How is AI availability measured?

Around a thousand recorded answers per category from ChatGPT, Gemini and Claude, across 8 to 16 buying situations with several phrasings each, in memory mode and with live search, reported as share of answers with 95% confidence intervals. The agent half is a ten-check review of the brand's public pages, summarised as four levels.

AI availabilityDefinition v1.0September 2026Kesterley

AI availability

AI availability is how easily a brand gets chosen when an AI assistant answers a buyer's question, and how easily an AI agent can read, verify and buy from that brand. It is the third availability, after the mental and physical availability that marketing science uses to explain how brands grow. This page is the reference definition we measure against. Plain text for machine readers: ai-availability.md.

The definition

AI availability has two halves. They are measured differently, they move at different speeds, and a brand can be strong in one and absent in the other, so we never add them into one number.

Two halves

Reported separatelyNever averaged

Answer availability

The mental half. Whether the brand comes up when assistants answer the buying questions of its category, and how often. Measured as the brand's share of recorded answers across the category's buying situations, in two modes: what the model already knows (memory) and what it finds when it searches the web (search).

Agent availability

The physical half. Whether an AI agent can find the brand's pages, read them without a browser or a login, trust the facts on them, and follow a path to purchase. Measured as a ten-check review of the public surface, summarised as four levels from Falling behind to Agent-ready.

Share of answers

For one buying situation: the share of recorded answers that include the brand, with a 95% confidence interval. Across situations: weighted by how important each situation is in the category.

AI penetration

The share of buying situations in which the brand appears at all, above the noise floor.

Network size

How many distinct buying situations the brand is attached to, weighted by their importance.

AI mental market share

AI penetration multiplied by network size, as a share of the category. The machine analog of mental market share in the Ehrenberg-Bass tradition, where mental market share equals mental penetration times network size.

Agent-readiness level

Checks passed out of ten, summarised as level 1 Falling behind, 2 The basics only, 3 AI-readable, 4 Agent-ready.

Where it comes from

The Ehrenberg-Bass Institute's research established that brands grow through two things: mental availability, being thought of in the situations where buyers enter a category, and physical availability, being easy to find and buy. Its work on category entry points gives us the sampling frame: the buying situations a category is entered through, phrased the way real buyers phrase them.

AI assistants have added a layer between those two. They are asked the buying question, they answer from memory or from a live search, and increasingly they buy on the buyer's behalf. AI availability is the availability framework applied to that layer: the same logic, a new respondent. It is our own extension of published research, not an endorsement by its authors.

Primary sources for the framework: Sharp, B. (2010), How Brands Grow, Oxford University Press. Romaniuk, J. and Sharp, B. (2016), How Brands Grow Part 2. Romaniuk, J. (2023), Better Brand Health.

How it is measured

The machine is the respondent. No surveys. We put the buying questions to ChatGPT, Gemini and Claude directly, around a thousand times per category, through their official developer interfaces.
Buying situations, not random prompts. 8 to 16 situations per category, several phrasings each, in memory mode and with live search, reported separately.
Statistics that admit noise. Every share ships with its confidence interval, every report with its own reliability and minimum detectable change. When a difference is within noise, the report says so.
Pre-registered designs. For published reports the design is frozen before the first question is asked, so nothing is selected after seeing the data.
The agent half from evidence. Crawler access, entity records, structured data and its freshness, readability without JavaScript or login, a plain path to purchase. Checked live, not self-reported.

The complete measurement rules, versioned, are in the methodology.

What it is not

Not rank tracking

Rank order inside an AI answer is close to random between runs. Membership in the answer is the construct that persists, so we report inclusion, never position.

Not share of voice from a prompt list

A handful of prompts someone picked, run once, produces a precise-looking number with no meaning. Availability needs a sampling frame, repeats and intervals.

Not a sales promise

AI availability is a leading indicator. The evidence that it drives sales is early, and we say exactly how strong it is rather than promising an outcome.

Not guaranteed

Nobody can guarantee a place in a stochastic system. We measure, we say what to fix, and we measure again.

Questions

Is AI availability the same as AI visibility?

No. AI visibility tools count how often a brand appears in a set of prompts someone chose. AI availability is sampled from the buying situations of a category, reported with confidence intervals and stability, and includes the agent half: whether an AI agent can read and buy from the brand at all.

How fast does it change?

The search half moves within weeks when the sources assistants read change. The memory half moves over quarters, because it depends on what the models learned in training. Agent availability changes as soon as the pages do.

Can it be bought?

Paid placements inside assistants are beginning to appear. They are a separate line. We measure organic availability, and where paid placement exists in a category we report it separately.

Who should care?

Any category that is bought after comparison: software, services, travel, finance, durable goods. If your buyers ask an assistant before they ask you, AI availability decides whether you are in the answer.

Cite as: Kesterley (2026). AI availability, definition v1.0. kesterley.com/ai-availability. The definition is versioned; changes are listed here with their date.

Read the methodology