Most measurement of an AI recommendation stops at the click. The more interesting number is what happens on day six.
Similarweb and Statista measured something unusual in The State of Ecommerce 2026. They took pairs of competing brands. Then they found the shoppers an AI assistant had recommended one brand to, and watched where those shoppers went over the next week.
Sephora was recommended, Ulta was not: 7.9% of those users visited Sephora within seven days, against 3.3% who visited Ulta. Then they flipped it. When Ulta was the recommended brand, 7.6% went to Ulta and 4.6% to Sephora.
The effect followed the recommendation, not the brand.
Plug In SEO checks the structured data AI assistants read when they decide which store to name.
What the AI recommendation data actually says
The figures come from Similarweb clickstream data, US desktop only, covering July to December 2025. Each pair compares two shares. How many users visited the recommended brand, against how many visited its named competitor, within seven days.
- Amex recommended: 7.2% went to Amex, 3.1% to Capital One
- Capital One recommended: 14.2% went to Capital One, 3.8% to Amex
- Skyscanner recommended: 9.5% went to Skyscanner, 7.6% to Kayak
- Kayak recommended: 12.0% went to Kayak, 3.4% to Skyscanner
- Sephora recommended: 7.9% went to Sephora, 3.3% to Ulta
- Ulta recommended: 7.6% went to Ulta, 4.6% to Sephora
Similarweb and Statista, State of Ecommerce 2026. US, desktop, July–December 2025.
Two things make this more useful than a typical citation count. First, the design is symmetric. Each brand appears as both the recommended and the unrecommended side, so the gap is not just one brand being bigger. Second, the window is seven days. So the recommendation is not only catching a click in the moment — it still shapes behaviour almost a week later.
The honest limits: these are large brands in finance, travel and beauty, not Shopify stores. The measurement is US desktop. Nothing here proves the same multiple holds for a store with 400 products and no brand recognition. But it does establish one thing. An AI recommendation behaves like a durable referral, not a single click.
What an AI recommendation means for a Shopify merchant
So if a recommendation keeps paying out for a week, two things matter. Can an assistant name your store confidently? And does anything break when the shopper finally arrives, days later?
1. Make the facts an assistant needs easy to extract
Assistants name stores they can describe concretely: what you sell, price range, shipping, returns. Put those in text on the page, not in an image or a PDF. Product and Organization structured data give the same facts in a machine-readable form.
2. Check that your structured data is actually valid
Invalid schema fails silently — the page looks fine to a human and is skipped by the parser. Run your product and collection templates through a validator, and fix required-property errors first.
3. Assume the AI recommendation pays out late
This is the part the seven-day window changes. A shopper may act on a recommendation six days later. By then a URL you changed is a 404, and a product you unpublished is a dead end. So redirect rather than delete, and leave collection URLs alone during a campaign.
4. Keep stock and price accurate
An assistant that described your price three days ago sends a shopper expecting it. A mismatch at arrival is a bounce you paid for with the recommendation.
5. Measure the AI recommendation over seven days, not one session
If you are looking only at last-click attribution from AI referrers, you will undercount this entirely. Look at a seven-day window for direct and branded-search lift after any period where you know you were being cited.
What not to take from this AI recommendation data
This is not evidence that AI referrals are about to replace search traffic. The same report found that 89% of shopper journeys involving AI also involve search, and that journeys using both convert better than either alone. The recommendation effect sits on top of search; it does not substitute for it.
It is also not a reason to write pages aimed at assistants rather than people. The brands here were not recommended for stuffing keywords. They were recommended because they were the obvious, well-documented answer. One more caution: six brand pairs in three categories is a finding, not a law. Treat the direction as solid and the exact multiple as unproven.
Where to go next
Want the tactics rather than the measurement? Our playbook on getting a Shopify store recommended by ChatGPT covers what influences the naming decision. If you are deciding whether AI traffic deserves budget at all, we looked at whether AI traffic is worth chasing yet.
Frequently asked questions
How long does an AI recommendation keep driving traffic?
Similarweb and Statista measured visits up to seven days out. Across six brand pairs, the recommended brand still drew roughly two to four times the visits of its unrecommended competitor. The study did not test beyond seven days, so nothing is known about what happens after that.
Does this apply to small Shopify stores?
Unproven. The brands measured were Amex, Capital One, Skyscanner, Kayak, Sephora and Ulta — all large and well known, measured on US desktop. The mechanism is plausible for smaller stores, since the recommendation drives it rather than brand size. But no data in this report covers them.
Should I stop optimising for Google and focus on AI assistants?
No. The same report found 89% of AI-involved shopper journeys also use search. Journeys combining both convert at 23.0%, against 17.3% for search alone and 12.5% for AI alone. Optimising for assistants and optimising for search are the same job, not competing ones.
How do I tell whether an assistant is recommending my store?
Ask the assistants directly with the questions your buyers would use, and record which stores get named. Then watch direct and branded search over the following week, not just referral traffic. Much of the effect arrives as a later visit with no AI referrer attached.
Plug In SEO flags missing and invalid structured data across your product and collection templates, so assistants have the facts they need to name your store.