For marketing leaders: what AI search is doing to your category¶
This page is the short version. It has no jargon and only the handful of numbers the argument needs. If you want the full technical treatment, the rest of this manual is free and there is no form anywhere on it.
Your buyers are asking a machine first¶
A meaningful share of the research that used to happen on a results page now happens inside a generated answer. Someone types a question about your category, and an AI product replies with a short list of vendors and a paragraph about each. That reply is the shortlist. For many buyers it is the only shortlist they will see before they start booking calls.
The important thing about that reply is how short it is. In our published research across 860 scored AI answers, the average answer named about two vendors. Not ten. Two.
Which means most companies are simply missing¶
When there are only two places, most companies get none. Across 85 B2B software companies we measured, 35% were never named once in ten questions about their own category. Not ranked low. Named zero times.
That is the first thing to understand about this channel: it is not a leaderboard where you are somewhere in the middle. It is a door you are either through or not through. Being absent is the normal condition, not the exceptional one.
There are two doors, not one¶
This is the part that changes what you should do, and almost nobody separates them.
The first door is whether the engine thinks you are one of the companies in your category at all. When a buyer asks "who are the best tools for X", the engine assembles a handful of candidates before it writes a word. If you are not among the candidates, nothing else matters. In our data, companies that were never named lost mostly these questions: broad ones, that do not mention a competitor by name, that simply ask who exists.
The second door is whether the engine prefers you when it is comparing you with someone specific. Companies that were named in most answers had cleared the first door. What they still lost was almost entirely head-to-head questions of the form "you versus them".
These need different work. Getting through the first door is about whether credible information about you exists in enough places for a machine to conclude that you belong in the category. Getting through the second is about whether there is something specific and checkable it can quote about you against a named rival. If you are stuck at the first door and someone sells you work aimed at the second, you will pay for content very few people will ever see.
Why the number you were shown may not mean what you think¶
If an agency or a tool has already given you an AI visibility score, three things are worth knowing before you act on it.
It probably covers one AI product, not all of them. These systems disagree with each other much more than people expect. Of 32 companies we tested on more than one engine, 14 were visible on some and invisible on others. Almost half would get a different verdict depending on which one happened to be measured. A single number labelled "AI search" is usually a number about one product.
It was probably measured once. Ask the same AI product the identical question twice, minutes apart, and it agrees with only about half of its own previous list. That is not a fault, it is how these systems work. But it means a single measurement is one draw from a range, and the difference between your score in March and your score in June can easily be the tool rather than your company.
On some questions there was no answer at all. Google does not generate an AI overview for every question. When it does not, there is nothing to appear in. If those questions were quietly dropped from the calculation, the percentage you were shown is bigger than it should be.
None of this means measurement is pointless. It means a number without its engine, its date, its questions and its number of runs cannot be interpreted, and most numbers in this market arrive without them.
What to do about it, in order¶
First, find out which door you are at. You do not need to buy anything to do this. Write down the ten questions a real buyer in your category would type. Put each one into an AI product with web search on. Write down only whether your company was named. Then look at the questions where you were not named and sort them into two piles: broad questions about who exists in the category, and questions naming you against a specific competitor.
Whichever pile is bigger tells you which door you are standing at. That is the whole diagnostic and it costs you an afternoon.
Second, be sceptical of anything sold before that diagnosis. A proposal that recommends the same programme regardless of which pile is bigger has not looked at your situation. Ask a supplier which door your data says you are at, and ask to see the questions that support the answer.
Third, ask what will exist when the work is finished. Not how many pages will be published, but what credible information about your company will exist in places you do not own. That is what the first door responds to, and it is the question most proposals answer least clearly.
Fourth, insist on being able to check the work later. Fix the question set on day one and never change it. Require that each question is asked more than once every time it is measured. Ask for the raw answers, not just a dashboard. All three are cheap at the start and impossible to add afterwards.
An honest note about who wrote this¶
Broadcastwell ran the research this page is based on and sells services in the category it measures. That is a real conflict and the only useful response to it is to publish everything, which we have: all the data, all the analysis code, and the questions. If you think a number here is wrong, you can recompute it and publish what you get.
We also measure ourselves on the same terms. In our own most recent check, across four AI products and ten questions about our own category, we were named once in forty answers. The one mention was an engine quoting our research while recommending other agencies. We publish that because a firm asking you to demand evidence should be willing to be measured by it.
Where to go next¶
If you are evaluating suppliers or tools, Chapter 12: How to buy GEO without getting sold a number is the practical companion to this page. It has twelve questions to put to any supplier, a description of what a defensible deliverable looks like, and a section on when the right decision is to hire nobody at all.
If you want the underlying research, the manual itself is free, ungated and has no email capture anywhere on it. It is one chapter per page and every number in it traces back to a published dataset you can download.
About this manual¶
Author. Sairam Sivakumar, Broadcastwell.
Operator disclosure. Broadcastwell ran this measurement and sells services in the category it measures. Broadcastwell is excluded from the measured sample and from every ranking. The mitigation is not that the conflict is absent, it is that the raw data and the code are public and the result can be recomputed by anyone who disagrees.
Self-audit, July 2026. In the four-engine, five-run self-audit published alongside Volume II in July 2026, Broadcastwell was named in 0 of 200 answers and cited 0 times among 663 citations. That published figure stands with its date and is never replaced.
Self-audit, 18 August 2026. Re-measured on the same ten published questions across the same four engines, at one run per question rather than five, between 03:51 and 04:03 UTC on 18 August 2026: named in 1 of 40 answers, and cited once among 674 citations. The single naming and the single citation are the same answer, in which the engine quoted Broadcastwell's own published visibility page as a source. The two lines are not directly comparable, because one rests on five runs per question and the other on one.
Not peer reviewed. This is an independent industry study published as an open dataset with the analysis code that produced every figure in it. It has not been through academic peer review. Read it as measurement, and check the measurement. If you disagree with a number here, recompute it from the public data and publish what you get.
Licence. Prose and figures CC BY 4.0. Site code MIT.
The three volumes.
| Volume | What it covers | DOI |
|---|---|---|
| Volume I | 85 companies, 61 categories, 860 scored answers and 5,160 citations, one engine held constant. The dataset README additionally records 1,753 unique domains cited | 10.5281/zenodo.21537014 |
| Volume II | The Absence Ladder. All 616 absence records classified by question shape | 10.5281/zenodo.21586091 |
| Volume III | Cross-engine divergence. 280 questions, 40 categories, four engines, 853 answers | 10.5281/zenodo.21789120 |
Data and analysis code for all three volumes: github.com/Broadcastwell/state-of-geo-2026.
Version 1.0, August 2026.