Someone asks an assistant for a lamp that will fit a narrow bedside table. They compare a few options in the conversation, then click through to your product page.
They may know very little about your brand. What they want to check is specific: is this the lamp they were shown, will it fit, and when can it arrive?
A product page needs to answer those questions on its own. Shopify’s referral data gives store teams a reason to take that direct arrival seriously.
What Shopify’s referral data shows
The channel is growing, according to Shopify’s published figures. According to Shopify’s Q1 2026 commerce data, referral sessions from AI chatbots grew more than 8 times year over year, while AI-attributed orders grew nearly 13 times.
The report also compares what visitors do after arriving:
- 55% of AI-referred sessions begin on a product detail page, compared with about 20% from organic search.
- Among sessions beginning on a product page, AI-referred visitors converted nearly 50% better than organic-search visitors.
- AI-powered search outperformed organic conversion in 23 of 25 merchant categories Shopify examined.
- AI-attributed orders carried a 14% higher average order value.
Those are Shopify-wide relative figures, not a promise for every store. Shopify has not published the underlying sample sizes, absolute conversion rates, or a category-by-category table. Treat the data as a strong reason to measure your own store, not as a forecast you can paste into a revenue plan.
For your own store, check whether those visitors reach different pages or buy different products before planning changes around the aggregate figures.
Some of the comparison happens before the visit
A conventional search journey is scattered. Someone searches, opens a collection page, reads a guide, checks a competitor, returns later, and perhaps buys in another session. Your navigation, category pages, homepage, and retargeting all get a chance to move the decision along.
An AI conversation can collapse much of that research into one place. The shopper explains the requirement, adds constraints, asks for comparisons, and receives a shortlist. Shopify calls this journey compression.
That changes the role of the landing page. An AI-referred shopper is less likely to arrive asking, “What does this brand sell?” They are more likely to ask, “Is this the exact product I was promised?”
Don’t assume a visitor has read the homepage or seen the delivery information elsewhere. Put the facts needed to check the recommendation on the landing page.
Your product page has to stand on its own
Most stores still treat the product page as one chapter in a longer brand story. That works when the visitor starts at the homepage and browses in order. It is riskier when the first page they see sits halfway down the funnel.
A shopper arriving this way should be able to answer five questions quickly.
Is this the right version of the product?
Variants need specific names, accurate images, and understandable differences. “Standard” and “Pro” are weak labels if the shopper cannot see what changes. Materials, size, compatibility, capacity, fit, and included parts should be explicit.
This is useful to both sides of the journey. The assistant needs structured facts to match the product to the request. The shopper needs the same facts to confirm it matched correctly.
Why is it right for this use case?
A product description should say who the product suits and where it does not. “Premium performance for everyday life” gives an assistant nothing it can defend. “Fits mattresses up to 38 centimetres deep and can be machine-washed at 40°C” does.
Write the first few sentences around the buying decision, then provide the fuller brand story below. Our guide to product descriptions that stay factual and on-brand covers that balance in more detail.
Can I trust the claim?
Reviews, ratings, certifications, guarantees, test results, and clear comparisons do more work than another row of adjectives. A shopper may want independent evidence before trusting a claim, even when an assistant has recommended the item.
Do not manufacture authority with vague badges. Show the source, the review volume, the warranty terms, or the test method.
What happens after I buy?
Shipping time, delivery coverage, returns, warranty, care, assembly, and subscription terms should not require a scavenger hunt. Remember that this visitor may not have seen your shipping strip, policy page, or homepage reassurance.
Keep terms that might change the purchase close to the price and purchase controls.
Can I complete the purchase without friction?
The variant state, price, availability, quantity, and add-to-cart action should remain unambiguous on mobile. Heavy review widgets, personalization tools, and upsells can undermine the high-intent visit if they make the page slow or move the purchase action around.
This is why we treat Shopify speed and conversion work as one system. A more persuasive page is not an improvement if the added weight costs the sale.
The catalog behind the page matters too
The visible product page is only one layer. AI shopping tools also rely on catalog data: titles, attributes, variants, price, availability, product category, shipping, and return policies.
A polished page cannot fully compensate for a vague product title, missing material field, stale variant, or incomplete returns policy. An agent cannot confidently recommend a detail it cannot find, and it should not invent one.
Structured data helps machines read the same facts consistently. Product and review schema should agree with what the customer sees. If the page says an item is available but the markup says it is out of stock, you have created doubt exactly where the system needs certainty.
Assign someone to keep those fields consistent when the catalog changes.
Measure AI traffic as its own channel
Do not wait for AI traffic to become a large percentage of revenue before separating it in reporting. Small channels disappear inside blended numbers, especially when their behavior differs from the average.
In Shopify Analytics, filter reports by the AI answer engines referrer channel. Review ChatGPT, Perplexity, Copilot, Claude, and similar sources; Shopify recommends filtering the referrer host for gemini.google.com when examining Gemini traffic.
Compare AI referrals with organic search on:
- landing-page type;
- conversion rate;
- average order value;
- revenue per session;
- device mix;
- products viewed and purchased; and
- return or cancellation rate, if volume is sufficient.
The last measure matters. A confident recommendation built from incomplete product data could create a fast purchase and a later return. High conversion is useful only when the customer receives what they expected.
Start with ten products, not the whole catalog
You do not need to rebuild every product page to learn something.
Start with ten commercially important products: a mix of bestsellers, high-margin items, products already receiving AI referrals, and products customers frequently compare. For each one, test the real questions a customer asks in ChatGPT, Gemini, and Perplexity. Record whether the product appears, what claims are made, which sources are cited, and where the recommendation becomes uncertain.
Then inspect the corresponding product page and catalog record. Fix missing facts, unclear variants, conflicting schema, weak proof, buried policies, mobile friction, and slow scripts. Re-run the same prompts monthly and compare the referral report with your baseline.
That review is also the basis of our Shopify AI Discovery & Product Page Audit: check the facts assistants can find, inspect the landing experience and prioritize the gaps. Visibility in a particular answer can’t be guaranteed.
Give important product pages a complete introduction
AI search is not replacing organic search, and organic still sends far more traffic. Your homepage, collections, editorial content, and conventional SEO remain important. In fact, much of what makes a store visible in search also makes it easier for AI systems to retrieve and cite.
Choose a few important product pages and review them without navigating anywhere else. Any essential answer you have to leave the page to find is worth reconsidering.