Picture the last time you bought something that mattered. You searched, then opened nine tabs. One was a review site and another was a forum thread. A third was a comparison article written by someone who had clearly never used the product.
That was shopping in 2019.
Shopping in 2026 can look like this. Your customer types one sentence into an assistant. "I need a commercial coffee machine for a 40 seat cafe in Newcastle, under twelve grand, serviced locally." A few seconds later she has three options, a comparison of warranty terms and a recommendation.
She never opened your website or scrolled your Google reviews. A machine read them for her, then handed over a shortlist. Whether you make that shortlist comes down to consistent facts, specifics a machine can quote, and what other people say about you.
When Riskified surveyed 5,400 shoppers across nine markets including Australia in September and October 2025, 73% said they already use AI in their shopping journey. Mostly that means research: 45% use it for product ideas and 37% to summarise reviews. Only 13% had completed a purchase after an AI referral. So the assistant is not doing the buying yet. It is doing the shortlisting, and that happens before you get a say.
Which leaves one blunt question. When the machine builds that shortlist, is your business on it?
The shortlist is the new shopfront
For twenty five years the goal was to be found. Rank on page one, run the ad, get the click, then convince them on your website.
Agentic commerce breaks that chain in the middle. The assistant does the finding and a fair share of the convincing. By the time a buyer reaches your website, she is mostly there to confirm a decision.
ChatGPT already surfaces products from merchants in its answers. OpenAI tried letting people pay inside the chat, then reworked its shopping features in 2026 so the purchase is completed through the merchant's own checkout. Amazon has Rufus, which it describes as "an expert shopping assistant" trained on its catalogue and the wider web. Discovery sits with the assistant. The transaction stays with the merchant.
So the contest has moved from winning a click to being one of a few names in a paragraph.
Being on page one used to mean being one of ten. Being in the answer can mean being one of three.
Why the machine picks one business over another
Assistants are not magic and they are not random. They assemble an answer out of whatever the open web says about you, checked against whatever structured data they can read. Three things decide how well you come out of that.
One. Whether your facts are consistent. Suppose your website says you service the Central Coast, your Google profile says Sydney, and a directory from 2021 says you closed. The model now holds three contradictory facts and has no way to resolve them. Naming a competitor it is sure about is the safer move.
Two. Whether your specifics can be lifted. "We pride ourselves on quality workmanship and customer service" is unquotable, because there is no fact in it. Compare "We service commercial espresso machines across the Central Coast and Newcastle, with a 24 hour callout guarantee and a two year parts warranty." An assistant can repeat that sentence word for word.
Three. What third parties say. Reviews, forum threads and industry directories are where an assistant can find confirmation that did not come from you. Comparison articles too. If nobody outside your own website has ever described what you do, you are asking a machine to take your marketing copy on trust. Reviews do more work here than most owners realise, which we unpack in AI reads your reviews before it recommends you.
How to make your business easy to recommend
This is the order we would tackle it in, starting Monday.
- Write your one-line fact sheet. What you sell, who for, where, at what price range, with what guarantee. Plain sentences, no adjectives. Everything else gets checked against it.
- Make every listing match it. Website, Google Business Profile, social bios, marketplaces, industry directories. Same name, same service area, same description. Hunt down the old listings too, because stale pages are where contradictions hide.
- Rewrite key pages so the facts come first. Put the specifics in the opening lines of your homepage and your service or product pages. Each page should contain at least one sentence an assistant could quote on its own.
- Add structured data. Mark up your business details, your prices and availability, your reviews and your FAQs with schema so machines do not have to guess.
- Publish comparison and buying-guide content. Buyers ask assistants things like "which is better for a small cafe, a two group or a three group machine?" An honest comparison page on your site gives the assistant something to draw from.
- Earn outside mentions. Ask customers for reviews that mention the specific job or product. Get listed in the directories your buyers trust. Contribute useful articles to industry publications.
- Check you're reachable. Look at your robots.txt and make sure it is not blocking the crawlers you want reading your site. For product sellers, the technical side is covered in the protocols deciding which brands AI agents can buy from.
The four second audit you can run right now
Stop reading. Open ChatGPT, Gemini or Perplexity. Type the exact sentence one of your customers would type. Leave your brand name out and describe the problem they have.
A Gosford espresso technician might try "best commercial coffee machine repairs on the Central Coast", then "who should I hire to service a cafe espresso machine in Newcastle", then "compare commercial coffee machine suppliers in NSW". Swap in your own trade and town.
Read the answer the way a customer would. One of three results usually comes back, and each tells you what to fix.
- You are not mentioned. The model cannot place your business with any confidence. Start with the fundamentals: the same facts on every listing, plus mentions from third parties that are not paid ads.
