A report published on 22 September shows that only 11% of small and medium businesses are ready for customers who arrive through AI assistants, compared with 19% of large merchants (PYMNTS). If you sell products or services and you have a website, that number concerns you directly. The good news: getting ready is not about new technology, it is about putting order into information you already have.
AI recommends, you get paid
A year ago it looked like the future was "buy inside the chat". OpenAI had launched Instant Checkout in ChatGPT, where the customer paid without leaving the conversation (OpenAI). Since then the direction has changed: payment moves back to the merchant, while the AI assistant keeps the discovery and recommendation part (Checkout.com). Google followed the same logic with the Universal Commerce Protocol, an open standard announced in January 2026 together with Shopify, Etsy, Target and Walmart (Google Developers).
For a small company in Romania, the translation is simple. The customer asks an AI assistant, gets a few options, then lands on your site to buy or to ask for a quote. You do not need to integrate with any protocol to matter in that first step. You only need the information about what you sell to be findable, clear and correct.
What "ready" means, in plain words
From the same report: only 15% of merchants have structured product data, meaning information written in a format a program can read without guessing (PYMNTS).
It sounds technical, but in practice it is a short list:
- Every product or service has its own page, with the price, what is included and the delivery time written as text, not only inside an image or a PDF.
- Availability is up to date. An assistant that recommends something you no longer have brings you an unhappy customer, not a sale.
- Company details (address, opening hours, area served) are identical everywhere: on the site, on Google, on social networks.
- Pages load fast and read well on a phone.
I wrote at length about how to appear in AI answers and about cleaning up your company data.
You cannot fix what you cannot see
The part most often skipped is measurement. Only 23% of merchants can clearly identify both the visits coming from AI and the completed orders, while 37% cannot separate AI traffic from the rest of their digital traffic at all (PYMNTS).
Here is what you can do this week, with no budget: open your website statistics, go to the traffic sources section, and look for names containing chatgpt, perplexity, gemini or copilot. Write down today's number somewhere. Compare it in a month. Without a starting number, any discussion about "AI is bringing us customers" stays an opinion, and you do not make budget decisions on opinions.
The last step is the most expensive one
87% of merchants say their checkout process needs improvement, and only 13% consider it fine as it is (PYMNTS). That is the irony: we pay to be found, then we lose the person right at the end.
Run the test yourself, as a first time customer. Time how long it takes from the product page to a submitted order. Count how many fields have to be filled in. Check whether the final price, with shipping and VAT, shows up before the last click. If you sell services, the same logic applies to your quote request form: three required fields bring in more requests than eleven.
What you do not send to an AI
Getting ready for AI-driven customers is about public information: price, description, availability. It does not mean, in any way, taking data out of your company that you are not allowed to share.
The rule I repeat on every project: in the window of a public AI assistant you do not put customer personal data (full name, national ID number, address, phone, e-mail), you do not put contracts, you do not put individually negotiated prices, and you do not put passwords or access keys. If you want help with a text that starts from a real case, replace the name with "Customer A", delete the contact details and round off sensitive figures. The result is just as good, and the risk disappears.
For a company in Romania this also matters legally. GDPR requires you to know where your customers' data ends up and on what legal basis you process it. Whether an AI provider uses conversations to train its models is something you check in the account settings, not something you assume. I covered the topic in the guide on GDPR and AI.
Today's step
Pick your three best selling products or services. Open their pages and read them like a stranger: within ten seconds, is it clear what it is, what it costs and how to buy it? If the answer is no to any of those, you have found your first repair. You need no protocols, no integrations and no agency budget to do it.
The advantage of a small company is speed. Sara Craven, at Visa, says small businesses have the ability to pivot and to learn faster (PYMNTS). An afternoon of cleanup on three pages is exactly the kind of move a large merchant needs three months of meetings to make.
Read next
How to appear in AI answers: a 2026 GEO guide, Clean data: how to prepare your company information, Your ChatGPT ad can now talk to customers