Make your store easy for customers' agents to understand
Your next customer may send software to research you. No verified figure shows how much Australian shopping agents handle today, but the work that helps them (clear product data, honest policies, a path to a person) helps customers and search.
Real and arriving, but unproven here.
In Build for customers and their agents, we argued that your next customer may send software to research your business before they visit it themselves. This article is the practical follow-up for established Australian merchants: what to fix first, and what to leave until later.
The honest position is that agents are real and arriving, while their volume in Australia is still unproven. We don't know of a public, verified measure of how much Australian shopping is done or referred by agents. So the case for acting doesn't rest on a forecast. It rests on the fact that most of this work also helps the people who shop with you, and the search engines they use.
Start with the foundations. Then add bounded actions.
What the named agents are.
Two agents come up often in conversations with merchants. We describe only what public documentation says.
Muse is Meta's personal AI agent. CNBC reported on 3 October 2026 that it can search retail sites, surface products and prepare a purchase for the user's approval. It shows an approval card with the order details before checkout. It can pay through Meta's payment partners, such as Link by Stripe and Shopify's Shop Pay, which the user must connect. PayPal was announced but isn't yet available. Meta announced retail partnerships, including Walmart, Gap, Sephora, Expedia, Wayfair and Best Buy, as an "official fast lane", though CNBC reports that not all connectors are active yet. A Meta spokesperson said shopping has become one of Muse's biggest usage drivers, with no figure given. [1]
That is US reporting. None of the sources we've seen confirm whether Muse is available to Australian users, whether its checkout works for Australian shoppers, or whether Link and Shop Pay payments through Muse are supported in Australia. Treat Muse as an example of where agentic shopping is heading, not as a service Australian merchants can plan around today.
Instinct is a personal AI assistant that researches products and websites on its user's behalf and prepares purchases, but only checks out with the user's approval.
We say nothing here about how any agent ranks, chooses or reads sites, because the public sources don't say.
The numbers, and what they measure.
Several figures circulate, and they measure different things. Here they are, Australian first.
Australia Post's Annual eCommerce Report 2026 (March 2026) found that Australians spent a record $82.6 billion online in 2025, up 13.9% year on year, with 9.8 million households shopping online. The average online basket fell to $96, and the average shopper bought from 16 different online brands. This measures Australian consumer online spending, not agent activity. [2]
The Australian Bureau of Statistics (ABS) reported in its final Retail Trade, Australia release for June 2025 (31 July 2025) that online sales were 12.7% of total retail turnover in original terms, up from 11.6% in June 2024. Non-food retailing was 19.0% online and food 7.2%. The ABS ceased this publication after that release, so this is the last reading of the series, and there is no newer ABS online share figure. [3]
Looking ahead, retailbiz reported on 18 March 2026 from the Australia Post eCommerce Report that Australia Post executive general manager Gary Starr estimates agentic AI could influence 30% of digital commerce transactions by 2030. That is an estimated, executive view of future influence, not a measurement of current activity. [4]
And the gap: there is no verified public figure for how much Australian shopping is done or referred by AI agents today. We haven't adapted an overseas number to fill it.
The global figures below are labelled as global because we have no verified Australian equivalent.
- Imperva's 2025 Bad Bot Report, announced 15 April 2025, found that automated traffic was 51% of all web traffic in its 2024 dataset, measured across Imperva's global network. That counts good and bad bots. It isn't a measure of AI shopping agents. It only shows that the web already has a large non-human audience. [5]
- Cloudflare reported on 30 September 2026 that daily requests from AI agents on its global network grew by more than 1,700% over the preceding year. These are requests, not unique visitors, customers or orders, and it is a vendor-reported network observation. Growth from a small base can look large. [6]
- Adobe Digital Insights reported on 29 October 2026 that AI-referred traffic to US retail sites grew 71% year on year in September 2026. Adobe defines this as users clicking a link from an AI source. It is US data, it counts human clicks rather than autonomous agent visits, and it isn't an Australian benchmark. [7]
- Google Search Central's generative AI optimisation guidance, updated 10 July 2026, says Google Search doesn't use llms.txt for visibility or ranking, and that no special AI markup is required beyond standard SEO foundations. That applies to Google Search everywhere, including Australia. It says nothing about other systems. [8]
Taken together, these figures show an audience that is changing and a base of online shoppers that is large. They don't show that agents drive a material share of Australian sales.
Practical steps, foundations first.
The steps below follow a rough order: the foundations that help every customer and search engine, then the bounded actions. No step guarantees that an agent cites you or a customer buys.
1. Fix product data first.
Agents and product feeds depend on complete, consistent product information. That means clear titles, plain-text descriptions, brand, GST-inclusive Australian dollar pricing, availability, images, variants, Australian sizing conventions, delivery coverage and returns.
Two platforms document what they need, and both are specifications for those platforms, not general rules:
- OpenAI's product feed schema lists required fields: item_id, title, description, url, brand, seller_name, image_url, availability and price, plus target_countries and store_country for geographic tagging. Variants get one row each with a stable group_id, and IDs should stay stable when price or stock changes. [9]
- Google's Universal Commerce Protocol guide says that to enable agentic checkout on Google, products need a native_commerce attribute in the product feed, consumer_notice data where legal warnings apply, and a product ID that matches the Checkout API. [10] We explain the protocol in what the Universal Commerce Protocol changes for brands.
Availability and eligibility vary by country and change over time. Confirm whether each platform's agentic checkout is currently offered in Australia before you plan around it.