- You are mentioned but described wrong. Old pricing, a service you dropped, a suburb you left. The model is quoting stale pages, so update the pages it is quoting as well as your homepage.
- A competitor is described better than you. Read their description closely. It will contain specifics yours does not, and that gap is your content brief for the next quarter.
For a fuller version of this audit, including how to trace each error back to the page that caused it, see what AI is telling your customers about you.
A worked example: the coffee machine shortlist
Go back to that cafe owner in Newcastle. Say you're one of the businesses that could serve her. You sell and service commercial espresso machines from a workshop in Gosford, you've been doing it for years, and your customers love you.
She types her sentence. The assistant names three suppliers. You're not one of them.
So you go looking for why. The homepage headline says "Passion for the perfect pour." Your service area gets one mention, in the footer. Google lists you in Gosford but says nothing about Newcastle, and an old directory entry still shows a number you disconnected. Price ranges live in a PDF. You have plenty of happy customers but only a handful of reviews, and none mention Newcastle or callout times.
All of it is fixable in a few weeks. Start with the homepage opening: "We sell and service commercial espresso machines for cafes across the Central Coast and Newcastle, with a 24 hour callout guarantee." Every listing gets updated to say the same. Price ranges move into plain text on the product pages, with markup. Then you ask your last thirty cafe customers for a review that says where they are and how fast you turned up.
The business is the same one it was a month ago. The difference is that a machine can now read it, confirm it and repeat it.
What this changes about your marketing
Start with what you publish. Vague brand copy has always been weak, and an assistant simply skips it. Every page needs at least one line a machine can lift and repeat without the paragraph around it.
Measurement shifts too. If a buyer spends nine minutes in an AI conversation and then types your business name into Google, your analytics records a branded search out of nowhere. The assistant did your selling, and you should be counting it.
That takes a few new habits. Set up a separate channel in your analytics for visits referred by AI tools. Watch branded search volume month to month. Add "an AI assistant" as an option in your "how did you hear about us?" question, and keep a saved list of buyer prompts you run again every month. In our view these visitors arrive further along in their decision, a point we explore in why AI traffic converts better.
Your budget should follow. We would put one strong comparison page, a well structured service page and thirty real reviews ahead of another rebrand.
Doing nothing has a cost as well. Every month you stay ambiguous is a month a clearer competitor gets named in your place.
The mistakes that keep businesses off the shortlist
- Only checking your brand name. Buyers rarely ask about you by name. They ask about their problem, so audit the problem prompts.
- Hiding facts in PDFs and images. Price lists, warranty terms and service areas locked in a brochure are hard for a machine to read and easy for it to miss.
- Fixing the homepage and stopping. The model reads far more than your homepage, so old listings and profiles need fixing too.
- Chasing tricks. Hidden text and fake reviews add risk and give the model nothing true to repeat.
- Assuming it's only for ecommerce. People ask assistants to shortlist plumbers and accountants as readily as products.
The window is open
The good news is that this work is still rare. Plenty of business owners have never typed their own category into an assistant to see what comes back.
That gives you a head start if you move now. Fix your facts, publish specifics a machine can lift, and go earn third party mentions while the competitor down the road is still arguing about a new logo.
If you'd like help doing it properly, this is exactly what our digital marketing team works on: fact cleanup, quotable pages, structured data, reviews, plus the monthly prompt checks that show whether it's working.
Go type the question. See what the machine says about you. Then decide whether you are comfortable letting that answer stand.
Frequently asked questions
What is agentic commerce?
Agentic commerce is shopping where an AI assistant or agent does much of the research on the buyer's behalf. The buyer describes what they need, and the assistant finds options, compares them and recommends a shortlist. In ChatGPT's current setup, discovery happens inside the assistant and checkout happens through the merchant, so the sale is often decided before the buyer visits you.
How do AI shopping assistants choose which businesses to recommend?
They draw on what the web says about you and what structured data they can read. A business is easier to recommend when its facts match on every listing, its pages carry specific quotable details, and reviews or other third-party sources back it up. Contradictory or vague information makes a model more likely to name a competitor it is surer about.
Do service businesses need to worry about AI agents?
Yes. People ask assistants to shortlist tradespeople and compare consultants, not only to find products. For a service business, the equivalent of product data is service data: what you do, where, for whom, at what price range and with what guarantee. Publish that clearly and consistently and you're doing the same work as an online store.
How do I track sales that come from ChatGPT or other AI tools?
Set up a separate channel in your analytics for referrals from AI tools, watch branded search volume over time and add AI assistants as an option in your "how did you hear about us?" question. Many buyers research in an assistant, then search your name directly, so rising branded searches can be a sign the assistants are recommending you.