What to show in an AI visibility reporting meeting (and why you should never do the demo live)
There's a huge temptation in AI visibility reporting meetings: open ChatGPT in front of the client, type the prompt and let the magic happen live. "Look, I'll show you right now."
Don't do it. Ever.
The AI doesn't answer the same question the same way twice. Last week your client came out recommended first; today, in front of them, with the projector on, they might not come up. Or worse: their competitor might. And at that point there's no reporting meeting left — there's a crisis meeting that you brought on yourself.
This article covers both halves of the problem: what to bring so the meeting works, and how to turn that AI volatility — which looks like your enemy — into your best sales argument.
Why the live demo is Russian roulette
AI search engines generate each answer on the spot. There's no fixed "ranking" like in Google, where yesterday's position 3 is usually today's position 3. Every query is a roll of the dice: the model can cite different sources, order the recommendations differently or simply change its mind depending on the nuance of the question, the time of day or nothing at all.
That means a single query proves nothing — for or against. If you do the live demo and the client comes up, you got lucky. If they don't, you've lost the meeting. And in both cases you've taught the client a flawed way of measuring: now they'll open ChatGPT on Sunday afternoon too, ask one question and email you the result, whatever it is.
The serious way to measure AI visibility is with periodic samples: the same prompts, in the same AIs, every week, logged with a date. Just as nobody measures a website's traffic by looking at the visitors in one specific minute, nobody should measure AI presence with a single isolated query.
That's the line to drop into the meeting, almost word for word:
Key data
"I'm not showing you live because a single query is a coin toss. What I'm bringing you is the same 40 questions, sent every week to the 4 AIs, over the last month. That's a measurement."
You've just turned your "no" to the demo into a demonstration of rigour. And along the way you've explained why your service makes sense as a monthly retainer rather than a one-off glance.
What to bring: the four pieces
1. Dated screenshots, not descriptions
"You appear in Perplexity" is worth nothing without proof. Bring screenshots with the date visible: the question, the full answer and the client's mention highlighted. The dated screenshot does two jobs: it's today's evidence and it'll be the baseline against which you prove improvement three months from now.
If the client comes out badly — outdated data, wrong description — bring that too. Showing the problem with a screenshot builds more trust than hiding it, and it opens up new work: for that case we have a full protocol.
2. Trend, not a still photo
The figure that justifies your invoice isn't "you appear in 12 answers", it's "in January you appeared in 4, today in 12". Bring a time series, however modest: the percentage of prompts the client appears in, week by week, per AI.
The first meetings won't have a trend to show — they'll have a baseline. Say it plainly: "this month we've established the starting point; from here on, each meeting compares against this." It's a legitimate deliverable and the client understands it.
3. Comparison with the competitor
Nothing puts a figure in context better than the competitor next door. "You appear in 30% of the questions" sounds so-so; "you appear in 30% and your main competitor in 10%" sounds like a win; and "your competitor is at 55%" turns the meeting into an action plan rather than a reckoning.
The comparison also protects you: if the whole sector loses visibility one month (it happens, because the models change), the relative figure proves it's not down to your work. In our study of 9,865 Spanish SMEs across 30 sectors, 91% only appeared in 1 of the 4 AIs (full data here) — in most sectors, beating the competitor is surprisingly within reach.
4. The actions and their why
Always close with what you've done and what's coming: which sources have been corrected, what content has been published, which new prompts have been added to tracking. If you're unsure which questions are worth tracking, we have a guide for choosing each client's prompts.
This matters because AI visibility takes time to move and sometimes moves for no apparent reason. In the months when the chart is flat, the work delivered is what holds the meeting together. Which metrics to commit to and which to only report is a meaty topic in its own right: we develop it in what you can promise about AI visibility.
The structure of the meeting, in order
- The headline (2 minutes): this month's coverage figure against last month's. One number, one sentence.
- The evidence (5 minutes): three or four selected screenshots — the best answers and, if there is one, the worst. No screenshot marathon.
- The competitive context (5 minutes): the comparison with the rival.
- The work (5 minutes): done this month, planned for next.
- Questions — and this is where someone will say "can you show me live?".
When that question comes — it will — don't get defensive. Offer the deal: "We can try it if you like, but for the record a single query isn't the data; the data is this", pointing at the time series. If the live query goes well, great. If it goes badly, you've already explained why that doesn't invalidate the month of measurements. You've taken the bullet out of the chamber before pulling the trigger.
Generating all this material by hand — sending dozens of prompts across 4 AIs every week, saving screenshots, maintaining the series — isn't viable for a client roster. It's exactly what Surfeo for agencies automates: weekly tracking of each client across ChatGPT, Gemini, Perplexity and Claude, with history, competitor comparison and PDF reports ready to take into the meeting.
Frequently asked questions
What if the client runs the query live on their own and doesn't come up?
They'll write to you, for sure. The answer: "That's exactly why we measure with weekly samples and not with single queries: an individual answer can vary, the trend doesn't lie. In this week's measurement you appear in X of Y questions." If you explained the volatility at the first meeting, this email gets answered in two lines. If you didn't, now you have to give the explanation you should have given then.
How often should the reporting meeting happen?
Monthly for the meeting, weekly for the measurement. AI visibility moves slowly: meeting every week means showing noise; measuring only once a month means missing the movements. A monthly report with data from 4-5 weekly samples is the balance.
How many screenshots should I show?
Three or four, chosen. The temptation to show all forty to justify the work is understandable and counterproductive: the client switches off by the eighth. The volume goes in the report annex; the meeting is for the headline and the trend.
What do I do the first month, when there's no trend to show?
Sell the baseline for what it is: the foundation. "Today we know exactly where you appear and where you don't, question by question and AI by AI. It's what lets us prove the improvement to you from here on." And pair it with the action plan, which is the real deliverable of the first month.
If you don't yet have your clients' baseline, start today: take the free AI visibility test with any of their websites and you'll have the first dated screenshots for the next meeting.
