How the AI Visibility Index works

The questions

Each category is measured on a fixed battery of around 150 questions, locked before scanning begins and run verbatim on every surface for the full window. The battery is built from the commercial questions with real query volume in that category, not questions a brand would write about itself. Because it stays locked between windows, a change in a brand's score is a change in the answers, not a change in the questions.

Saying what we ask is the point. We don't publish all three thousand questions, but every battery in every category is built from the same eight shapes:

1. Best in category: the open "who should I buy from" question
2. Segment: men's, women's, kids', plus-size, professional
3. Use case: the specific job the shopper is buying for
4. Gifting: buying for someone else, where brand recall matters most
5. Value: affordable, premium, and whether the price is justified
6. Sustainability and ethics: including locally made
7. Comparison against the category's dominant international brand
8. Where to buy: which store or retailer the shopper is sent to

Ask any of these in your own category and you'll see what we see: a shortlist of five or six names, and no page two.

The surfaces

ChatGPT, Perplexity and Google's AI Overviews, with location set to Australia. The same battery runs on all three on the same weekly cadence, so a brand's split across platforms is a like-for-like comparison rather than three different tests. Further surfaces join as their answer formats settle.

The population

Every category's brand population is frozen before scoring begins, so it cannot drift between windows. To be ranked, a brand must be Australian-founded or Australian-headquartered, sell direct to consumers from an active store, and have the category as its primary business. Every brand considered and left out is recorded against the rule that excluded it.

Some brands are tracked but unranked. They fail an inclusion rule and still keep turning up in answers: international majors, mass-retailer sub-lines. They stay in the dataset as context rather than being scored against local brands.

The score

The Visibility Score is an appearance rate: the share of all answers collected in the window that name the brand, from 0 to 100. A score of 25 means the brand was named in one answer in four. Scans run weekly and accumulate across the full window, so no single scan's answers can carry a brand, and a brand can't luck into a score on one good run.

Detection combines automated brand extraction with an alias match run directly against the answer text, because extractors miss spelling variants, ampersands and abbreviations. Every brand in the population is detected by the same method. There is no per-brand tuning. Detection is symmetric, or the scores mean nothing.

Average position is a secondary measure: where a brand lands inside an answer when it appears, taken from order of first mention. It separates brands with similar scores, and it is reported only for brands that actually appeared. A brand never named has no position, not a bad one.

Zero is a result

In every category we measure, some tracked brands are not named once in their scan window. These are real, trading businesses, with stores, staff, customers and orders going out the door, that are simply absent from the answers shoppers now get. We publish them beside the ranked brands rather than trimming the list, because a zero is the most useful number on the page: it is the only one that tells a brand its problem is being found at all, not where it sits.

What we publish

Every category page carries the full ranked ladder, not a top ten. Every brand in the population has its own page with its score, its rank, its split across the three platforms and what moves it. Each page states the scan window it reflects. Movement against the prior window appears from a category's second window onward.

What this doesn't measure

Sales, traffic, product quality or customer loyalty. A low score doesn't mean a bad brand. Plenty of excellent businesses score zero. It means AI assistants don't name that brand when shoppers ask. That is a different problem from being a worse product, and a more fixable one, which is the point of measuring it.

Independence and disclosure

No brand can pay to be included, excluded, ranked higher or removed. There is no sponsorship, no paid placement and no advertising anywhere in the Index. Answer capture runs on third-party monitoring infrastructure; the question batteries, the population rules and the scoring are ours.

Flux works with ecommerce brands, and some of them appear in this Index. Where a ranked brand is a Flux client at the time of publishing, we say so on that brand's page. Detection is symmetric: the same extraction and alias matching is applied to every brand in the population. A client brand is measured exactly like every other, and neither we nor they can move a score.

Cadence

Scans run weekly and accumulate into a scan window. Results are recomputed and republished at the close of each window, and every page shows the window it reflects. The battery stays locked between windows; any change to a battery or a population is recorded when it happens.