Machine availability: the step between being thought of and being bought
Asked three times which firms handle brand repositioning, ChatGPT never named Interbrand, Landor, Wolff Olins, Siegel+Gale, McKinsey or Bain. It named two small firms every time. Whether an assistant names you is its own availability, and fame does not reliably buy it.
You know the planning meeting. The 2027 plan has a line for brand and a line for distribution, and somewhere in the third hour someone asks where AI fits. The answer is usually that it sits inside brand: be famous enough and the assistants will know you.
That answer rests on two ideas that have organized brand growth for more than fifteen years. Mental availability asks whether a buyer thinks of you when the need arises. Physical availability asks whether they can find you and buy you when they look. Advertising builds the first. Distribution builds the second.
A third step now sits between them. A buyer hands the question to an assistant before acting: who should I hire, which one should I buy. The assistant does the remembering, and more and more it does the buying too. In September we found Walmart and Amazon had each used the word “agentic” in eight SEC filings since January 2025, while none of the twelve largest US packaged-goods makers had used it once (AI agents are learning to buy). Whether the assistant names you is the step between being thought of and being bought. Call it machine availability.
Machine availability sits between the two availabilities brand plans already fund, and neither of them guarantees it.
The decision belongs to the CMO, with the CFO holding the budget: does machine availability get its own owner and line in 2027, or is it assumed to follow from fame? The two answers lead to opposite plans. So we tested the assumption on our own category, B2B marketing advisory, with eight buyer questions put to ChatGPT between November 10 and 12, 2026. Each prediction was written down before the answers came back.
Fame did not get the famous firms named
Times ChatGPT named any of Interbrand, Landor, Wolff Olins, Siegel+Gale, McKinsey or Bain when asked three times which firms handle brand repositioning. Suits & Sandals and Zephyr Haus were named all three times, and seven of the nine firms it named recurred in every answer.
That is one question, and it is not a rule. Asked which firms do marketing mix modeling, one answer in three named the famous incumbents: NielsenIQ, Kantar, McKinsey, Bain. The other two named smaller specialists, and across the three answers only one firm, Analytic Partners, appeared every time. On some questions, fame is not how you get named. On others, the answer is a draw.
How the named firms got there
November 10, 2026On the marketing mix modeling answer whose sources could be read, five firms were named from eight cited sites. Only one, mma.com, was a named firm’s own site. The rest were a listicle, two directories, a press-release wire, Crunchbase, Wikipedia and a trade title. In a later asking, the model cited a shoe shop’s page as a source for a modeling firm that shares its name.
On this evidence, you get named through the lists other pages are built from, not through your own site.
How the buyer asks decides whether there is a list to be on
Asked the way a buyer would ask (“who advises a new CMO in their first 90 days?”), the model answered five of eight questions with job titles and named no firms. Asked “which consultancies…”, it named firms on five of five, 36 firms in all. A buyer who asks naturally may never see a vendor list at all. Our answer on whether wording changes which brand an AI names found one rewording took a brand’s own site from zero citations in three answers to first place in all three.
Then the answers moved. On November 12 we asked five questions three times each, in the same minute.
| Question | Firms named in all three askings | What it means |
|---|---|---|
| Budget allocation | 8 of 8 | A stable list. Being on it, or off it, is a real result |
| Category assessment | 6 of 8 | A stable core with a rotating edge |
| Board presentation | 3 of 11 | Three firms hold it; the rest rotate |
| New CMO | 2 of 9 | Two identical answers, then a different third |
| AI visibility | 0 of 19 | Two different answers to one question |
Source: CULT+MATH runs fd31033e and c613ae47, ChatGPT (gpt-4.1-mini), web search on, US, November 12, 2026. Each question asked three times, worded to ask for firms.
It also moved over days. The willingness-to-pay answer that named the same eight firms three times on November 10 kept six of them on November 12. Our answer to whether ChatGPT gives the same answer every time reaches the same place: ask once and you have measured once.
What we withdrew, and what we have not tested
Two of our early readings did not survive the second asking. We first said fame and machine availability were close to unrelated; that holds for brand repositioning and fails for marketing mix modeling. We first said the AI-visibility question had no incumbent firms; asked for firms, the model named eight brand-design agencies, and in one asking of three, small AI-visibility specialists. None of the category’s best-known vendors, Profound, Semrush or Athena HQ, was named in any asking.
The test that would settle whether machine availability is truly a third availability, rather than a reflection of the first, is registered and not yet run: a survey of real buyers compared with what the assistant names. Until then this rests on one budget-tier model, one category, US answers and at most three askings per question. Our retrieval study adds one more limit: on many questions the assistant searches nothing and answers from what it already holds, so the routes above apply only when it searches.
- Your buyers ask assistants vendor questions — you have seen it in sales calls or inbound, not assumed it.
- The questions produce a list — asked the way your buyers ask, the answer names firms rather than job titles.
- The list holds — the same question asked three times returns mostly the same names.
- You are not already on it — if you are named on a stable list, protect it; the budget belongs elsewhere.
- Not yours if your category’s answers rotate on every asking. Spending to be named in a draw buys a draw.
- CEO The six best-known firms were named 0 times in 3 on brand repositioning while smaller firms were named every time. That cuts against the belief that brand spend covers it. Give machine availability its own owner.
- CFO Answers held anywhere from 8 of 8 firms to 0 of 19, and one moved from 8 to 6 in two days. Against the urge to fund visibility now. Fund measurement first, on your buyers’ own questions.
- CRO Asked naturally, five of eight buyer questions returned job titles, not vendors. Against chasing every AI mention. Compete only on the questions that produce a list.
- CPO In the one readable answer, one named firm of five was cited from its own site. Against publishing more on the company site. Get into the directories, listings and entity records the answer is built from.
- Legal Assistants describe firms from pages the firms do not control, and one cited a shoe shop for a modeling firm. Check what those listings say about you as you would an ad claim.
- Risk One model, one category, one country, three askings at most, and a two-day drift. Report every result with its stability and its date.
Reconciled. The CEO and CRO want to act on an open position: fame is not the gate, so smaller firms can be named. The CFO and Risk want the measure settled first, because the answers move. The course that satisfies all six: give machine availability an owner and a line in 2027, spend the first quarter measuring your buyers’ own questions with repeated askings, and fund listing and directory work only on the questions that both produce a list and hold steady.
Put your buyers’ five most common questions to ChatGPT, three times each, before the 2027 budget locks.
Take the five questions your sales team hears most, worded the way buyers word them. Ask each three times in fresh sessions, then once more asking for firms. Record whether the answer names firms, which ones, and whether the names repeat. Fund listing and directory work only for the questions where a stable list exists and you are not on it.
- What it costs
- About two hours by hand for twenty askings, or a few cents each through an API. The real cost is holding visibility spend for a quarter while you measure.
- How you’ll know
- By the end of January you can say how many of the five questions produce a stable list. If fewer than two do, machine availability is not yet your budget problem in this category.
See which of your buyers’ questions hold a list before you pay to be on it.
The AI Visibility Decision Brief sets out the sourced, dated evidence a CMO needs for this call, with its limits stated. Or commission one built on your own buyers’ questions.
See the AI Visibility brief Or commission a custom brief →