Shopping habits are changing fast, and the data backing that up is hard to ignore. New research from Adobe shows AI tools like ChatGPT are no longer just helping people research before they buy, they’re increasingly sending shoppers who are already close to purchasing straight to retailer websites.
This data is specific to the US retail market, drawn from more than 1 trillion visits to US retail sites and a survey of more than 5,000 US shoppers. Adobe hasn’t published equivalent Australian figures, so treat the scale of these numbers as a US signal rather than a direct read on local behaviour. The trend itself is still worth paying attention to here.
AI traffic keeps climbing, and converting better
According to Vivek Pandya, Director of Adobe Digital Insights, AI traffic to US retail sites grew 71% year-on-year in September 2026, and Adobe expects that to spike further over the holidays. Visits arriving via AI converted into purchases 90% better than other traffic that month, generated 83% more revenue per visit, and saw shoppers spend 73% more time on site than other visitors.
Here’s the catch: getting any of that benefit depends on whether AI systems can actually read a retailer’s site. Adobe’s Citation Readability Score measures how much of a page’s content AI can effectively understand and cite. The average US retail homepage scored just 63%, meaning more than a third of typical content isn’t being picked up by AI at all. General merchandise homepages scored worst, at 40%, typically because they’re built around navigation menus and promotional banners rather than content that actually explains anything. Electronics retailers scored best, at 79%, helped by detailed specs, setup guides and troubleshooting pages, exactly the kind of explanatory content AI tools read well.
What government sites get right (and retailers can steal)
Here’s a genuinely useful, free blueprint hiding in plain sight: the Australian Government Style Manual, the writing and content standard used across government sites like business.gov.au, Services Australia and the ABS.
It wasn’t built with AI in mind, it predates the current AI boom. But its core rules happen to line up almost exactly with what makes content easy for AI systems to read and cite: plain language over jargon, one idea per paragraph, descriptive headings that state what the section actually covers, and content structured so it can be scanned rather than read top to bottom. Government sites also lean heavily on FAQ-style formatting, answering specific questions directly rather than burying the answer in marketing copy.
That’s not a coincidence. AI systems and time-poor human readers want the same thing: a direct answer, clearly labelled, without having to dig for it.
Five practical fixes
A few concrete things SME retailers can do, based on what both Adobe’s research and government content design point to:
Write real FAQ and buying-guide pages, not just product listings. Answer the specific questions a customer would actually type into an AI chat, delivery timeframes, sizing, compatibility, return policy, in plain sentences an AI system can lift directly.
Use descriptive subheadings, not clever ones. A heading like “Shipping and returns” gets read and cited. A heading like “The Fine Print” often doesn’t.
Keep product and policy pages in plain language. Short sentences, one idea per paragraph, no unexplained jargon, the same principle government style guides have used for years to make dense material scannable.
Make specs and policies explicit text, not images or PDFs. AI systems can’t reliably read specs buried inside a product image or a scanned PDF. If it matters, put it in real text on the page.
Check your actual return policy, shipping info and FAQs are current and easy to find. Adobe’s data suggests these unglamorous pages are exactly what AI systems lean on most when deciding what to recommend.
None of this requires a redesign or specialist AI tools. It’s largely the same discipline government agencies have applied to their own websites for years, just newly relevant because the audience reading that content now includes AI systems as well as people.