Three numbers first.
This client has been in its local market for 15 years. Its monthly digital marketing budget runs into six figures, and has done for years. Over that time it has been through several agencies.
Trackable customers from the web: zero.
Not few. Zero.
Every agency he hired was busy hitting its own metrics
Each one had a handsome report ready: impressions climbing, clicks climbing, keyword rankings moving up. The data was real. Nobody faked anything.
The problem is that not one of those numbers was for him.
Rankings need to move? Buy backlinks. Metrics need to look good? Pick a few long-tail keywords nobody actually searches and take the number-one spot. Report needs to clear review? Draw a prettier curve. These are the standard moves in that industry, and every one of them serves Google’s algorithm — not one of them serves his customers.
In all those years, not a single agency ever asked him: how many people walked into your shop this month because of these numbers?
Because that was never their KPI. Their KPI was whether the monthly report cleared review.
So the only move left was to switch agencies. He switched, and switched again. Each new agency arrived with a fresh dashboard, the numbers looked just as good, and the customers stayed exactly as absent.
By the third switch this should have been obvious: the problem was never their execution. It was that they were never standing on the same side as his business. They sell metrics. He needs customers. Those two things stopped being the same thing the day he signed.
So if you have already changed agencies three times and nothing moved, the fourth will not move it either. What needs replacing is not the vendor. It is the whole way you are buying.
Same managed model — a different thing being managed
One thing to be clear about first: the agencies he used before were also managed services. None of them handed him a to-do list and told him to do it himself; they did the work. He was never short of people to execute.
The trouble is what they managed was metrics. You pay, they keep your rankings, traffic and impressions in good shape, and they hand you a report at month end. The work got done — but it was aimed at Google’s algorithm.
What we manage is the result. You still pay and we still do the work, but what we watch is not rankings — it is whether anyone walked into your shop because of it. That difference sounds small. It decides how each of the four things below actually gets done.
What we did after we took over
We do four things:
One — take the site apart and find out why AI cannot see him. A 12-dimension deep diagnosis: structured data, E-E-A-T signals, whether AI crawlers can even get in, whether the content is written in a shape an engine can quote. This is not running a tool and showing him a score. It is identifying the specific reasons he does not exist in an engine’s eyes.
Two — do the work ourselves. Engineering and content, both done by our people. He does not hire an engineer, does not build a content team, does not read a line of code. That part is the same as his old agencies — the difference is direction: every change we make is there to get AI to say his name when a customer asks, not to nudge some ranking up one spot.
Three — verify it on our own backtesting platform. We put the questions his actual prospects ask to ChatGPT, Gemini, Claude, Perplexity, DeepSeek and Meta AI at the same time, and check whether his name comes back in each answer. Not once — on a schedule, continuously, because AI answers drift and a single result proves nothing.
Four — when the engines change, we change with them. How AI picks its sources keeps moving. This work has no end date, so we do not have a delivery date either.
Not one of those four is something his previous agencies sold him. And not one of them is something he could finish on his own.
Month one: inbound customers up 109%
In our first month, his inbound customer count grew by 109% — it more than doubled.
One thing needs saying plainly, because it is easy to misread: that 109% is the growth in total inbound customers. It is not the growth of web-sourced customers. Web-sourced customers started at zero, and zero cannot grow by 109%.
What actually happened: the extra customers — a whole additional set — came entirely from the new channel that had been built from nothing: AI recommendation. It brought in roughly as many people as all his existing channels combined, so from the moment it went live it was his single largest acquisition channel.
For a 15-year-old brand, doubling the customer count is hard — not because doing more is hard, but because its existing channels had long since hit their ceiling. What opened up was a road six figures a month had never managed to buy.
And this was not a one-month firework. We have been running his GEO for several months now, and the web has been his largest and steadiest channel throughout.
Today, more than half his customers come from AI
His prospects go to ChatGPT, Perplexity and other mainstream AI engines, describe their situation in their own words, and ask who they should turn to. His name comes back in the answer. They follow it to his door.
More than half his customers now arrive that way.
That number did not come from us. He keeps it himself: every person who walks in gets asked “how did you hear about us?”, and it goes into the record, one customer at a time. So this is not our marketing claim. It is his ledger.
Compare it with what he used to pay for. SEO ranks you near the top of a results page; paid ads spend money to push your name in front of people. Both sell exposure — more eyes on you. Whether those eyes trust you, or come to you, is the user’s call, and no amount of budget buys it.
AI recommendation sells trust. The user asks who to turn to, and the engine says your name — that is not exposure, it is an endorsement. Exposure you still have to convert; an endorsement is already converted before the person walks in.
So the people who walk into his shop know what they want. The reason this channel went from zero to his largest source is that it does not send strangers interrupted by an ad — it sends people who have already been recommended, who have already made up their minds.
Whether you have the same problem, you can find out today
Open any mainstream AI engine, ask the question your customer would ask, and see whether you appear in the answer.
If you do not, you do not — and the budget you spend every month is not buying you that position.
To check it properly, the method is here: Screenshots lie, backtests don’t.
Then take this question to your agency: how many customers came from the web last month?
This is not something you can do yourself
Run an audit, ask an engine a few times — those are worth doing and you can start today. But these three, no company carries alone:
Continuous backtesting across engines and across time. Asking once proves nothing. Each engine picks sources differently, and the answers drift. What you need is a measurement system that keeps running, not one curious afternoon.
Content and engineering moving together. Structured data, E-E-A-T, crawler accessibility, rewriting the content — done separately they cancel each other out. Done together, the engines start recognising you. That is rarely a skill set one person holds, and that person is expensive to hire and harder to keep.
Changing when the rules change. How AI engines select sources keeps shifting. What you got right last month may not be enough this month.
Which leaves two roads: have someone run it for you, or don’t do it. The middle road — “we’ll learn it ourselves, slowly” — is a fantasy.
What to do next
If this sounds like your company — budget in place, agencies changed, reports gorgeous, no customers arriving from the web — there are two concrete things to do:
Today: run your site through the free audit on geoweb.tw and see where you land across the 12 dimensions. It costs nothing, and you walk away with a report you can take to any agency and use as a cross-examination.
Next: write to [email protected], or book a demo. We will run a backtest first and show you exactly how the mainstream AI engines talk about you right now. Once you can see it, we can talk about working together.
Where these numbers come from: they were compiled by the client himself — he asks every customer who walks in how they found him, and writes it down. This is his operating data, not our estimate. The 109% is the change in inbound customers during the first month of managed GEO; “more than half from AI search” is the standing position to date, not a single-month spike. The client’s identity is de-identified under NDA.