Ask most brand teams what AI is doing to ecommerce and you get one of two answers. Either it changes nothing, or it changes everything. Our 2026 Marketplace Consumer Report, based on a survey of 1,000 Australian online shoppers, is actually pointing somewhere more useful than either.
Only 3% of Australians say they begin a product search on ChatGPT or a similar AI tool. Google is still where more than half of all product journeys start, at 54%, and it grew YoY.
However, there is a different side to the story as well. Almost a third of consumers say that they rely on AI more than they did a year ago when discovering products online. Approximately 26% used a search engine AI summary (like Google AI Overviews) while researching or buying something in the past 12 months. Another 23% used ChatGPT or a similar LLM for the same job. So, while AI may not be where the journey starts, it is increasingly where the journey gets narrowed.
That distinction matters more than it sounds. Discovery is the part of the funnel brands have always spent marketing budget to influence. The shortlist is the part they have never really been able to see.
The people using AI are the people you want
Among 25-34 year olds, 39% used an AI-assistant to research a product and 36% used AI search summaries. Among people over 65, those numbers fall to 6% and 14%. Household income splits it the same way. In homes earning $160,000 to $199,000, 42% used an AI assistant to research a purchase. In homes under $40,000, it was 11%.
So AI-assisted shopping is currently concentrated in younger, higher-income, metro households. If you sell anything considered, anything with a price tag that invites a second opinion, that cohort is disproportionately your growth.
Trust is genuinely split, and that is the opportunity
45% of Australians say they trust AI-generated summaries and recommendations when deciding what to buy, while 44% they do not.
A near-even split like that usually means the behaviour is still being formed. People are testing whether the answer is any good, and the thing they test it against is everything else they can see, including reviews, the brand’s own D2C website, the product listing. Which is why the brands that come out of this well will not be the ones that mastered a prompt, but the ones whose product information is consistent, specific and verifiable wherever a model looks.
What the models are reading
Google’s own data shows queries are getting longer and more conversational. Around 70% of keywords are now estimated to be long-tail, and conversational queries of more than eight words are up 800% year on year. Google’s AI Mode and AI Overviews rollout has pushed AI summaries into roughly 39% of searches.
What this means is that when someone asks a nine-word question about which moisturiser suits sensitive skin under makeup, the model does not answer from a display ad. It answers from structured, plain-language content that addresses the question directly. Most brand content in Australia is still written for a two-word keyword.
Experts from Pattern have expressed their views on why this changes the mechanics of organic in the new rules of search and how AI has changed discovery, and there is a practical piece on why long-form copy is now the fast track into the AI-first web.
Marketplaces still decide
Here is the part that gets missed in most AI commentary. Discovery is fragmenting, but conversion is consolidating.
60% of Australians who see a product on social go to Google next. 41% go straight to Amazon. And 48% say they have visited a brand’s own website after first finding it on Amazon, up 18% YoY. The AI answer, the social post and the search result are all feeding the same few destinations where the decision actually happens.
Which is why the risk of getting this wrong is not abstract. When Australians go to Amazon looking for a specific brand and can’t find it, 14% now buy a comparable product from a brand they have never heard of. Absence does not pause demand. It hands it to whoever is present.
What to actually do about it
Three things, in order.
- Write your product content so a model can answer a question with it. That means the specific claim, the specific use case, the specific comparison, in prose a person would say out loud. If your listing copy and your site copy contradict each other, you have given the model a reason to cite someone else. We have explained what this means in our blog on an LLM access audit covering the technical side of whether AI systems can read your site at all. Get your complimentary LLM Access Audit now.
- Then check whether you are even eligible to be recommended. Reviews, ratings, availability and accurate imagery are the trust signals AI systems lean on because they are the trust signals shoppers lean on. Thin, inconsistent or unmanaged listings fail both audiences at once.
- Treat the marketplace as the close, not the channel. Social and AI create consideration. Amazon and the retailer sites convert it. If those two halves are not carrying the same story, you are paying to create demand that someone else collects.
None of this requires a new AI strategy. It requires the content strategy you already have to be good enough that a machine can quote it.
If you want the full picture of how Australians are shopping in 2026, including the category-by-category data behind these numbers, download our 2026 Marketplace Consumer Report here or get in touch and we’ll walk you through what it means for your category.
