The share of American buyers who ask an assistant before contacting a supplier, and what is honestly measured about it

There is no one number, and a supplier who quotes one is almost always quoting a survey that measured something adjacent. The figures in circulation differ on three things at once: whether the question asked about shopping, research or choosing a supplier, whether the sample was consumers or business buyers, and whether an AI answer inside a search results page counted as using an assistant. Those three choices move the answer by tens of points. So the useful question for an American business is not what the national share is. It is what the engines answer when somebody asks about your category, and that is measurable this week with a denominator you control.

Why the published shares disagree, and what to check before quoting one

Choice the study made	Why it moves the number	What to check
Shopping against supplier selection	Consumer product research is far more common than business supplier selection	The exact question wording, not the headline
Consumers against business buyers	Different populations with different tools and different approval steps	Who was sampled, and how they were recruited
Whether an AI answer in search counts	Including Google's AI answers raises the figure substantially over standalone assistant use	Whether AI Overviews was in the definition
Self report against observed behaviour	People under report routine tool use and over report novel tool use	Whether anything was observed or only asked
Recency	This behaviour is changing fast enough that a figure from two quarters ago is a different market	The field dates, not the publication date

None of that makes the published work useless. It makes the headline unusable as a plan input, which is a different thing. A figure with the question wording, the sample and the field dates attached is worth reading. A figure quoted as a percentage with none of those is a number somebody will quote back at you.

What we measured instead, and what it establishes

On 22 October 2026 we put a 198 question American buyer bank to the ChatGPT and Gemini product surfaces through a data provider, with the United States stated inside the text of every question. The provider accepts gl, hl and location parameters and ignores all three, so the prompt is the only market lock that works. Batch 33 names us in the question and therefore measures a ceiling rather than a market, so it is excluded from everything below. That leaves 64 blind American readings, 32 from ChatGPT and 32 from Gemini.

• The engines produced 2,866 citation URLs across 1,143 distinct hosts.
• They named us in 0 readings and cited our domain in 0 readings.
• They named a supplier that was not us in 24 readings, with 53 namings in total.
• They named no supplier at all in 38 readings, the largest single outcome in the American set.
• Every one of the 53 namings was a software product. Semrush 14, HubSpot 12, Ahrefs 9, BrightLocal 5, Peec 3, Profound 2, SE Ranking 2, Otterly 2, Scrunch and Conductor 1 each and Surfer 2. No American agency was named once.

That establishes something a share figure cannot. It says what an American buyer who does ask receives, which is the part a supplier can act on. A 20 per cent share and a 60 per cent share lead to the same action if the answer those buyers get names nobody in your category.

The engine split, and why it matters to the question of share

ChatGPT ran a live web search on all 32 of its American readings. Gemini ran one on 16 of its 32. So even the question of whether an assistant answer reflects the current web has two different answers depending on which assistant, and any single national share statistic flattens that. The two engines are reported separately here and never averaged.

The measurement to take in one afternoon

• Write the twelve questions an American buyer in your category would actually ask out loud, with the United States or the state named in the text.
• Put each to ChatGPT and to Gemini. Save the full answer and the source list. Record the date and the engine.
• Score each reading three ways. Named you, named a rival, named nobody. That is the scorecard, and it needs no share estimate at all.
• Count the repeat hosts across the source lists. Those hosts are where the answer is coming from and they are your target list.
• Repeat in thirty days on the same twelve questions, and report ChatGPT and Gemini in separate columns.

The city level version of this behaves differently again, which matters if your buyers are local, and the four metros we measured are in two of the four winning sources are the city's own. If you want to see the source mix behind a much larger single engine run before building your own, it is counted in the source mix behind 306 readings.

Questions about assistant use before an American purchase

So what percentage of American buyers use an assistant first?

We did not measure that and will not publish a figure we did not take. What we measured is what the engines answer when asked an American buying question, 64 blind readings on 22 October 2026, and that is the part a supplier can act on.

Can I use a published industry figure in a board paper?

Use it with the question wording, the sample and the field dates beside it, and label it as what it measured rather than as assistant use in general. A figure presented without those three is the one that gets challenged.

Does AI Overviews count as using an assistant?

That is exactly the definitional choice that makes the published numbers disagree. Decide for your own reporting whether it counts, write the decision down, and keep it constant between readings.

Is my own twelve question reading statistically valid?

It is a census of the questions you chose rather than a sample of a population, which makes it directly useful for your decisions and not suitable for a claim about the market. Keep the question set constant and the comparison between months is sound.

Which engines should I include?

Start with the two that answer most readily and add the others as your reporting matures. Five surfaces are sold in this category: ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, and reading two of them is a real measurement provided you say which two.

The number worth having instead of a national share

A share statistic tells an American business how many buyers might ask. A reading tells it what those buyers are told, and on 22 October 2026 across 64 American readings that was a rival in 24 cases, nobody in 38, and us in none. AI Knows Us reads ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews for the buying questions in a category, reports each engine separately with its denominator and keeps the raw answer text so the figure can be audited. In the United States the cheapest tier that reads all five engines is Growth at 299 dollars a month, read on aiknowsus.com/pricing on 22 October 2026. SEO Start at 149 dollars for the United States and the United Kingdom reads zero engines and is not an AI visibility plan.