0 of 3
Markets where a tool was stable
Twelve runs across the US, Spain and France. Not one tool was cited consistently in more than one market.
Agencies and consultancies
Report AI visibility numbers your client can defend.
Clients are asking what AI says about them, and the honest answer is harder to produce than the market admits. EchoWi runs your clients' buying questions repeatedly, per market, and gives you the sources behind each answer rather than a score you would have to explain away.
A best-tools comparison
Who wrote the twenty seven we read
- A vendor listed in its own comparison 22 of 27
- Placing themselves first or second at least 11
- Someone with nothing in the list 1 of 27
What we measured
Five findings that change what you can safely tell a client
All of these are our own runs, with country and language set explicitly, published with the query and the date.
97.7%
Of a vendor's search volume was other companies
One AI tool showed 27,100 searches a month. Disaggregated it was 627, and three names in a row behaved the same way.
22 of 27
Tool comparisons written by a vendor in the list
At least eleven place themselves first or second. One of twenty-seven came from someone with nothing to sell.
2 of 5
Spanish categories with no AI answer at all
Thirteen runs, zero AI Overviews, zero errors. There is nothing to optimise for in those categories yet.
58%
Of cited sources appeared in exactly one run
Fifteen agency buying questions, four markets, three surfaces, 38 runs. Only about a third of what was cited held across every run.
EchoWi runs between March and August 2026, each with country and language set explicitly. Every figure links to the study it came from. Read the three-market measurement.
Why this is your problem before it is your client's
An agency signs up to a number. If the number cannot survive a sceptical client asking how it was produced, the renewal conversation is already lost.
- A client selling into four markets wants one report. Our three-market study says the stable citation sets do not overlap at all, so an average across four can hide being invisible in three and still look healthy.
- Most tools default to the United States and English and do not announce it. Run a Spanish question without changing that and you will report the American market to a Spanish client, formatted identically to a correct answer.
- Answers vary between identical requests. A month-on-month move of three points, on a tool that runs each prompt once, is noise with a number on it.
- If your client's brand name is a common word, mention counting is counting other companies until the vendor can show you the answer text behind it.
The number that moved when we asked the same thing twice
The most useful thing this method produces is not a score, it is a specific gap you can put in an email. In one run, a hotel chain's own official site was the first source Google cited, and the chain is not named anywhere in the answer. When the model has already read the client's page, crawling, rendering and relevance are all ruled out, and what remains is that the page does not answer the question that decides the sale. That is a content brief, and a content brief is billable work.
Seven client sectors, one question shape, and how little holds
We asked which marketing agency is best in seven client sectors, in the United States in English, repeating each question rather than asking once. The column that matters is the second: how many of the cited domains came back in every run instead of appearing once.
- Domains cited
- Held in every run
Law firms 13/20
White label SEO 8/25
Ecommerce 7/24
B2B SaaS 7/19
Healthcare 6/20
Hotels 6/17
Higher education 6/23
| Client sector | Domains cited | Held every run | Runs |
|---|---|---|---|
| Law firms | 20 | 13 | 2 |
| White label SEO | 25 | 8 | 2 |
| Ecommerce | 24 | 7 | 3 |
| B2B SaaS | 19 | 7 | 2 |
| Healthcare | 20 | 6 | 2 |
| Hotels | 17 | 6 | 2 |
| Higher education | 23 | 6 | 3 |
Google AI Overview, AI Mode and Gemini, United States in English, measured 9 and 10 August 2026. Run counts differ because a run that fails is not a run that returned nothing, and the denominator is always what answered. Nothing here says these agencies are good; it says which domains the answer was assembled from on those dates.
What an agency team gets, per client and per pitch
Something you can put in a client report and defend when they ask where it came from.
Every client market run separately
Country and language fixed per run, never averaged into a single figure that hides three markets inside a fourth.
The sources, not just the score
Which domains built each answer, so your recommendation names the publications to get into rather than telling a client to "improve authority".
Repeated measurement by default
Because one run cannot tell a stable citation from a coincidence, and a client who works that out on their own will not take the next report seriously.
Findings you can bill against
A gap that names a page, a query and a competitor turns into scoped work. A visibility percentage turns into a discussion.
We confirm the sector has an answer layer before you promise a client anything
AI coverage is not universal. Measuring five categories in Spain in August 2026, two returned no AI Overview at all across thirteen runs, with no errors, so a client in consumer electronics or eye surgery would have been paying to watch a surface that does not render. Saying that costs us a subscription and saves you a renewal, which is the trade we would rather make with an agency than with a brand.
The agencies we already work alongside
We work with GEO and SEO agencies. And because we publish what we measure: in six runs asking which GEO tool to use, no funded vendor appeared even once. Reddit and YouTube were cited every time. We printed that too.
What agency teams ask us first
The questions that come up on the first call, answered here so you do not have to book one.
Can I put this number in a client report?
Only with its run count next to it, and that is the point of this page. We measured the law-firm agency question twice in the same market: the first reading gave thirteen stable domains out of twenty cited, the second gave seven out of twenty-four. Neither is wrong. A single reading is a draw, and a client who repeats it and gets a different figure will ask you why, so the run count and the date belong in the report.
What do I say when a client asks whether their competitor is winning?
That being named and being cited are different outcomes and they move independently. We report both separately for exactly that reason. A competitor named in the answer while nobody's site is cited is a different situation, and a different remedy, from a competitor whose own pages are the source.
Do you white-label, and can we resell this?
We work with agencies as agencies, not as a badge on a dashboard. What you get is the measurement, the source list and the method written out plainly enough to hand to a client, including the parts that are uncertain. We would rather you keep a client because the numbers survived scrutiny than lose one because they did not.
How is this different from the AI visibility feature in our SEO suite?
Mostly in what it refuses to tell you. Any tool can print a percentage; the questions to ask yours are whether its runs bypass the provider's cache, how many wordings it tracks per intent, and whether it can name the domains a figure is made of. If it cannot answer those, the trend line it draws may be a property of one cached phrase rather than of your client's market.
Which client sectors are worth pitching this in?
The ones where an answer layer exists, and that is not all of them. In our August 2026 sweep, questions asking which local business is best returned no AI Overview at all across thirty questions in three markets, while category and destination questions did. Pitching AI visibility to a client whose questions have no answer is the fastest way to lose the account after the first report.
What does the first measurement deliver?
One client sector, its buying questions run repeatedly on the surfaces that answer them, with which brands were named, which domains built the answer, and which of those held across every run. You get the source list, not a score, because the source list is the part a client can act on and the part you can defend.
See what an agency account looks like
Tell us how many clients and how many markets. We will run five buying questions for one of them and show you the report you would be sending, sources included.
Measure a client sector (opens in a new tab)No card. If a client's sector has no answer layer yet, we tell you before you pitch it.