AI evidence check · 8 figures · About three minutes · No sign-up - Guess the figure. Then read the source.
Most numbers quoted about AI at work arrive without the study behind them. Here are 8 we have published, each with its source. Put your own number down first, then see the real one, how it was produced and what it does not tell you.
Guess eight figures
Figure 1 of 8 · Copilot
Of employees who have a Microsoft 365 Copilot licence, what share use it regularly?
Your guesses stay in this page and are gone when you leave it. Nothing you enter is stored or sent.
How we read a figure - Four questions before a number goes on this site.
The same four apply to any statistic a vendor, a consultant or a headline puts in front of you, including ours.
01
Who was asked, or what was counted
A survey of 418 organisations with 20 or more staff says nothing direct about a firm of twelve. A figure for 47 countries is not an Australian figure.
02
Whether anyone held a stopwatch
People asked whether a tool saved them time nearly always say yes. Where the same work was timed against a group without the tool, some tasks were faster and some were slower and worse.
03
Who paid for it
A vendor-commissioned survey can be right. Read it as a second opinion and look for a source with nothing to sell.
04
What it would take to check it in your business
Usually two weeks of counting one repeatable job. That number is worth more than any of these, because it is about you.
The ledger · 22 figures · Every one linked to its source - Every figure in the posts, and how it was produced.
Of the 22 figures, 10 come from asking people, 8 from timing, counting or crawling, and 4 from a law, a list price or an official finding. Filter by method and the difference shows.
Showing 22 of 22 figures
- CopilotPeople were asked
35.8 %
of employees with a Microsoft 365 Copilot licence use it regularly.
Reached us second-hand, through Redress Compliance. First-year figures below 40 % are normal before anyone intervenes deliberately.
Source: Recon Analytics, January 2026
From the post: Two in three Copilot licences sit unused
- CopilotOn the record
AU $31.40
per user per month ex GST is about what Microsoft 365 Copilot Business lists at in Australia.
Source: Microsoft list pricing, July 2026, via Otto IT
From the post: Two in three Copilot licences sit unused
- MeasurementOn the record
13 minutes
a week moved from low-value to high-value work was enough for a mid-level officer to cover the licence in the Australian Treasury trial.
The evaluation could not say whether that happened. Nothing had been timed before the trial, so there was nothing to compare with after.
- MeasurementPeople were asked
26 minutes
a day saved per user is the headline of the UK Government trial of Copilot with about 20,000 civil servants.
Self-reported. The report says it was not possible to identify how the time saved was spent.
- MeasurementPeople were asked
54 %
of businesses using AI said it saved them time, in a survey of 1,019 businesses.
The chart is labelled "perceived" time savings, which is the honest word for it.
- MeasurementTimed, counted or crawled
14 %
more issues resolved per hour by customer support agents given an AI assistant, and 34 % more for novice staff.
The firm, with 5,179 agents, already counted issues per hour and rolled the tool out in stages, so there was a before, an after and a group without it.
- MeasurementTimed, counted or crawled
19 %
longer is what experienced software developers took with AI in a 2025 randomised trial, after expecting to be 24 % faster.
Afterwards they still believed they had been 20 % faster. The researchers have since said the result is out of date; the gap between the feeling and the clock is the part that travels.
- MeasurementTimed, counted or crawled
Over 7 minutes
faster on average for Copilot users writing PowerPoint slides than for non-users, with worse quality and accuracy.
The observed-task samples were small, and the department says so.
- MeasurementPeople were asked
Nearly 40 %
of medium and large Australian firms told the Reserve Bank their use of AI was minimal. Two-thirds had adopted it in some form.
Returns on investment were described as mixed, and few firms reported material productivity gains.
- GovernancePeople were asked
67 %
of employees whose organisation has banned generative AI report using AI in ways that break the rules, against 33 % where there is no policy.
48,340 people in 47 countries. Some of the gap is arithmetic: you cannot break a rule that does not exist.
- GovernancePeople were asked
48 %
of employees say they have uploaded company information, such as financial, sales or customer information, into a public AI tool.
The global figure. It counts anyone who has done it at all, including rarely, and there is no Australian breakdown in the published snapshot.
- GovernancePeople were asked
44 %
of Australian employees say they have used AI in ways that contravene policies and guidelines.
In the same snapshot, 57 % say they have relied on AI output without checking its accuracy.
- GovernancePeople were asked
56 %
of Australian workers using AI at work were using tools their employer had not provided or approved.
Commissioned by a vendor, so read it as a second opinion and not as the main evidence.
- GovernanceOn the record
Almost 13 %
of a Victorian department's workforce, nearly 900 staff, had visited the ChatGPT website in six months.
The Commissioner found the department had no controls aimed at the specific privacy risks of generative AI tools.
- GovernancePeople were asked
25 %
of organisations have clearly designated roles with responsibility for the responsible use of AI.
418 organisations with 20 or more employees, so it says nothing direct about the smallest firms.
- GovernancePeople were asked
29 %
of organisations had implemented the relevant practices, while 78 % agreed with statements about ethical AI performance.
- GovernanceOn the record
10 December 2026
is the date from which the Privacy Act requires a privacy policy to say where a computer program makes, or materially helps make, a decision that significantly affects someone.
A decision can be in scope even where the output is subject to human review, and a tool supplied by a vendor is still yours to disclose.
- AI visibilityTimed, counted or crawled
No measurable link
between publishing an llms.txt file and how often a site is cited by AI assistants, in an analysis of roughly 300,000 domains.
We publish one anyway. It costs twenty minutes, it is accurate, and it is not a growth lever.
- AI visibilityTimed, counted or crawled
10.13 %
of domains have an llms.txt file, so nearly nine in ten sites do not.
- AI visibilityTimed, counted or crawled
0.32 %
of all website traffic comes from AI platforms, up from 0.24 % in 2025 and 0.02 % in 2024.
Small, and growing fast. People arriving from an assistant have usually done their comparison already.
- AI visibilityTimed, counted or crawled
74.8 %
of AI referral traffic comes from ChatGPT. Gemini is next at 11.6 %.
Its share is falling even as its absolute traffic grows, because the rest of the market is growing faster.
- AI visibilityTimed, counted or crawled
Under 4 %
is the share of US ChatGPT prompts that cite a source in professional services, the least-cited category. Travel and hospitality is at about 23 %.
US prompts, May 2026, reported second-hand.
Each figure is repeated from the post named under it, where it sits in context with the rest of the evidence. Where a figure could not be traced to a source it is not on the site. This page is general information, not advice about your business.
Free diagnostics · About three minutes each · No sign-up
These are other people's numbers. Get yours.
Each diagnostic scores one gap from your own figures, in your browser, and names the next step.
Want the whole picture, not one figure at a time?
Nine questions, three minutes, nothing leaves your browser. The scorecard shows which of the five steps you are on and which service fits.
Want your own numbers measured, not estimated? Book a call.
Twenty minutes. We tell you which of your figures is worth measuring first, what it costs to measure, and whether you need us at all.
+61 490 083 850 · Wollongong, NSW