AI assistants and buying agents can now propose suppliers before anyone visits a website. Shortlist Share measures how often they propose you, and why they leave you out.
AI agents shortlist B2B vendors by matching what they can read about each supplier against the constraints in a requirement, such as application, certification and region. We call the share of repeated runs in which a company is proposed its Shortlist Share, and each missing appearance points at a constraint the agent could not confirm.
AI agents shortlist B2B vendors by matching what they can read about each supplier against the requirement they were given, either from the web at the moment of the request or from what their model learned in training. A buyer who asks an assistant for suppliers of machine vision systems for food packaging lines, with a named certification and a support team in Europe, gets a short list back in seconds, and the companies on it are the ones whose public information made the match easy. A supplier missing from that list never enters the comparison at all.
We call the measure of this Shortlist Share. We give an assistant or a buying agent a real requirement, written the way a procurement team would write it, ask it to propose suppliers, repeat the run several times, and count the share of runs in which the company appears. If a supplier appears in 6 of 20 runs, its Shortlist Share for that requirement is 30 percent.
The share moves because the agent's answer depends on what it can find and confirm about each supplier against each part of the requirement. A requirement is a list of constraints, such as an application, a certification, a region, an integration or a price range, and an agent can only propose a supplier with confidence when each constraint is stated plainly somewhere it can read. A company that lists its certifications on a public page, names the applications it serves and the systems it integrates with, and is described the same way by third parties gives the agent fewer reasons to leave it out. A company whose website describes what it does in general terms forces the agent to guess, and an agent that cannot confirm a constraint has every reason to drop that supplier.
Measuring it starts with the requirements rather than with a tool. Write five to ten requirements the way your real buyers would write them, each with the constraints that matter in your market, and keep them fixed so the results can be compared over time. Run each one several times in each assistant your buyers use, such as ChatGPT, Claude, Gemini and Perplexity, because answers vary between runs and a single run says little. Record which suppliers appear, in what order and with which constraint the agent cites for each, then compute your share per requirement and per assistant, and repeat the same set every month.
Take a sensing company with eight requirements, each run five times in four assistants, which gives 160 runs. If the company appears in 40 of them, its overall Shortlist Share is 25 percent, and the view per requirement says more than the total: a share of 70 percent for warehouse requirements and zero for anything mentioning a medical certification tells the team which page or which third-party listing is missing. These numbers illustrate the method and describe no real company.
The measure has limits we state plainly. Assistants change their models and their sources without notice, so a drop in one month may come from the assistant rather than from anything the company did, which is why the trend over several months counts for more than one reading. Runs are samples, so a share from a handful of runs carries wide uncertainty and should always be reported with its run count. And Shortlist Share measures being proposed rather than being chosen, since the human buyer still compares the list and the close still happens between people.
For a marketing leader, the share turns a vague worry about AI search into a list of fixable gaps, because each missing appearance points at a constraint the agent could not confirm. It sits next to two related measures we use: Machine Recall, whether an assistant names you in open category questions, and Answer Fidelity, whether it describes you correctly once your name is in the question. This work belongs to the Visibility practice of Torke Service, and a team that prefers to run it in-house can do so from its own assistant with Torke Engine.
A shortlist built by an agent is a comparison you were either part of or never heard about, and measuring the share is the only way to know which. The same method applies to any channel where a machine filters options for a person, from procurement platforms to marketplace search. Write down three requirements your best customers would have typed a year ago, run each one five times in two assistants this week, and count how often your name comes back.
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.
The ones your buyers use. For most B2B markets that means at least ChatGPT, Gemini, Claude and Perplexity, each run several times, because answers vary between runs and between assistants.
One common reason is that it could not confirm a constraint in the requirement, such as a certification, an application or a region, from a source it could read. Stating each one plainly on public pages gives it fewer reasons to drop you.
Enough to see a pattern rather than one answer. We start with at least five runs per requirement in each assistant, and report the run count next to the share so a small sample is never read as a firm result.
It means you entered the comparison. The human buyer still compares the suppliers on the list, and the evaluation and the close still happen between people.
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.