Dashboards full of AI visibility metrics look impressive, but which means what? Why the emperor's new dashboard is old vanity metrics one floor up.
I had a client once who only ever asked one question.
He ran e-commerce at a large UK electronics retailer, and he was one of my favourite people to present to, which is a strange thing to say about a man who made quarterly reviews genuinely uncomfortable. We'd bring our slides. Someone on my team would say, "Organic visibility is up three per cent." And he'd let it sit for a beat. Then: "Which means what?"
If you survived that, there was a second one behind it. "So, I should do what now?"
He wasn't being difficult. He was doing his job. He was in that room to make decisions, not to be informed. What he understood, long before most of us did, was that much of what is presented in marketing meetings is not information at all. It's decoration. I've been thinking about him a lot this month because there is a spectacular amount of emperor's new clothes going around at the moment, and almost all of it is gathered in one place.
Everyone can see the suit
AI visibility is the thing now. Are you showing up in ChatGPT? Are you cited in AI Overviews? What's your share of voice against your competitors inside a model that doesn't publish anything about how it chooses?
There is an entire tooling category built around answering that. Beautiful interfaces. Confident scores. Trend lines that go up and down with real conviction..
And I want to be careful here, because I'm not saying the people building these tools are frauds. Some of them are doing serious work on a genuinely hard problem. But I've read the methodologies, and the honest ones will tell you this themselves: nobody can see the prompt stream.
The platforms don't publish query volumes. So, the tools estimate them from consumer panels, clickstream data and extrapolation models. They then run a set of invented prompts against the models and record whether your brand appears. That's the mechanism. An agency that happily buys several of these platforms put it plainly earlier this year: the volume figures are probabilistic estimates, and the mention rates vary between runs.
Which is not a scandal. It's a structural property of the medium. Large language models generate answers probabilistically. Ask twice, get twice.
But it does mean the number on the dashboard is a simulation of your market, not a measurement of it.
Alright, but Google publishes it now
This is the part where I declare that something has changed.
On 3 June this year, Google launched dedicated generative AI performance reports in Search Console. These provide first-party data from Google on how your pages appear in AI Overviews and AI Mode.
Genuinely useful. And I'd encourage every retailer reading this to go and look at theirs.
Now, here's what it gives you. Impressions. Which pages. Which countries. Which devices. Change over time. And here's what it doesn't give you. Clicks. Click-through rate. The queries. Conversions. Position.
So, the single most authoritative source of AI visibility data hands you a number that has gone up, with no way of knowing whether anybody clicked, what they were asking, or whether any of it sold a single thing. Which means what? So, I should do what now?
That was the whole point, and it's worth saying twice: the problem was never that the data was hidden. It's that the data, hidden or published, doesn't tell you what to do.
We have done this before
I’ve got three words for you: social media monitoring.
Remember that? We absolutely had to understand the Twittersphere. Who was talking about us, in what tone, trending where, against whom. An entire industry appeared to tell us. And it told us, and we looked at it, and then most of us quietly stopped.
Before that, it was big data. Before that, machine learning. Each arrived as a measurement layer, years ahead of any action layer. Each produced a dashboard first and a decision second, if at all.
The martech landscape is the receipt. Around 150 solutions in 2011. 15,505 this year. Roughly a hundredfold in fifteen years. And 2026 is the first year it stopped growing, up by less than one per cent, with 1,488 products added and 1,367 removed.
That's not consolidation. That's churn. That's a market where a tool arrives, is bought, is ignored, and is replaced by another tool that measures the same thing slightly differently.
It’s the same disease, just one floor up
And while all this is going on, we've quietly installed the next meaningless metric.
Token usage. Consumption is the new pricing model, so consumption has become the new measure of productivity, which is roughly like assessing a writer on the number of words they typed.
The bills have started arriving. Sixty-eight per cent of US companies now say at least some of their AI initiatives ran over budget in the past year, and a third say it happens mostly or always. Only nine per cent say more than three-quarters of their AI work produced a measurable financial return.
