Most teams check AI answers by typing their own company name. What decides a buyer's shortlist is whether an assistant names you when the buyer doesn't.
Ask ten unbranded questions a buyer would ask before knowing any vendor, in ChatGPT, Claude, Gemini and Perplexity, three times each, and count the answers that name your company. Torke calls that share Machine Recall, the AI version of unaided brand recall, and it shows whether a buyer who starts with an assistant ever hears your name.
Ask most marketing teams whether ChatGPT recommends their company and they will type the company name into it, read a reasonable description and conclude that it does. That check measures something real, but it can't tell you whether an assistant would bring you up for a buyer who has never heard of you, which is the question that decides whether you reach a shortlist at all. A buyer asking which companies make inspection systems for semiconductor fabs doesn't put your name in the question, so what counts is whether the assistant puts it in the answer.
We call this Machine Recall: the share of unbranded category questions in which an AI assistant names your company without being prompted with your name. It is the machine version of unaided brand recall, the survey measure in which people name the brands they know in a category before being shown a list. Its partner is Answer Fidelity, which checks what the assistant says once your name is already in the question, and the two answer different problems, since a company can be described perfectly when asked about and still never come up when it isn't.
An assistant names a company in an unbranded answer when its training data, or the pages it retrieves at the moment of the question, connect that company to the problem in the words the buyer used. That connection is built from comparison articles, directories, conference programmes, trade press and your own pages, when they state plainly what you do and for whom. When those sources describe you in product language and the buyer asks in problem language, the connection is weak, and the assistant reaches for the vendors whose names appear next to the buyer's words most often, which in a small category is often two or three companies.
The check fits in an afternoon. Write ten questions a member of your buying group would ask before knowing any vendor, in their words rather than yours, such as which suppliers can add vision inspection to an existing line or what to look for in a perception system for forklifts. Ask each question in ChatGPT, Claude, Gemini and Perplexity from a logged-out session or a fresh account, so earlier conversations don't steer the answer, and run each question three times, because answers vary between runs. Record every company named in every answer, then divide the answers that name you by the total.
Take ten questions, four assistants and three runs each, which makes 120 answers. If your company appears in 18 of them, your Machine Recall is 15 percent for that question set on that date. The number says little on its own, so record next to it which competitors appeared and how often, because a competitor named in 70 of the 120 answers shows you how the assistants currently describe the category and which sources you need to appear in to change that.
The check has limits worth stating. Answers change when models are updated and when the pages they retrieve change, so one afternoon gives you a reading for that date and nothing more, and the repeatable version, with a question set that stays the same from month to month, is what we describe as AI Retrievability. Ten questions is also a small sample, so a difference of two or three answers between two checks is noise. And in a category that has no settled name yet, a low score often means the buyers' words connect to nobody so far, which leaves the vocabulary open for whoever supplies it first.
For a marketing leader, Machine Recall changes what goes on the content plan. If the assistants name competitors and never you, another page on your own site describing your product rarely closes the gap, because the assistants match the buyer's problem language against many sources at once. The work is to describe what you do in the buyer's words on the pages they and the assistants read, and to be named in the third-party sources the assistants already cite for your category, which is the work of the Visibility practice in Torke Service.
Recall in AI answers behaves the way recall in people's heads always did, since the company that comes to mind first gets the first call, and the same check works in any category where buyers ask before they search. Write your ten questions this week in the words your last three new customers used in their first email, run them across four assistants, and count how often you appear without being named, because that count tells you whether a buyer who starts with an assistant will ever hear of you.
Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is.
Usually because the sources the assistant draws on connect your competitors to the buyer's problem in the buyer's own words, through comparison articles, directories and trade press, while your pages describe you in product language the buyer doesn't use.
Use a logged-out session or a fresh account. Memory and earlier conversations can steer an assistant towards companies you discussed before, so the check would measure your own history instead of what a new buyer sees.
Ten questions asked three times in four assistants gives 120 answers, enough to see which companies dominate a category. A difference of a few answers between two checks is too small to read as a change.
Tell us what your team sells, who buys it and where deals stop moving. You'll get a written reply saying which way of working fits you, or whether a Diagnostic makes sense, and if neither does, the reason.