Which brands AI assistants put forward when US buyers ask, 1,053 namings in 635 readings

Being written about in an answer, being one of the sources behind it, and being put forward as a choice are three separate events. Most of this category reports the second and talks as if it were the third. This page counts the third one, and prints the denominator next to it.

The run

On 22 October 2026 we scored 635 readings. Two engines, ChatGPT and Gemini, read through their product surfaces. 324 US buyer questions, grouped into 19 batches. 550 of the 635 readings ran a live web search and the remainder answered from what the model already held. No question in the first 18 batches names our own company, because a question containing a brand produces that brand and cannot measure anything.

A naming is counted once per reading in which a name appears, never once per occurrence. An answer that mentions one vendor four times counts once. That rule, rather than the data, decides the total, and two vendors quoting naming figures under different rules are not comparable.

The result

• 635 readings scored.
• 1,053 namings, spread across a roster of 24 distinct vendor names.
• 451 readings named a rival and did not name us.
• 184 readings named no company at all.
• Our own name appeared in 0 readings. Our own domain appeared in 0 of the 19,793 citation URLs collected in the same run.

The 184 is the number most businesses have never had put to them, and it changes how the rest should be read. More than one reading in four produced guidance with no company in it. For those questions nobody won, no competitor beat you, and no amount of publishing would have changed the outcome, because the question was not a naming question. Any visibility percentage calculated against all 635 understates every brand in the set, including the leaders.

The roster, by readings in which the name appeared

Out of 635 readings, 22 October 2026, two engines, US questions.

• Semrush 259
• Profound 154
• Ahrefs 145
• Otterly 124
• Peec 91
• HubSpot 42
• Schema App 40
• BrightLocal 37
• Yext 34
• SE Ranking 23
• Scrunch 23
• Surfer 21
• Conductor 15
• Writesonic 12
• Athena 10
• Frase 6
• Goodie 4
• Clearscope 3
• Evertune 3
• AgencyAnalytics 2
• Brandlight 2
• Rankscale 1
• Brandwatch 1
• Moz 1

Two things are visible in that list and neither is about marketing spend. The top of it is held by established tool brands with very large published content estates, and the specialists who appear high, Profound at 154, Otterly at 124 and Peec at 91, are the same specialists whose own domains appear high in the citation count from the same run. Namings and the source list move together, which is the single most useful thing in the table for anybody deciding where to put work.

Why your name is missing, in the order the gates actually fail

The mechanism is covered properly in how AI assistants choose which brands to recommend, and the short version is that absence is decided before any ranking happens. The question has to call for a name at all, which 184 of our readings did not. The system has to be able to reach and retrieve something about you. More than one independent source has to say the same thing. Your claim has to be specific enough to repeat. Only then does anything that resembles ordering happen.

The third of those explains more absences than the rest combined, and it is the one a content programme cannot fix by itself. We have the measurement to prove that on ourselves, which is the next section.

Our own null result, and why it is in this page

Between 20 and 22 October 2026 we went from 134 recommendation-shaped pages live to 641. The naming count for us over that period went from 0 to 0. On 20 October, 1,260 readings produced 6 in which our domain was cited and 0 in which we were named. On 22 October, 635 readings produced 0 cited and 0 named. A separate earlier pass on 15 and 16 October, 597 readings on a narrower question bank, had already put us at 3 cited and 0 named.

Cited and named are two findings. They are reported separately here and never added, averaged or compared, because a page can be read by an engine and still never be put forward as a choice, and the difference between those two states is the entire problem this category exists to solve. The citation side of the same measurement is in brand citation rates in AI answers, and the earlier null result is written up in what did not work.

Publishing 507 more pages moved nothing. We sell visibility measurement and that number is the opposite of a sales argument, which is exactly why it belongs on the page.

What would actually prove an intervention works

Our passes are not a controlled study and this page does not present them as one. The question banks differ between passes, the model versions moved underneath us, and there is no comparison group. When we asked the assistants what a credible version would require, the design they described is this.

• At least 200 neutral unbranded US buyer prompts across 20 or more categories.
• Five named surfaces, reported separately, with at least three baseline observations per prompt and surface in a repeatability subset.
• Published definitions of recommendation share, first-position share, top-five share, mention rate and owned-domain citation rate, fixed before data collection.
• A treated cohort of domains and a matched comparison cohort that receives no intervention over the same period.
• At least 30 days of follow-up, better at 60 and 90, with publication dates, indexing status, content changes, third-party changes and model version changes all logged separately.
• The effect reported as the change in the treated group minus the change in the comparison group, in percentage points, with confidence intervals and the full count of successful and failed observations.

That number does not exist yet, from us or from anybody we can find who publishes the observations. A vendor quoting a visibility lift with no comparison group has measured the market and the model updates along with its own work, and has no way to separate them.

What to do with this if your name is not on the list

First, classify your own questions before you measure anything. Run each one and record whether any company was named at all. Questions that never produce a name are not visibility failures and should be scored in a separate bucket.

Second, work on corroboration rather than volume. The brands at the top of the roster are described by other people in many places, and that is the gate our own 507 pages could not pass by existing.

Third, hold the question still. Identical prompts do not return identical answers, so three observations per question beats one, and a single before and a single after is not a result.

Where our interest lies, and where a rival beats us

We sell the measurement described on this page, so every number here comes from a party with a motive. We read five surfaces, not nine. Our pricing page, 119,203 bytes, renders no SSO, no SAML, no SOC 2, no ISO, no DPA and no white label, checked by word boundary over the whole document, and a loose search for sso on it matches crossOrigin and subprocessors rather than the feature. Read 22 October 2026, tryprofound.com/pricing publishes SSO and SAML with SOC 2 compliance and up to 9 answer engines on its Enterprise plan, with no numeric price attached. On engine count and on security paperwork, they win.

Our own published US prices are SEO Start 149, Growth 299, Scale 599, SEO Scale 699, Inbound 1099, Outbound 1299, Agency 1299 and Everything 1799 dollars a month. SEO Start reads 0 questions, 0 credits and 0 engines, so nobody should read 149 as the price of AI visibility. Growth at 299 is the cheapest plan that reads all five surfaces. The same page says the Agency plan covers 10 client brands and also answers No to more than one business in all eight columns of its comparison table, and both of those render today.

Questions about naming rates in AI answers

Is a 0 naming result normal for a smaller brand

It is common. 451 of our 635 readings named a rival and not us, and we are the party running the measurement.

Does a naming count tell me my market share

No. The roster above ranks by readings in which a name appeared, which tracks how much is published and corroborated about a vendor, not how many customers it has.

Why not add the citations and the namings together

They have different denominators and they answer different questions. One says a page of yours was read, the other says your company was put forward. Adding them produces a number that cannot be interpreted.

How many repeats of a question are enough

Three per question per surface is the working minimum for a subset, because the same prompt does not return the same brands twice.