Is your store visible to AI shopping agents?
AI referral traffic to online stores has grown more than tenfold in two years, and an AI visit converts about 60% better than an ordinary one (Adobe, US market). At the same time, buying "inside the chat" does not work in Europe — what matters today is whether an agent finds and recommends you. We tested it on our own content. Below: how it works, the numbers, and what to do first.
Are customers really buying "through AI"?
They are increasingly asking AI before they buy. Adobe Analytics has tracked AI referral traffic to US retail sites since late 2024: by July 2026 it had grown by more than 1,200% Such a visit converts about 60% better and generates 53% more revenue than a visit from other sources. The market is also diversifying: according to Similarweb, ChatGPT's share of generative AI traffic fell from about 76% to 53%, while Gemini passed 25%. So it is not worth optimising for a single platform.
A significant gap separates those numbers from actual purchases. The much-publicised Instant Checkout, launched by OpenAI in September 2025, was moved out of the main chat window after six months and into a merchant-app model. According to trade press, of more than a million eligible Shopify stores, roughly a dozen actually integrated it. The hype ran ahead of market readiness. The lesson for a store owner: what you can win today is visibility, not in-chat checkout.
Can ChatGPT and Google see my store?
This is a question about crawlers, not magic. Each platform runs separate bots with different roles — and this is where stores make their most expensive mistake: they "defend themselves against AI" by blocking everything, and disappear from shopping results.
The mechanics, per OpenAI's documentation, Perplexity's and Google's: GPTBot collects training data — blocking it has no effect on visibility. ChatGPT's search and shopping results rely on a separate crawler, OAI-SearchBot: block that one and you vanish. Likewise, Google-Extended only concerns Gemini training and does not affect Search, AI Overviews or AI Mode — those run on the ordinary Googlebot. Perplexity indexes with PerplexityBot, which does respect robots.txt.
The discovery layer is fully live in Europe. Google's AI Mode has been available in Poland, in Polish, since October 2025. In a simple test from Poland, ChatGPT — asked for a robot vacuum under 1,500 PLN — checked prices in złoty and linked to two Polish electronics retailers. Agents already point to specific local stores. The question is whether yours is one of them.
Can an agent actually BUY from my store?
In Europe — no, and through no fault of your own:
| Platform | What it does today | What it needs from your store | Availability in the EU |
|---|---|---|---|
| ChatGPT — search and recommendations | finds and recommends products, links stores | crawlability (OAI-SearchBot), product data | works |
| ChatGPT — in-chat checkout | checkout via merchant apps | ACP integration + merchant programme | buyers in the US only |
| Perplexity — search | finds and recommends, links stores | crawlability (PerplexityBot) | works |
| Perplexity — Buy with Pro | in-chat checkout, 0% commission | feed + merchant programme | store must sell and ship to the US |
| Google — AI Mode | shopping answers with links | crawlability + Merchant Center | works, in local languages |
| Google — agentic checkout | agent buys when the price drops (Google Pay) | Merchant Center + a feed attribute | products in the US, Canada, Australia |
The standards meant to tie this together — the Agentic Commerce Protocol (OpenAI + Stripe) and the Universal Commerce Protocol (Google + Shopify) — are open and evolving. For a store selling in the EU they are reading material for now, not an implementation project. You are not missing out on checkout, because in the EU it does not exist yet. What you can miss out on is visibility.
Do I need an llms.txt file?
Short answer: do not start there. llms.txt has acquired a mythical status, but the facts are more modest. No major platform has stated that it uses the file, and the official bot documentation from OpenAI, Perplexity and Google does not mention it. Google says outright that its AI features in Search require no "AI files" and no special schema. In one study of bot traffic, llms.txt accounted for 408 requests out of half a billion AI-bot visits. We have such a file ourselves; it costs nothing and does no harm. But we treat it as a bet on the future, not as something that changes anything today.
What must my product data contain?
Three layers, in order of importance:
- Crawlability — the platforms' search bots must be able to reach and read your product pages; the section above lists which ones not to block.
- Structured data that matches the visible content — Google officially requires nothing beyond standard SEO; structured data should describe what is actually shown on the page (price, availability, variants).
- A product feed — the ChatGPT product feed specification has nine required fields: identifier, title, description, URL, brand, seller name, image, availability and price. If you already run Merchant Center for Google Shopping, most of that work is done.
Why does AI recommend my competitors and not me?
Here we let our own data speak. We tested a fixed set of 36 shopping and compliance queries across three AI platforms — the conclusions are unambiguous:
- In 36 out of 36 answers, ChatGPT cited only official sources and documentation — never company content. On that platform a store blog will not outrank legislation or the manufacturer's own site.
- On Perplexity our articles earned 5 citations across 21 queries — all on situational questions of the "does my store have to…" kind, none on definitions. Best result: position 1 of 14 sources, ahead of finance.yahoo.com.
- 4 of the 5 citations went to the English version — even for questions asked in Polish.
- Google AI Overviews did not appear for any of the 36 main queries. Within 28 days, though, the same articles collected 163 impressions in generative AI features from the long tail (Search Console report).
What this means for a store: category pages and guides that answer a specific customer question ("which robot vacuum for a flat with a dog, under 1,500 PLN?") have a real chance of being recommended. Generic "what is a robot vacuum" copy does not. An English version is not an add-on — it is the version agents actually cite.
Where to start: five steps
- Audit robots.txt — check that you are not blocking OAI-SearchBot, PerplexityBot or Googlebot; you can keep GPTBot and Google-Extended blocked if you do not want to feed model training.
- Crawlable product pages — no login walls, no pop-up gates, full content in the HTML.
- Product data — structured data matching the visible content; a Merchant Center feed; the nine ChatGPT feed fields as a checklist.
- Content built around customer questions — situational, concrete, with prices and dates; in your language and in English.
- Measurement — split out chatgpt.com and perplexity.ai referrers in your analytics and watch the generative AI report in Search Console before you trust any "AI visibility" tool.
If you would rather someone took those five steps for you, that is exactly the scope of our AI readiness audit for online stores. It takes 1–2 weeks and costs €1,500–2,500 net. It includes a controlled query test in ChatGPT, Perplexity and Google AI — run by people who do the same on their own content.
