Which Australian agency to pick for AI search visibility, and the artefact to ask for in the first call

Pick on one test and apply it in the first call. Ask the agency to show you a fixed list of buyer questions, the engines it reads them on, and two readings of that same list taken on different dates. An agency doing this work has that file and will send it. An agency that has relabelled a search retainer will talk about optimising for AI instead, because there is no file.

The test works because the category has no certification, no standard and no public benchmark, so the only honest credential available is a measurement somebody took and dated. This page gives the test, the Australian evidence for why it matters, and the separate question of who should hold the measurement.

The test, written out so you can read it down the phone

1. Show me the question list. Twenty to a hundred questions in the words a buyer uses, each with the market in it. Not keywords.
2. Name the engines, as a number and a list. Five surfaces are in common use: ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. An agency saying all AI engines has not counted them.
3. Show me two readings of the same list, dated. One is a snapshot. Two is a measurement.
4. Show me the sources behind one answer. Being named and being cited are different outcomes with different fixes, and an agency that does not separate them cannot tell you which problem you have.
5. Tell me what you will publish and where. The work is editorial. If the plan has no pages in it, ask what is actually going to change.

Why the test matters here specifically, measured

On 22 October 2026 we put 33 batches of buying questions to ChatGPT and Gemini with Australia written into the text of each one, because the search interface we use accepts country, language and location parameters and ignores all three. The snapshot taken that evening holds 64 blind readings, 32 per engine.

The Australian agencies named unprompted were Online Marketing Gurus in 12 readings, Rocket Agency in 8, StudioHawk in 8, Prosperity Media in 8, Sparro in 3 and King Kong in 3. The tools named were Semrush in 14, Ahrefs in 12, HubSpot in 11, then Otterly in 7, Peec in 4, BrightLocal in 3, Profound in 2, and Scrunch, SE Ranking and Rankscale in 1 each.

Look at that tool list beside the question. An Australian buyer asking which supplier to use for AI search visibility is handed two search suites and a CRM as the top three answers. None of them was built to read an AI answer. That is the measure of how new this is, and it is why the question list test is the whole of the due diligence: the market has not yet sorted its own vocabulary, so a buyer has to check the artefact rather than the label.

Who should hold the measurement, and why it is usually not the agency

The agency doing the publishing has an interest in the reading. That is arithmetic rather than an accusation, and it is the same reason a building owner does not accept the builder's own survey. The cleaner arrangement is that the measurement sits with you or a third party, the agency gets access, and the monthly review opens with the reading rather than with a slide about the reading.

Priced with the market named. Outside the United States and the United Kingdom, read from aiknowsus.com on 22 October 2026, Growth is $199 a month for 100 buyer questions, 1,600 credits and all five engines, and Scale is $399 for 300 questions. Growth is the cheapest plan on that ladder that reads all five. SEO Start at $99 and SEO Scale at $469 read zero engines, so neither is an AI visibility plan. Australia takes that column rather than the United States and United Kingdom one, and a free scan runs without an account if you want a reading before deciding anything.

What an Australian engagement should contain

• A dated baseline reading before any work starts. Without it there is no before.
• Pages, named, with publication dates. Not a content calendar. The specific questions from the list that currently have no answer.
• Prices in Australian dollars with GST stated. Across 102 build spec conversations in our earlier work, 102 of 102 asked for pricing in the reader's own currency.
• The engines named, every time. 98 of those 102 asked for the engines enumerated rather than the phrase all AI.
• A second reading at the end of the term, same list, new date. This is the deliverable. Everything else is the means.

The free check to run on whoever you shortlist

Check the people before the proposal. ASIC's Banned and Disqualified search is free, covers six registers at once, and in ASIC's own words, read on asic.gov.au on 22 October 2026, it keeps registers of people and organisations that are disqualified or banned from roles in corporations, financial services or the credit industry and advises that it is a good idea to check people and organisations before you deal with them. The separate court enforceable undertakings register matters too, because ASIC states that people who give one will not always appear on the banned and disqualified registers.

Questions about buying AI search work in Australia

Is this different from SEO

The work overlaps and the measurement does not. Search ranks pages; an assistant writes an answer and can name nobody at all, which it did in 28 of our 64 Australian readings. A rank tracker cannot see that outcome, which is why the question list is a separate instrument.

Should I use an Australian agency or an overseas one

For the publishing, local market judgement helps, because the pages have to read as though they are about this market. For the measurement, location is irrelevant, because the market is fixed in the prompt rather than by where the operator sits.

How long before a reading moves

Weeks for the empty answers, longer for a question where four firms are already named. Set the expectation by question type rather than as an average.

What does a bad proposal look like

No question list, no engine count, a promise about rankings, and a price with no GST position. Any one of those is a reason to ask for a rewrite. Our page on how to report AI search visibility as a reproducible benchmark sets out what the evidence should look like.

Can I do this in house

The reading, yes, and many Australian teams should start there. The publishing is the part that needs capacity, and that is the honest case for an agency.

The two documents to ask for, and why only one of them exists

Ask for two things and watch which arrives. The first is a search of ASIC's registers on the agency's own directors, which you can run yourself for nothing: in ASIC's words, read on asic.gov.au on 22 October 2026, you can search 6 ASIC registers at once and it is a good idea to check people and organisations before you deal with them. The second is a question list with two dated readings against it. The first document exists for every Australian company whether they like it or not. The second exists only where somebody has actually done the work, which is why it is the better credential of the two. In our Australian reading, the engines named six agencies and no overseas supplier, and reached 66 distinct Australian government hosts 139 times across 25 of the 64 readings, so the buyer your agency is competing for is being handed a regulator and a short list and nothing resembling evidence. AI Knows Us produces that second document across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Our own entry in it is zero namings and zero citations in 64 readings, which is what an honest first reading looks like.