How the monitor works

Measuring AI is hard because most of it is invisible. It sits inside email, office software and customer systems, and the people using it often don’t know it’s there. So this monitor doesn’t try to count all AI use. It tracks signals, and is open about where the signals run out.

Five areas, from access to impact

Each indicator belongs to one of five areas. They form a chain: each one is harder to measure than the one before, and matters more.

Access and readiness. Who could use AI: connectivity, skills, policies and licences.
Use. Who does use it, how often and how deeply.
Task mix. What it is used for, and whether it helps people work or replaces their work.
Outcomes. Time, quality, cost, service levels and new products.
System effects. Productivity, jobs and skills, who benefits and who is left behind, incidents and trust.

Most measurement today stops at the first two links and implies the last three.

What the shading means

Every indicator is tagged by what it would take to measure it.

Three measurement traps

Tokens measure compute, not value

The amount of text AI systems process is a useful signal of how intensively a tool is used inside one organisation. It is a poor proxy for value. Newer models and agents use far more tokens for the same task, units differ between vendors, and a target on token use invites people to burn tokens.

Surveys are good for who, weak for how much

Surveys show who uses AI, what they think of it, and use that no system logs. They are weak on frequency and time saved: people misremember, stigma hides some use, enthusiasm inflates the rest, and AI built into familiar software often goes unreported. Self-reported time savings tend to exceed measured ones.

Embedded AI is visible only to vendors

AI inside office software or customer systems shows up only in vendors’ admin dashboards, licence counts and their own aggregate reports. As AI becomes infrastructure, “AI use” stops being a useful category, and the better question becomes how tasks and outcomes are changing.

Why sources disagree

Three credible sources put the share of people using AI at 67%, 54% and 44%. ONS asks adults whether they have knowingly used AI in the past year. Ofcom asks whether people use AI tools. Microsoft estimates from device data how many people aged 15 to 64 used a generative AI product in a quarter. None is wrong; they measure different things. The monitor shows them side by side rather than picking one.

Updating

This is a prototype compiled in October 2026. The main official sources update on their own timetables: the ONS business survey every two weeks, with AI questions asked regularly; the ONS household survey monthly, with AI questions in some waves; productivity figures quarterly; and the AI sector study once a year.