Practical Ecommerce reported on 1 November 2026 that Persa Flores, a family-owned florist in Spain, now gets one in five of its online orders from shoppers who found it in ChatGPT. José Pérez Sánchez of the company gave the figures, month by month. Most of those shoppers landed first on one of the shop’s city pages, the ones about sending flowers to Madrid, Marbella or Alicante, published in Spanish and English.

Month (2026) Share of orders from ChatGPT Share of revenue
July 4% 2%
August 12% 11%
September 21% 18%
October (to the 27th) 23% 26%

Source: Practical Ecommerce, 1 November 2026, figures supplied by Persa Flores.

The claim is that ChatGPT has become a sales channel

That is the exciting reading, and the orders are real. But I always ask what actually changed and what didn’t.

What changed is buyer behaviour. People now ask an AI engine “send flowers to Marbella today” and act on the shop it names. Four months ago that was a Google search. Now a share of it happens inside a chat window, and the florist’s own numbers show that share growing every month.

What didn’t change is the florist. The article says the company did not apply any optimisation for the chatbot until mid-October, after the rise had already happened. There was no new tool, no agency, no trick with the model. The pages ChatGPT sends people to were already there.

One more number tells the story. Of the customers who came from ChatGPT, 84% arrived first on a city page. Of the customers who came from Google, about a third did. The engine was picking the specific page over the homepage far more often than Google ever did.

A city page is the shape an engine can quote

The AI doesn’t pick the biggest brand. It picks the one it understands most clearly and can quote most precisely. A page called “send flowers to Alicante” is one specific answer to one specific question. It has a place, a service and the delivery facts for that place. When someone asks about flowers in Alicante, there is nothing to interpret.

Compare that with a homepage that says “beautiful bouquets delivered across Spain”. That line fits a thousand florists. An engine has nothing precise to name, so it names someone else. Numbers and specifics survive in an AI answer. Adjectives get dropped.

This is the super long tail. Everyone asks their agent the same question slightly differently: by city, by occasion, by day, in their own language. Persa Flores caught that spread the only way it can be caught, with one plain page per variant. Your own pages get you cited fast. Other people’s pages keep you there.

What this looks like outside a flower shop

Take a clinic. Patients don’t ask “which clinic is good”. They ask whether a specific treatment is available near them, what it costs, and whether they can be seen this week. Each of those is a page, or it is a question the engine answers with a competitor.

Take a plumber or an aircon service. The questions come with an area and a job attached. “Aircon chemical wash in Tampines, same day.” If the only page is “our services”, there is nothing for an engine to match to that question.

Take a marketing team with a product catalogue. The buyer asks “does this work with the software I already use”. We do this at AeroChat: we publish many pages, each about one integration with one specific piece of software, and each one catches one long-tail question. None of them is clever. Each one is precise.

Two honest limits on the florist’s story. The figures are one shop in one country over four months, reported by the shop itself. And AI answers change week to week, so what ChatGPT names today is not a fixed position. Treat it as consistent with the pattern I see, not as proof that it will repeat for you.

What a business should do

  1. Write down the questions your buyers actually ask, with the place, product or situation attached. Not topics. Questions.
  2. Check which of those questions has a page of its own on your site. Count them. The gap is usually large.
  3. Write the missing pages plainly, one question per page, with the specific facts an engine can quote: areas served, cut-off times, prices, what is included.
  4. A month later, ask the engines those same questions and note which ones name you. Date the results. Then write the next batch.

Source: Practical Ecommerce: Florist Gets 1 in 5 Orders from ChatGPT