Can AI referrals bring buyers, not just curiosity?

Imagine a Polish homeware shop selling a desk lamp. Its page has a photo, a price and one line: “A stylish lamp for any interior.” A buyer asks an assistant for a compact lamp with adjustable colour temperature and USB-C power. If the shop sells a suitable model but never states those features, it makes the match harder to judge.

This is not an “AI SEO” trick. Replacing vague copy with useful answers can create an opportunity to be discovered without paying for each visit. Content still costs effort, and neither mentions nor sales are guaranteed.

What the conversion data says

The April 2026 Adobe Digital Insights report analysed US retail visits. In March 2026, AI-referred visits converted into purchases at a rate 42% higher than visits from other sources. That is not a forecast for Poland or proof that rewriting descriptions caused the difference. Adobe sells digital marketing products, so read its findings with their method and commercial perspective in mind.

Clicks are not guaranteed either. Pew Research Center studied the browsing behaviour of 900 US adults. On Google searches showing an AI summary, people clicked traditional result links less often than on searches without one; clicks on cited sources within the summary were rare. That is not a study of Polish shopping, but it puts a useful limit on the promise of “free AI traffic”. Being visible, earning a visit and completing a purchase are distinct outcomes.

Start with buying decisions, not a list of magic keywords

A shopper may need to compare dimensions, compatibility, materials, use cases, delivery costs or returns. OpenAI describes ChatGPT shopping research as a conversation that clarifies constraints and compares options by factors such as budget, features and trade-offs. This describes a service, not the frequency of any particular query in Poland.

These are illustrative questions written for this article, not observed customer prompts or claims about search volume:

• “Which lamp fits a narrow desk without shining into my eyes during video calls?”
• “How do these models compare if I need adjustable light colour and a replaceable light source?”
• “What will delivery cost in Poland, and how long do I have to return it?”

Pick real products and list the decisions their pages leave unresolved. Service conversations, reviews and site-search data can reveal questions. Treat questions you invent as hypotheses, not customer quotes. Put answers on useful pages rather than creating a landing page for every wording.

Five product-content changes worth testing

1. Be specific about the product and its limits

“Desk Lamp Model X” tells a buyer less than a title with its type, relevant dimensions or power source. Give verified specifications and explain when the product does not fit. If its light source cannot be replaced, do not hide that behind “perfect for every desk”. This is useful product information, not a ranking trick.

One lamp, two ways to describe it

Before

Desk Lamp M1

A stylish lamp for every interior. The perfect choice for your home.

The buyer cannot tell its size, bulb type or limitations.

After

M1 lamp with an adjustable arm

Position the arm to direct light onto your desk without moving the base.

• Base 18 cm wide, height 45 cm
• Replaceable E27 bulb, not included
• No colour-temperature control

2. Put the answer near the facts

Replace the manufacturer's generic paragraph with a use case, features, dimensions and honest fit guidance. Original photos can show scale, but keep critical details in text. Google recommends accessible text and internal links. Its guidance concerns Google Search, not a guarantee of citations by every AI system.

3. Keep price and availability consistent

The page, stock system, feed and structured data should agree. Google's documentation describes Product markup and Merchant Center feeds; using both can improve eligibility and verification, not guarantee display. For Poland, check delivery and returns too. Free listings can show eligible products without an advertising charge, though exposure is not guaranteed and data maintenance takes work.

4. Make the page discoverable before chasing an “AI schema”

Check indexing, crawler access and links from categories. Google says AI Overviews and AI Mode need no special file or extra schema.org markup. For ChatGPT search, OpenAI advises allowing OAI-SearchBot access to content intended for summaries and snippets. Thin pages made for imagined prompts cannot replace a useful catalogue.

5. Support the decision beyond the product page

The buyer may need a comparison, sizing guide, delivery details or first-hand review. Google lists category descriptions, reviews and service information as useful shopping content. Connect guides to products with links. If you claim to have tested an item, show your method and evidence rather than repeating marketing copy as experience.

How will you know whether the work paid off?

Start with one category. Record Search Console visibility and clicks, page visits, add-to-cart events, orders and revenue. Compare the same measures after editing, allowing for seasonality, promotions and price changes. More visits may not mean better buyers; a higher conversion rate alone does not prove the copy caused it.

OpenAI says some ChatGPT search referrals can be identified through utm_source=chatgpt.com. Track that traffic separately, but do not assume the referral field captures every AI-influenced journey. An observational preprint that has not been peer reviewed found that a brand recommendation in an assistant response can be followed by ordinary brand searches and website visits. Its authors did not observe purchases and could not rule out every alternative explanation. It is a reason to watch brand-search patterns as well, not proof of incremental sales for your store.

Make a small, measurable start

Do not rebuild your entire catalogue because of one attractive chart. Choose products buyers must compare, add verified facts and useful answers, align the data across channels and measure what follows. Better product content may help people discover your shop through unpaid surfaces and decide whether an item really fits. Whether it brings valuable traffic and more orders to a Polish ecommerce business is a question for that business's own data.