How we decide what to publish.

Denominators

A percentage without the population it was counted over is a decoration, not a finding. In our 300-business study, the headline 41% never named is counted over the 137 businesses that returned complete records from both engines, not over the 300. The study's key-findings table carries a "counted over" column for exactly this reason, and every figure in it names its own denominator.

The rule bites hardest on the numbers we would most like to quote loosely. When a figure has no honest denominator, it does not get published.

Raw data

Research is published with the file behind it. The State of AI Visibility in Australia 2026 ships its full 300 rows as data.csv and data.json under a CC BY 4.0 licence, so a sceptic, a journalist or a competitor can recompute every headline number without asking us for anything.

Two published engine scores and a published site readiness score reproduce the published overall score. We ran it across every row: 300 of 300 match exactly.

// recompute the overall score from published columns
Math.round(0.75 * mean(engine_scores) + 0.25 * site_readiness_score)
  === ai_visibility_score

Unpublished data

If a reader cannot open the source, we do not cite it. No internal test run, no private client scan and no unreleased sample is used as evidence on this site. Individual scan reports are unlisted and set to noindex by design, and a client's numbers do not become a statistic here unless they have agreed and the page says whose numbers they are.

The practical effect is that our published claims rest on a smaller pile of evidence than we actually hold. That is the trade, and it is the right way round.

Naming businesses

Businesses are named factually and never disparaged. "Was not named in our test questions" is a statement about our questions on one day, not about the quality of anyone's work, and the study says so in those words on its own page. Household names sit in the invisible column of that study, and they are listed without commentary.

Any business named in our research can write to us. The route and the commitment are on the corrections page.

Our own weak results

A result that argues against what we sell is published on the same page as everything else, at the same size. The study's own correlation between site readiness and per-engine visibility is a Pearson r of 0.116 across the 137 complete records, which is close to nothing, and site work is a thing we sell.

We also published our own score: 29 out of 100 on 31 July 2026, with site readiness four out of four and ChatGPT naming us in none of the five buyer questions it was asked.

Method before conclusion

The scoring formula is published in full before any number derived from it, including the parts that are not flattering: 70 per cent of an engine score is a count, and the remaining 30 per cent comes from two model-graded inputs that a third party cannot reproduce from the stored answers. That is stated at the technology page rather than left for someone to discover.

Gaps stay visible

Where nobody has measured something, the page says so rather than supplying a number. Where we hold a figure but not with enough confidence to stand behind it, we do not round it into a claim. An estimate presented as a measurement is the failure mode this whole site exists to argue against.

Who writes this

Pages are published by Answerable as an organisation. The firm is founded by Jackson Wilson, and the about page names him and the people who work on the business. Individual pages are not bylined to a person yet. That will change when there are public profiles to point at.