AI search optimisation: how retail brands get recommended by ChatGPT, Gemini and Perplexity

Introduction

Australians spent AUD 82.6 billion online in 2025, with 24% of all retail spend now happening online, and in Australia Post's 2025 survey for that report, 32% of shoppers said they already use AI for shopping advice. Neither number tells you the thing you actually need to know, which is whether the engines those shoppers use can see your catalogue at all. For two of the three engines in the title of this guide, if you sell only into Australia, the answer today is: not through the front door.

That is a different problem from the one most AI search advice is written to solve. Australia Post tells retailers they need to show up in AI search, not just Google, without addressing whether they can. This guide covers how each engine sources a product recommendation, what our own visibility tracking on an Australian retailer showed, the five things worth doing first, and what nobody has published reliable answers to.

‍

What AI search optimisation actually means—and where it parts company with SEO

AI search optimisation — making your brand and products retrievable and citable inside an AI-generated answer, rather than rankable in a list of links — inherits most of its mechanics from SEO. Crawlability, clean markup and third-party credibility still matter. If your technical SEO is broken, this is not the project to start with.

Where it parts company is supply. Retail product recommendations reach these engines by two routes: a structured merchant feed you push to the engine's product graph, and the open web it crawls. Almost all competing content on this keyword addresses only the second. The first decides whether your SKUs are eligible to appear as products at all — as curated panels rather than links, in Shopify's vendor-published description of how AI Mode presents results — and it works differently at each engine.

‍

How each engine sources a retail product recommendation

Google—the feed you already maintain is the input

Google's AI surfaces do not build a separate product index. They read the Shopping Graph, the dataset your Merchant Center feed populates, which Google says holds more than 50 billion listings with 2 billion refreshed every hour. The same graph grounds the Gemini app, which Google describes as surfacing shoppable listings, comparison tables and prices from Shopping Graph data. When a shopper narrows a request, AI Mode runs what Google calls query fan-out — several simultaneous searches behind one answer, which is why a feed built for one keyword underperforms against a multi-part question.

The mechanism is well documented. The weighting is not: Google has not stated that Merchant Center data is a direct ranking signal in AI Mode, only that the graph grounds the answers. Treat feed completeness as a strong eligibility bet and a weak ranking claim.

ChatGPT—a pushed feed, a US shop window, and a crawler that still counts

OpenAI does not crawl your catalogue for product data. Merchants push a structured file against a published feed specification, delivered through the Agentic Commerce Protocol, which OpenAI has extended to carry feeds and promotions. Results are ranked on availability, price, quality and whether you are the maker or primary seller, and OpenAI states they are not ads and cannot be bought.

Two details matter most for an Australian operator. First, OpenAI's merchant page states that shopping in ChatGPT is currently live in the U.S. and will expand to additional regions over time. Second, and more usefully, the same page confirms a feed is not mandatory: feeds give greater control, but crawled pages remain a path in. If you sell through Shopify or Etsy, your catalogue is already integrated and there is nothing to apply for.

Perplexity—a citation engine first, a merchant program second

Perplexity's product cards are, in its own words, unsponsored recommendations rather than paid placements, and its free merchant program is pitched at large retailers sharing product specifications. But the experience launched on a Shopify integration covering businesses that sell and ship to the US, with expansion to new markets promised rather than dated.

That announcement is from November 2024, and we could not find a published feed specification, refresh cadence or selection rule from Perplexity since. Any guide listing field-level Perplexity feed requirements is extrapolating from Google Merchant Center. For an Australian retailer, the available lever here is not the feed. It is being a source worth citing.

‍

What our own tracking showed—and how much weight to put on it

This is our own visibility tracking, not published research, and it covers one retailer in one vertical. Treat it as illustration, not evidence: 60 prompts tracked daily across ChatGPT, Google AI Overviews and Perplexity for a multi-site Australian pharmacy retailer, producing roughly 10,000 answers in the three months to 12 August 2026. The prompt set skews to location and service questions, which shapes several findings below.

