Do assistants recommend different kinds of supplier in different American sectors, and what that changes about who you hire

In our American data the answer is that the engines reach for the same kind of supplier in every sector, and that kind is software. Across 64 blind readings of ChatGPT and Gemini on 22 October 2026, every one of the 48 supplier namings was a product rather than a firm or a practitioner. The engines did not switch to agencies for the agency questions, to consultants for the strategy questions or to local firms for the local questions. They named software or they named nobody, and they named nobody in 38 of the 64 readings. The hiring consequence is specific and slightly counterintuitive, so it is worth stating plainly: if you want to be the thing an engine names in your sector, you need somebody who can produce the artefacts software companies produce, and most American sectors do not currently hire for that.

What the readings show, by the shape of the question

Question type in the bank	Supplier type named	Readings naming nobody
Choosing and vetting a supplier	Software, or nothing	A large share
Pricing and contracts	Mostly nothing, with ranges instead of names	A large share
Channel and platform questions	Software, consistently	Lower
Sector specific buyer questions	Software, with sector context around it	Moderate
Local and city questions	Software and directories, with local domains appearing only when a city was named	Moderate

The full blind set was 64 readings, 32 from ChatGPT and 32 from Gemini, producing 2,866 citation URLs across 1,143 distinct hosts. The United States was named inside the text of every question, because the data provider accepts gl, hl and location parameters and ignores all three, so the prompt is the only market lock that holds. Batch 33 names us in the question and measures a ceiling, so it is excluded throughout.

The 53 namings, in full

• Semrush, named in 13 readings
• HubSpot, 11
• Ahrefs, 8
• BrightLocal, 4
• Peec, 3
• Profound, SE Ranking and Otterly, 2 each
• Scrunch, Surfer and Conductor, 1 each
• American agencies, 0. Individual practitioners, 0. Local firms, 0.

That list is the whole of the American supplier namings. It is a software list with no services entry anywhere in it.

Why that happens, mechanically

The explanation is in the citation list rather than in any preference. The hosts the engines leaned on were documentation and reference material: developers.google.com in 17 readings, support.google.com in 15, searchengineland.com in 13, blog.google in 11, clutch.co in 11, semrush.com in 10, ftc.gov in 9, blog.hubspot.com in 9, with brightlocal.com and searchenginejournal.com at 7 and goodfirms.co at 6. One entry, ccn.web.tr at 8, is the provider's own fetch infrastructure rather than a publisher and is excluded from source analysis.

An engine composing an answer needs something it can attribute. Software companies publish prices, specifications, documentation and changelogs, all dated and all checkable. Services firms publish approaches and case studies they wrote themselves. Given that choice, the engine quotes the first. The sector does not change the mechanism, which is why the supplier type does not change by sector in our data.

The hiring consequence

• Hire or train somebody who can write a specification. What arrives, how many, in what format, by when, what is excluded. This is a documentation skill rather than a copywriting one and most American marketing teams do not have it.
• Put a price position in somebody's objectives. Getting a dated, defensible figure published is usually an internal negotiation rather than a writing task, so it needs an owner with standing.
• Make one person responsible for counting something. An original dated figure with a denominator is the most quotable asset a services business can produce and nobody owns it by default.
• Do not hire for assistant optimisation as a discipline. The experienced pool is tiny. Hire for judgement and documentation skill and train the specifics.

One contrast worth knowing before you generalise

We ran the same 198 question bank for Canada on the same day. There the engines named six Canadian agencies 21 times between them, with Major Tom named in 8 readings and Konstruct in 6, alongside 48 software namings. Same bank, same day, same engines, and a different answer on supplier type. So the finding above is a finding about the United States rather than about assistants, and the reason is the published evidence in each market. That is a reason for optimism rather than the opposite, because it shows the American result is not structural.

How to check your own sector

• Write twelve buying questions a buyer in your sector would ask, with the United States named in each.
• Put each to both engines, save the answers and the source lists, and record the date and the engine.
• Classify every named supplier as software, agency, practitioner or local firm. The mix is your answer.
• Count the readings that named nobody. Those are your openings and they are the cheapest to take.
• Report ChatGPT and Gemini separately. On our reading ChatGPT ran a live search on all 32 of its American readings and Gemini on 16 of its 32.

If you want to see how the five gates an engine applies before naming anybody actually work, they are set out in the five gates. If the question behind yours is why a competitor gets named and you do not, that is worked through in why your competitor is recommended and you are not.

Questions about supplier types in American assistant answers

Does my sector get different treatment?

Not in our American data, where software was the named supplier type across every question shape and no American services firm was named in 64 readings. Check your own sector with twelve questions rather than assuming, because our bank is broad rather than deep in any one vertical.

Should a services firm act more like a software company?

In what it publishes, yes. Prices, specifications, documentation and dated counted figures are the artefacts the engines quoted. That is a publishing choice rather than a change of business model.

Why do the readings name nobody so often?

Because the engine had nothing specific enough to attribute. Thirty eight of 64 readings named no supplier, which is the largest single outcome and the cheapest opening in the set.

Is this likely to change?

The Canadian result from the same bank on the same day, where six local agencies were named 21 times, shows the American outcome follows the available evidence rather than a fixed rule. The lever is what gets published.

Which asset moves it fastest?

A dated price position, because it is short, it is almost always missing, and it is directly quotable with attribution.

Classifying your own namings rather than guessing at them

The question of what kind of supplier an engine names in your sector is answerable only by reading the answers and classifying them, which is exactly what produced the 53 namings above. AI Knows Us reads ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews for the buying questions in a sector, records who was named and what each answer cited, reports the engines separately with their denominators, and keeps the raw answer text so a classification can be checked. In the United States that is the Growth plan at 299 dollars a month with 100 buyer questions watched, the cheapest tier that reads all five engines, read on aiknowsus.com/pricing on 22 October 2026.