2. Publish the answers agents will be asked.
Put specifications, compatibility, sizing, lead times for metro and regional addresses, delivery areas by state and territory, trading terms, warranty and returns in plain text on the page. Don't leave them only in images, PDFs or tabs that need interaction.
Under the Australian Consumer Law, consumer guarantees apply regardless of what a returns policy says. Publish policies that sit alongside those rights rather than replace them. The ACCC is the recognised reference. [11]
If you sell to businesses, add account terms, minimum orders, ABN details, who can buy and how to get a quote.
3. Use structured data and standard SEO foundations.
Mark up products, offers and availability. Keep pages crawlable, sitemaps working and pages fast. Google says these fundamentals are what apply to its AI features, with no special AI markup required. We limit the claim to that. [8] Google's page on AI features in Search is the supporting reference. [8]
4. Offer a plain-text route.
Markdown versions of key pages and an llms.txt index make content easy for software to read. Be clear about what that is worth. Google doesn't use llms.txt for ranking, and we know of no source that shows it is a visibility factor in general. [8] Treat it as a low-cost convenience for agents that choose to read it, not a growth lever. We explain the format in why your website needs an llms.txt file.
Matter's own site is an example. It serves a .md version of each article at the same address and lists them in /llms.txt, and it runs on Matter Content Studio and Matter Data Studio. [12] We make no claim about which other AI systems read these files.
5. Keep key information accessible without interaction.
Don't lock buying information behind logins, pop-ups or scripts that only run on a click. Good accessibility practice points the same way: what a screen reader needs is close to what an agent needs. The Web Content Accessibility Guidelines (WCAG) are the reference. [13]
6. Be honest on price and promotion.
Show the total price, including GST and delivery, as early as you can. Avoid drip pricing and misleading urgency claims. Agents comparing offers on a customer's behalf will surface the real total anyway, so an honest price costs you little, and a hidden one is easier to find.
The ACCC polices this. On 9 November 2026 it warned against misleading "storewide" and "limited time" Black Friday sale claims. [14]
7. Keep a person in the loop.
Plan a clean path to a person, or to a checkout the customer controls. Australian customers know cards, PayPal, Afterpay and Zip, so build around those familiar options, and don't promise agent checkout to Australian customers. On Matter's website, the assistant takes initial notes before the team picks up the conversation. It is a practical example of moving from reading about a business to beginning an interaction with it. It isn't evidence that every external agent can complete every step without difficulty.
8. Watch emerging protocols before you build on them.
WebMCP, in early preview from Chrome for Developers (10 February 2026), describes sites exposing structured actions to agents. [15] It is not a standard, it isn't universally supported, and Matter doesn't claim to have implemented it. The Universal Commerce Protocol is the same: emerging and platform-specific, so check its current status before you plan around it.
9. Keep control of your data, permissions and integrations.
Own your product data, the permissions you give agents and the integrations that connect them. Whatever you add should be AI you own: running under rules your team sets, so you can change or switch off an action without starting again.
Start with one useful piece, then one bounded action.
Choose a real customer task: finding the right size, checking whether you deliver to a regional postcode, requesting a trade quote. Test whether a person and an agent can both find the answer, understand its conditions and reach the next step. Don't use agent names in your tests as proof of readiness: a store that works for one named agent hasn't shown it works for the next. Fix the gaps in your product data and policies. Then connect one bounded action, with clear permissions and a route to a person.
One practical check is is-agentic.com, where you can scan your own site to see how agent-friendly it is. It is one check, not a certification, and a good score doesn't show that any agent will cite you or a customer will buy. As an example, matterdesign.com.au scored 100/100, described as a "strong technical baseline", in a snapshot taken on 19 November 2026. [16]
Measure answer accuracy, successful handoffs, qualified enquiries and completed tasks. Keep request volume separate from commercial value, because a rise in agent requests isn't a rise in orders.
If you'd like to look at where your store stands, start the conversation. Bring a customer task that is harder than it should be, and we can look together at what your customers need, what their agents need and where a practical first fix could help.
Sources
- CNBC: Meta's Muse shopping AI agent, 3 October 2026.
- Australia Post: Annual eCommerce Report 2026, March 2026.
- Australian Bureau of Statistics: Retail Trade, Australia, final June 2025 release, 31 July 2025.
- retailbiz: Australia Post: Aussies spend $82.6 billion in online shopping, 18 March 2026.
- Imperva (Thales): 2025 Bad Bot Report announcement, 15 April 2025.
- Cloudflare: The Internet has a second audience, 30 September 2026.
- Adobe Digital Insights: AI traffic and U.S. retail, 29 October 2026.
- Google Search Central: Generative AI optimisation guidance, updated 10 July 2026, and AI features and your website.
- OpenAI: Product feed specification, page undated, accessed 21 November 2026.
- Google: Universal Commerce Protocol guide for Merchant Center, page undated, accessed 21 November 2026.
- Australian Competition and Consumer Commission: Repair, replace, refund, cancel, accessed 21 November 2026.
- Matter: Matter Content Studio and Matter Data Studio, checked 21 November 2026.
- W3C: Web Content Accessibility Guidelines (WCAG), accessed 21 November 2026.
- Australian Competition and Consumer Commission: Warning on misleading "storewide" and "limited time" Black Friday sales claims, 9 November 2026.
- Chrome for Developers: WebMCP early preview, 10 February 2026.
- Is Agentic: is-agentic.com, site scan of matterdesign.com.au, snapshot 19 November 2026.