How the Index was measured

We measured 100 Australian businesses, ten in each of 10 industries. The point of the selection rules is that we did not hand-pick anybody: a business is in the study because a buyer looking for its category would find it.

1. The five product industries, which are Ecommerce, Supermarkets, Big Retail, Banks and FMCG, were picked by market position from public rankings. These are the brands Australians actually buy from, in the order the market puts them.
2. The five service industries, which are Healthcare, Real Estate, Builders, Manufacturers and Lawyers, were picked by who actually ranks on Google for the category’s own search terms, with directories and aggregators stripped out. If a business earns the category’s search traffic, it is in.
3. Three national brands failed that ranking test. We kept them in the study and said so, because a brand missing from its own category’s search results is a finding, not a sampling error, and quietly swapping them out would have tidied away the very thing the Index measures.

The full register, meaning every business we measured, is published at the bottom of this page and on each industry’s own page.

We wrote 473 distinct questions of five kinds. Every one is a question a real buyer would type, not a keyword dressed up as a sentence, and every one was put to all three engines.

The five kinds are not equal in what they reveal. Cost and category questions are where buyers actually choose, and they are where service businesses go missing. Brand and comparison questions ask about a business by name, which is why almost every industry scores well on them. The gap between those two groups is the story most of the Index tells.

Three engines, identical conditions, one pinned model each, all inside August 2026. Pinned means the same model version answered every question, so no difference between answers comes from the engine changing underneath us.

What each engine ran on, and how it responded.

Engine	Model or capture route	Answers given	No AI answer
Google AI Overviews	via DataForSEO, geo-coded AU	459	41 of 500
ChatGPT	gpt-5.5-2026-04-23	500	0 of 500
Perplexity	sonar	500	0 of 500

Every answer was captured word for word, alongside its date, the model that produced it and every source it linked. Nothing was paraphrased at capture time, because the scoring depends on what the engine actually said, not on what a note-taker remembered it saying.

Every answer was scored the same four ways, against written definitions.

1. Named. Does the answer mention the business by name? This is the number the Index table leads with, because a business the engines never mention is invisible no matter what else is true.
2. Linked. Does the answer include a link to the business’s own website? Named without linked means the engine is describing the business from what other people have written about it.
3. Accurate. Does the answer get its facts about the business right? The mistaken identity cases in the Index, like the bank that was reviewed as a Pixar film, were caught by this check.
4. Displaced. Does the answer send the buyer to somebody else instead? An answer can mention a business and still recommend its competitor, and that difference matters to the business more than the mention does.

An engine can also give no AI answer at all, showing its regular results instead. Of 1,500 business and question pairs, 1,459 came back with an answer we could score and 41 did not. Every one of the 41 came from Google AI Overviews. All percentages on these pages are rounded to whole numbers, once, at the end.

A methodology that only lists its strengths is a brochure. These are the limits we know about, stated plainly.

1. It is a snapshot, not a time series. Every answer was captured in August 2026, and the engines change constantly. The next edition is planned for roughly six months later, on the same businesses and the same method, which is what will make movement measurable.
2. It covers three engines. ChatGPT, Perplexity and Google AI Overviews are measured; Gemini is on the list for a future edition.
3. It measures visibility, not quality. The Index says whether the engines mention a business and whether its own site is the source. It says nothing about whether the business is any good at its work.
4. No figure was adjusted after scoring. The dataset ships as one versioned file in version control, and every number on these pages is computed from it, so the numbers cannot be quietly retyped and any future change arrives as a new dataset with a changelog entry.

The complete list, grouped by industry. Each industry name links to that industry’s own page, where these businesses’ numbers are broken down by engine and by kind of question.