Build it
Make it sell
Keep it running
AI search
One order in ten no longer starts with a search engine.
A growing share of orders in the shops we look after arrives from ChatGPT, from Perplexity, or from an AI answer inside Google. Not a forecast — WooCommerce records the source of every order, and we counted them.
What we measure, not what we expect
More than one order in ten.
Across five shops we look after — among them Apotheke zum heiligen Joseph and Natürlich by Dräxlmaier — more than ten percent of orders in recent quarters have come from AI sources. That figure is not modelled or estimated: WooCommerce stores the origin of every order, and we read it. For comparison, the industry currently reports well under one percent of traffic from these sources and treats fifteen to twenty-five percent as a projection for the coming years. Those shops are already there, and not one of them did anything unusual to get there.
Why some shops are found and others are not
Most of this was always the right thing to do.
An agent does not browse. It compares, and it decides in a fraction of a second whether your product is comparable at all.
- Product data that is actually complete
- A missing weight, a missing size, an unclear pack quantity — and your product drops out of the comparison silently. Nobody sees it happen, least of all you.
- Structured data that agrees with the page
- Price, availability and shipping in machine-readable form. Where the markup says one thing and the page says another, an agent trusts neither and moves on.
- Real HTML rather than assembled-on-arrival content
- Much of what a page builder produces only exists after the browser has run it. An agent reading the page as delivered finds an empty frame where the price should be.
- Delivery cost and date, stated plainly
- Agents avoid ambiguity because they have to act. “Ships in two to three days, depending on region” is not an answer they can compare, so the offer is skipped before a person ever sees it.
- A page that answers before it is asked
- How long it lasts, what it costs per use, what the difference to the neighbouring product is. The same four answers a customer needs — an agent needs them in text it can read. That is the product page work, unchanged.
- Structure a machine can follow
- Headings in order, real labels, lists that are lists. This is the same requirement as a screen reader has — which is why a shop built properly for people turns out to be readable by agents without anyone planning it.
Five of those six are things we would have done anyway, and did. That is the honest explanation for our figure: nobody set out to optimise for AI. The shops were built to be read, and something started reading them.
Where we draw the line
Your catalogue stays on your server.
There are plugins that promise to connect a WooCommerce shop directly to ChatGPT or to Google. Some are honest about being experiments — one of the better-known ones states in its own documentation that it is not production-ready. Others work, and work by copying your product catalogue onto a third party’s server so that agents can query it there.
For a pharmacy or a supplement brand, where your product data sits is a question that belongs before the technology rather than after it. We would rather raise it early than explain it later.
So we do the shop instead. Structure, product data, markup, speed — on your own installation, with no connector and nobody in the middle. That is where our figure comes from, and it is the part that keeps working regardless of which protocol wins.
And the limit, plainly: if you reach the point where you need a genuine protocol connection — a live checkout inside an assistant rather than a link back to your shop — that is a different project with a different price, and we will say so rather than pretend the groundwork covers it. For almost every shop we are asked about, it does not come up yet. The shop itself is still where the work is.
Check your own orders first.
WooCommerce already records where each order came from. Open the last few months and look. If AI sources are in there, the question is no longer whether this matters — and if they are not, that is worth knowing too, and we will tell you why.