The same brand had very different visibility by engine. One of the retailer's two consumer domains was retrieved in 47.9% of AI Overview answers but only 18.7% of Perplexity answers. Its other domain appeared in 34.7% of ChatGPT answers. A single "AI visibility" score would have hidden a gap that size.

Its own site was not the most-retrieved source. A government one was. The federal health service healthdirect was retrieved in 52.1% of Perplexity answers, 39.5% of AI Overview answers and 37.2% of ChatGPT answers — ahead of every retailer in the set, client and competitors alike.

The pages retrieved were store pages, not product pages. Almost every retrieval from the retailer's own domains was an individual store location page. On a location-weighted prompt set that is partly an artefact of the questions asked, but it is worth sitting with: those pages are usually templated, thin and owned by nobody in particular.

Directory sites and foreign sources leaked in. ChatGPT and Perplexity drew heavily on Australian business directories, while AI Overviews barely touched them, leaning on Google's own properties and social profiles instead. Perplexity also retrieved the UK's NHS site in 7.8% of answers, and the US chain CVS in 7.8% — on a prompt set that is entirely Australian.

‍

The five things to do first

1. Treat the Merchant Center feed as AI infrastructure, not ads plumbing. It is the only route fully open to an Australian-only retailer today, and it feeds three surfaces at once. Prioritise GTIN coverage, availability accuracy and attribute completeness. If your feed syncs nightly against a graph refreshing two billion listings an hour, your availability is wrong for most of the day.

2. Apply to the feed programs anyway, and diarise the region check. Both OpenAI and Perplexity run waitlists and both have said expansion is coming. Applying costs an hour. Finding out eighteen months late that your category opened in Australia costs you a competitor's head start.

3. Promote your store pages to first-class assets. If you run stores, each store page is a product page for the store itself — hours, services, stock visibility, a real address and structured local markup. On our tracking, these were the pages engines actually pulled.

4. Instrument before you optimise. Track 30 to 60 real customer questions across the engines your customers use, and report per engine rather than as one average. Google is piloting native AI visibility reporting in Merchant Center; until it lands here, this is third-party tooling or nothing.

5. Do not fund this out of your search budget. IAB Australia's 2026 commerce and discovery survey of 1,079 Australians and 877 New Zealanders who shopped online in the past year found search remains the dominant discovery source, with AI complementing existing research habits rather than replacing them. That is one survey and adoption is moving quickly, but the evidence does not currently support a reallocation.

‍

What we still don't know

• No engine publishes how product results are ranked. OpenAI names four broad factors; Google confirms the graph grounds answers without confirming it as a ranking signal; Perplexity publishes nothing. Everything more specific is inference.
• Our tracking covers one retailer, one vertical and a location-weighted prompt set. Whether the store-page finding holds for a fashion or electronics catalogue is untested.
• We could not find reliable Australian data on what AI-referred traffic converts at, split by engine. The circulating figures are US retail aggregates, and Australian basket sizes and delivery economics differ enough that they should not be transferred directly.
• Whether enrolling in a feed program helps at all in a region where the shopping experience has not launched is undocumented.

‍

The bottom line

If you sell only into Australia, two of these three engines are closed to you at the feed layer, and pretending otherwise is how budgets get wasted. What is open is substantial: a Merchant Center feed that grounds every Google AI surface, a crawler path into ChatGPT that OpenAI confirms still works, and a citation path into Perplexity that rewards being worth quoting. Fix the feed, fix the store pages, measure per engine, and leave your search program funded. Then check the region question quarterly — it is the variable most likely to move first.

If store pages are what the engines pull, they have to carry live information: trading hours, services, and what is actually in stock at that location. That is an inventory and checkout data problem before it is a content problem. If you want to work through what your store pages would need to expose to hold up in an AI answer, we are happy to go through it with you.