AI Shopping Agents Are Picking Products Now: A 10-Prompt Test for Your Online Store
Shoppers are starting to ask AI what to buy. Here is a simple test to see whether your store gets recommended, and a checklist to fix what’s holding it back.
For years, selling online meant ranking in search and winning the click. That still matters. But more shoppers now describe what they want to an AI assistant and get a short list of products back, sometimes with a way to buy right there.
If your store is not in that short list, you may never get the visit. This article gives you a repeatable way to check, then a prioritized list of fixes for small ecommerce stores.
A note on sources: the figures below are relayed from recent public write-ups and the research they cite. I have not independently verified every primary source, so check the originals before quoting any number.
What changed, in plain terms
Several big platforms have added shopping features to their AI products. One recent example: according to a write-up by Terry Hutchins on Medium, Google announced on October 15 the US rollout of “Universal Cart” across Search and the Gemini app. Shoppers can save products from different retailers, watch for price drops, and check out through supported options. In that description the merchant remains the seller of record and handles customer support.
Another write-up, by Abhinav Panse, describes how Google and Shopify co-developed an open standard called the Universal Commerce Protocol, and how Shopify syndicates product data to AI channels. The common thread is that these systems run on structured product data. If your data is thin or inconsistent, an AI has little to work with.
What does this do to traffic? Two data points that get cited often:
- Pew Research Center analysed the browsing of 900 US adults in March 2025. Users clicked a traditional result on 8% of visits that showed an AI summary, versus 15% when there was no summary. That study was about AI summaries in general, not shopping specifically, and Google has disputed the methodology.
- Adobe reported that traffic from generative AI tools to US retail sites rose 693.4% year over year over the 2025 holiday season, while noting the base was still modest. It also reported that AI-referred visitors converted better than other visitors in that period.
Treat these as signals, not forecasts. They suggest AI answers are reducing some clicks while the visitors who do arrive may be more ready to buy. Nobody can tell you exactly how that will play out for your store.
Why a single test is not enough
A September 2026 preprint, described in Hutchins’s article, audited 1,536 product-query responses across ChatGPT, Gemini and Google AI Overviews. It reportedly found that recommended products often changed when the same request was repeated, and that sources differed between systems and between a consumer app and its API.
Get Kanisshka’s stories in your inbox
Join Medium for free to get updates from this writer.
The practical lesson: one favorable answer proves little, and one unfavorable answer proves little. You need repeated observations. That is the idea behind the test below.
The 10-prompt test
You need a spreadsheet, about an hour, and access to ChatGPT, Gemini (or Google AI Mode) and Perplexity. Replace the brackets with your real product type, price range, competitors and customers.
- “Best [product type] under $[price]”
- “[Product type] for [specific person or use]”
- “[Your brand] vs [competitor brand]”
- “Is [your brand] legit? Reviews?”
- “Where can I buy [your exact product name]?”
- “[Product type] that ships to [country] within [days]”
- “[Product type] made of [material] / with [certification or feature]”
- “Cheapest [product type] with free returns”
- “Gift ideas for [recipient] under $[price]”
- “Alternatives to [bigger brand’s product]”
Run all ten on each of the three engines. Then pick the five that matter most to your sales and repeat them two more times on each engine, ideally on different days. For each run, log:
- Was your store or product mentioned?
- Where in the answer, and next to which competitors?
- Were your price, stock, shipping and return details correct?
- Which sources were cited?
- Did any link go to the right product page?
Look for patterns, not single answers. A product that appears in two out of three repeats is in a very different position from one that appears once.
What the results usually tell you
- Never mentioned: the AI may not know you exist or cannot read your pages. Start with the product data and crawler basics below.
- Mentioned with wrong details: your data conflicts across places, such as an old price on a marketplace or an outdated blog post. Fix the sources.
- Mentioned only for brand queries: you are known but not recommended for category searches. You need clearer product attributes and comparison content.
- Competitors win on trust questions: they probably have more reviews and clearer policies.
The fix list, in priority order
1. Product titles and attributes
Write titles the way a shopper would describe the item: product type, key attribute, size or quantity, material, and who it is for. Fill in every attribute field your platform offers instead of leaving them blank. AI systems match requests like “waterproof, under $150, fits under a seat” against exactly these details.
2. Accurate price, stock and identifiers
Make sure price and availability are correct and consistent everywhere. Add standard product identifiers such as GTIN or brand and part numbers where they apply. If a shopper’s AI shows a price that your site contradicts, you lose trust at the worst moment.
3. Policies in plain text
Shipping times, costs, regions and return terms should be written clearly on visible pages, not buried in images or pop-ups. Questions like “free returns” and “ships to my country” are common, and the answer has to come from somewhere.
4. Feeds and channels
Submit your catalog through the feeds the major platforms use, such as a Google Merchant Center product feed. If you sell on Shopify, check the settings for the sales channels and AI-related catalog features available to your store, since these change often. Look at each platform’s current documentation instead of relying on a blog’s steps, including this one.
5. Reviews and proof
Collect real reviews and show them on product pages. Do not fake or buy them. AI answers often lean on third-party reviews and discussions, so what others say about you matters.
6. Comparison and use-case content
Write honest pages for the questions in your test: “X vs Y,” “best for [use],” and “how to choose.” Keep them factual, and acknowledge where a competitor is a better fit. Pages that read like pure advertising tend to be less useful as sources.
7. Structured data and crawler access
Add Product markup with price, availability and review information where it fits, and validate it with Google’s testing tools. Check that your robots.txt and any bot-protection settings are not blocking the crawlers you want to reach you.
How to measure whether it is working
Repeat the same 10-prompt test every month and compare against your first log. Also check your analytics for referral traffic from AI products. Source names vary and change, so look at your referral list for ChatGPT, Perplexity, Gemini and similar domains. Compare conversion rates for that traffic against your other sources before drawing conclusions.
What not to do
- Do not rebuild your whole store around one screenshot of a good or bad answer.
- Do not stuff product pages with keywords written for machines. Clear, accurate and complete beats clever.
- Do not buy courses promising guaranteed AI rankings. Nobody controls these answers.
- Do not neglect classic search and email. They still bring most stores most of their traffic.
FAQ
Can I make an AI recommend my store? No one can guarantee it. You can make your products easier to find, read and trust, which improves your chances.
Do I need special software? Not to start. A spreadsheet and the AI tools you already have are enough for the test. Paid tools can help later if you outgrow manual checks.
How often do the answers change? Often. That is why the test uses repeats and why you should rerun it monthly.
The short version
Ask the AI what a shopper would ask, more than once. Fix the data it reads: titles, attributes, price, policies and reviews. Then check again next month. The stores that do this basic work now will be easier for shoppers, and their AI assistants, to choose.
Does your store show up when you ask AI for your own product? Tell me what you found in the comments. Follow me for more practical ecommerce and AI guides.
Sources: write-ups by Terry Hutchins and Abhinav Panse on Medium, citing Pew Research Center, Adobe and a September 2026 research preprint. Photos from Unsplash (free to use): rupixen, RoseBox and Julio Lopez.