My favourite detail in all of it: 41% of firms surveyed by KPMG said they'd consider "token-maxxing" – gamifying token consumption with incentives and leaderboards. We have invented a scoreboard for the thing we're being charged for.
That's two floors of the same building. Downstairs, we're paying for a report on our AI visibility. Upstairs, the CFO is opening an invoice no one forecast. In both cases, we've mistaken the instrument panel for the engine.
If you want to move the needle…
I ran an SEO agency for a long time. The hardest lesson I learned, and I learned it slowly, was this: no volume of reporting and no amount of account management ever improved anyone's visibility. Not once. Not ever.
The only thing that moved it was understanding what customers actually typed, then pushing everything you own in that direction. Relentlessly. Product data, category structure, page copy, navigation, attributes, the lot.
And if you think that sounds like an old man defending his old trade, take it up with Google, who published guidance in July that says the quiet part out loud. Their generative AI features are rooted in core Search ranking and quality systems.
Optimising for generative AI search, in their words, is optimising for search, and therefore still SEO. There is no llms.txt trick. You don't need to chunk your content. You don't need to rewrite pages for machines. And on the tools: “no third-party tool has access to our internal ranking or AI systems”. (Google)
These last twelve words from the only organisation that can actually see inside the box.
So, the thing that wins in AI is the thing that always won. Product data that speaks the language your customer uses, rather than the language your buying team files it under.
That's the gap we spend our lives in at FoundIt! Customers search by problem, occasion, situation, or by what's gone wrong in their kitchen at nine on a Sunday night. Catalogues are organised by supplier hierarchy and buying structure. And the space between those two things is where revenue quietly falls through the floor. Once you see it, you can't unsee it. You start noticing bad product data everywhere. You also start noticing opportunity everywhere, which is the better half of the deal.
No visibility score will ever surface that for you. It isn't pointed at your catalogue. It's pointed at the weather.
So, if the dashboard isn't the answer, what do you actually do on Monday? Here are my three suggestions to get you started:
1. Kill one report and fund one fix. Take the budget from your most impressive-looking AI visibility subscription and invest it in product data enrichment for one category. The diagnostic: if this report disappeared tomorrow, which decision would change? If nothing changes, you were never measuring. You were reassuring.
2. Go and read what your customers actually type. Your site search logs. Your zero-result queries. Your support tickets. The question-shaped queries in Search Console. That is real prompt data, from real customers, with real intent, and you already own all of it. The diagnostic: when did anyone on my team last read a hundred of these end-to-end?
3. Pick one category and be relentless for a quarter. Not a pilot across the whole estate. One category, enriched, restructured, optimised, then done again. The diagnostic: am I judging this by traffic and sales, or by a score?
That's it. Focus and execute relentlessly.
It isn't a new instrument panel that turns a business around.
David Marquet took the USS Santa Fe from the worst-performing submarine in the fleet to the best by consistently saying and doing the right thing until it stuck. He now runs the Intent-Based Leadership Institute, which I enjoyed more than I probably should have.
The emperor isn't wearing anything. He never was. But the crowd is having a wonderful time, and the tailors are doing very nicely indeed.
Get the basics right, and the score takes care of itself.
Until Tuesday after next.
Warren.
Sources Google Search Central, Introducing Search Generative AI performance reports in Search Console, 3 June 2026 · Google Search Central, Optimizing your website for generative AI features on Google Search, updated 10 July 2026 · Search Engine Journal, Google Now Reports AI Search Impressions, June 2026 · Search Engine Land, 7 hard truths about measuring AI visibility and GEO performance, January 2026 · Brainlabs, Your AI Visibility Data Is Wrong (And That's Okay), April 2026 · Digiday, Marketers question expensive AI visibility tools, June 2026 · chiefmartec / MartechTribe, 2026 Marketing Technology Landscape, May 2026 · CFO Dive, Nearly 7 in 10 firms report AI cost overruns, July 2026