precision of a study. Serious analysis with statistically robust samples, automated tracking and intent segmentation requires professional tools. But to get started and understand where you stand, it’s more than enough.
The mistakes I see most often in quarterly reports
Mixing brand queries and need queries. If you only measure “restaurant [name]” it will always cite you. Measure the queries where the customer doesn’t know you yet: that’s where AI visibility is won or lost.
Changing the queries from one quarter to the next. The measurement queries must stay the same for at least 4 quarters. Otherwise you’re not measuring your visibility: you’re measuring different queries.
Not noting the events. Without the timeline of your events and market events, the report becomes a collection of numbers with no cause. And without a cause you don’t know what to repeat and what to correct.
Reacting to the single data point instead of the trend. If in one quarter you disappear from 2 queries out of 25, it may be noise. If over the next two quarters you disappear from another 5, it’s a signal. Wait for the trend.
How does all of this tie back to visibility in AI answers?
Quarterly trend analysis is not a technical exercise: it’s the way to understand whether the things you do (work on E-E-A-T for the AI, inverted pyramid, author entity recognition) are truly shifting your presence in AI answers, or whether a model update penalized you and you’re chasing a ghost.
In the previous articles of this series I showed you how to define the right metrics and how to monitor the individual indicators. In the upcoming ones we’ll look at the real-time control dashboard for critical alerts, the industry comparison to position yourself against your real competitors, and the 12-month projection that lets you plan instead of chasing emergencies.
The quarter is the heartbeat of all this. Skipping it means noticing too late, and with AI visibility too late means months of lost bookings without knowing why.
