How the curve is built.

News, product launches and posts I have read and sorted by what they describe. This is the part that moves the line.

What I hear every week on enterprise calls. No names, no company details, ever. It shapes the descriptions of each stage, it does not move the line.

64 named companies placed from things anyone can read: product pages, press releases, court filings, job titles and reporting. They are the dots on the curve. They illustrate each stage, they do not move the line.

1. 1. Every signal I approve gets sorted into one of five buckets by the words it uses: talking about it, trying it, running it, measuring it, governing it.
2. 2. The number on the curve is the share of those signals sitting in the last three buckets. That share is the reading, and it is what moves week to week.
3. 3. A week is locked once it is finished, then averaged over the last four weeks, so the line moves weekly and not daily.
4. 4. Limits: it measures published language, not company behaviour. Work that is never published does not appear, sorting is done on wording, and any one company can sit two groups away in either direction.

Every line is a reading of published evidence on the date I checked. It is a statement about what a company has made public, and nothing else. Private work, internal programmes and client delivery are invisible here, so a placement is never a judgement of capability.

Innovators

Creating or testing genuinely new AI search capability before an established market practice exists.

Early adopters

Identifiable operational commitment beyond isolated experimentation.

Early majority

The capability is repeatable, operational and part of normal business. Dated public evidence shows the same AI search capability operating repeatedly across products, teams, customers or reporting cycles, rather than as a single launch, experiment or commitment. Evidence of ownership, cadence, standardisation or repeated delivery strengthens the placement but is not individually required.

Late majority

Active implementation has begun, after the practice was becoming established elsewhere.

Laggards

Public evidence shows activity remains primarily observational, planning or policy led.

Evidence describes what a company has done in public. It is an input to the reading, never a shortcut to a stage. Two companies can carry the same evidence type and sit in different places, because the question the stage answers is how established the behaviour has become inside the organisation.

Infrastructure

Shipped standards, protocols, datasets or market mechanisms that other companies then have to work with.

Measurement

A published method, metric or dataset for reading AI search performance that someone else could run.

Product or service

A named product or service in market, with a page anyone can read.

Structured access

An MCP server, a public API or a documented data feed.

Customer adoption

Named customers, case studies or reported usage of the above.

Dedicated hiring

Roles created and filled specifically for this work.

Budget

A stated spend, filing line or funding round tied to the work.

Pilot

A beta, trial or limited release, described as in progress.

Governance

Crawler policy, licensing terms, litigation or published rules of access.

Commentary

Posts, talks, research reads and opinion, with nothing shipped alongside.

Confidence is set by the sourcing, not by how sure I feel. It is derived mechanically from how many independent sources exist, whether any of them come from the company itself, and whether the earliest artefact carries a date.

High confidence

Multiple independent dated sources, at least one of them first party.

Medium confidence

Strong first party evidence, or two or more credible secondary sources.

Low confidence

Limited public evidence. Interpretation required.

Position on this curve is a reading of published evidence, not a judgement of the company. Pace is a choice, and a slower one is often the right call for the demand a business actually has.

Plenty of serious work never reaches a public page, so a company low on this curve may be well ahead of what it shows. Missing evidence is not evidence of nothing. If I have the reading wrong, send me the proof and I will move the dot.