How to Get an E-shop Into AI Recommendations: A Guide for 2026

E-shops spent years optimizing for one thing: being first in Google for a product query. But that query is moving elsewhere. The customer types their whole situation into ChatGPT or Perplexity – budget, use, constraints – and gets a narrow selection of two to four specific brands or shops. Whoever is not in that selection does not even get the chance to offer a price.

Why AI concerns e-shops more than anyone else

Shopping queries are an ideal discipline for language models: they have a clear brief, comparable parameters and plenty of public reviews. Models therefore answer shopping queries confidently and like to recommend concretely. At the same time, what the 2025 data shows holds: where an AI answer appears, clicks on classic results fall by roughly 30 to 50%. But the visitor who arrives anyway is, according to the Semrush study, on average 4.4× more valuable than one from ordinary organic search – they arrive already decided.

For an e-shop that means an unpleasant but simple equation: fewer visits, a higher value for each of them, and a much harder fight over who is in the game at all.

What the model picks its recommendations by

The model has no access to your revenue or to how nice your website looks. It works with what somebody else wrote about you:

• Editorial mentions and tests. An article in a respected medium where your e-shop is named in a specific context (“where to buy…”, “test of…”) is the strongest building block.
• Comparisons and rankings. Overviews like “the best e-shops for…” are among the most cited formats of all.
• Reviews and discussions. Perplexity takes up to a fifth of its citations from Reddit; on the Czech market the same role is played by Heureka, price comparison sites and industry forums.
• Your own content – but only as a supplement that confirms the claims, not as the main source. A brand’s website is biased by definition for the model.

Six steps that work

1. Find out where you stand. Write down 20 queries your customers actually deal with (not keywords – whole sentences) and put them to ChatGPT, Perplexity and Gemini. Note whom the models name. That is your baseline and at the same time the list of competitors you must catch up with.
2. Add facts to the key pages. The GEO research (KDD 2024) measured that concrete numbers, expert quotes and cited sources raise the chance of a citation by roughly 30–40%. For a product that means parameters, measured values, delivery time, return terms – written as a sentence, not as an icon.
3. Build categories on questions, not on keywords. A page “Marathon shoes for beginners” with a table and a clear recommendation is more usable for the model than a category “Running footwear” with 300 products.
4. Deploy structured data. Product, Offer, AggregateRating, FAQPage. It is not magic, it is about the machine not having to guess anything.
5. Let the AI crawlers in. Blocking GPTBot or OAI-SearchBot in robots.txt reliably removes you from the answers. For an ordinary e-shop it is a pure loss.
6. Earn mentions in the media. This is the only step you cannot manage purely on your own website – and at the same time the one that decides. Without external authority you remain one of a thousand shops to the model.

What will not work

Keyword stuffing brought no improvement whatsoever in the GEO study. Mass production of thin blog articles does not help either, nor do fake reviews – models and platforms gradually expose those, and they damage the brand long term. AI visibility cannot be clicked together; it can only be worked for.

Where to start when you have an hour

Put five shopping queries to ChatGPT and Perplexity. If neither mentions you, look at whom they mentioned – and in which media an article about that brand came out. There you have the media map of your niche. Exactly such titles are what you find on the Linketica marketplace, with parameters and price, so you can order a PR article or a mention without calling round newsrooms and waiting for a quote.