What we can't promise, and why nobody else should either
Nobody controls what these systems say. Anyone guaranteeing you a place in an AI answer is either lying or doesn't understand how they work.
synthesis · May 21, 2026 · 3 min read
This is the article that costs us business, and we would rather publish it than have the conversation go badly in month three.
We cannot promise that an AI will name you. Not for any amount of money, not in any timeframe. Here is precisely why, in four parts, and each one is measurable.
One: the systems are not deterministic. Ask the same model the same question twice and you often get a different list of companies. Across 2,961 runs, researchers found under a one-in-a-hundred chance of two identical brand lists. Even with the model’s randomness turned down to zero, repeated runs changed 9 to 28% of decisions. There is no stable position to guarantee, because there is no stable position.
Two: nobody outside these companies controls the inputs. A model’s answer is assembled from sources it selects, weighted by processes none of us can see, changing — in the case of AI Overviews — roughly every two days. We can improve what it finds about you. We cannot decide what it does with it.
Three: the field has no demonstrated causal chain to revenue. A survey of 45 studies published between 2023 and 2026 concluded: “No reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability or downstream behavior.” That is the state of the evidence. Anyone quoting you a revenue multiple from AI visibility work is extrapolating past the end of what has been measured.
Four: it is competitive. A peer-reviewed benchmark found that the effectiveness of these techniques declines as more competitors adopt them. Whatever advantage exists today compresses as your category catches up. That is not a reason to skip it — it is a reason the timing matters, and a reason to distrust anyone selling permanence.
So what can be promised?
That the work gets done. Every item on your list, delivered, documented and verifiable. If we can’t complete one, it comes off the invoice.
That you will be able to see it. The measurement switched on and the baseline set in the first week, so that six months from now there is a before to compare against.
That the obstacles on your side are removed. If a crawler is blocked, it gets unblocked. If your pages can’t be quoted, they get rewritten so they can. If your profile is wrong, it gets fixed. Those are all things we control, and they are the entire scope of what any honest vendor in this market controls.
And that we’ll tell you when something didn’t work. Which, given the four points above, will sometimes happen.
There is also a legal dimension, and it is worth knowing because it explains a lot of what you’re being sold.
In the United States, an objective performance claim requires substantiation before it is made — competent and reliable evidence, held in advance, generalisable to the population being marketed to. The FTC’s 2025 order against an AI company that advertised a “98% accuracy” figure turned on exactly this: tested against the actual marketed population, the real figure was 53%.
Which means “guaranteed to appear in AI answers” is not merely optimistic. It is an unsubstantiated objective claim about a system the seller does not control, in the category the FTC has been actively policing since Operation AI Comply in September 2024. And a refund offer does not cure it — the Commission’s position is that a money-back guarantee doesn’t license an unsubstantiated representation.
So when a competitor guarantees you a position, you are looking at one of two things: a company that hasn’t read the rules, or one that has and is betting nobody checks.
The version of this we hold ourselves to, in one line: we will tell you what we measured, what we changed, and what moved — including when nothing did.
If that isn’t worth paying for, don’t pay for it. We would genuinely rather you didn’t than have you expect something we can’t deliver.
