Imagine you sell hiking boots.
A customer asks ChatGPT which waterproof boots suit wide feet. It recommends a model you stock, explains why it fits, and shows somewhere to buy it.
The link goes to another retailer.
Your product category is visible. The manufacturer gets a mention. Another store gets the opportunity to make the sale.
That distinction is easy to lose in reports about AI visibility. A product appearing in an answer can look like progress, even when your shop is absent from the buying decision.
For an ecommerce business, the useful question is specific:
When someone asks for help buying what you sell, does your store become an option?
To answer that, you need to look beyond the lists of websites ChatGPT and Perplexity cite most often. You need to see which sources appear for your customers’ questions, and where those answers send the customer next.
A big citation percentage can tell a surprisingly small story
Take the claim that Reddit dominates AI citations.
In Profound’s analysis of 680 million citations, covering August 2024 to June 2025, Reddit accounted for 46.7% of citations within Perplexity’s ten most cited sources.
Its share of all Perplexity citations in that dataset was 6.6%.
Both figures are correct. The denominator changes the meaning.
The same research found Wikipedia leading ChatGPT’s overall citations. Even within one dataset, the platforms had different source patterns.
Then there is timing. The Ahrefs Perplexity leaderboard updates monthly. In its September 2026 snapshot, Reddit leads with 21.6% and YouTube follows with 20.8%, measured against the combined citations of the top sources. Those percentages cannot be treated as directly comparable with Profound’s overall citation shares.
None of these figures tells a hiking retailer whether customers asking about wide-fit boots are being shown specialist shops, outdoor magazines or a manufacturer’s sizing guide.
Use the big studies to discover places worth investigating. Before putting budget into any of them, check whether they appear in the answers that matter to your business.
A citation, a recommendation and a place to buy are different outcomes
An assistant might cite your guide to waterproof materials while recommending a competitor’s store.
It might recommend your shop based on an independent review, without citing your website.
Or it might show a product you sell, with a purchase link to the manufacturer.
All three belong in your measurement. They tell you different things.
For a retailer selling other companies’ products, I would track the store name separately from the product brand. If you sell Salomon boots, a mention of Salomon does not count as a mention of your shop.
Even when you manufacture your own products, check which merchant receives the purchase link. Brand exposure and direct sales opportunities can go to different places.
This also changes how you interpret competitors. A review site may be the source shaping the answer, while a large retailer is the seller benefiting from it. Record both. Otherwise, you can end up trying to outrank the wrong business.
ChatGPT shopping needs a separate check
Text citations do not capture everything a customer sees.
OpenAI’s shopping documentation explains that product results can use structured data from merchants and other providers, alongside third-party content. The shopper’s question and context also affect selection.
For your audit, that means keeping two separate records: links used as sources in the written answer, and product results with their displayed merchants.
There is evidence that the balance between retrieval methods can change quickly. In a September 2026 analysis, Profound reported that feed-integrated retrieval in its tracked shopping data rose from 8.26% to 61.54% around July 10.
That is one vendor’s observation, using its own sample and classification. It does not establish the share for every category, prove the cause of the shift, or mean that stores without a direct feed cannot appear.
In fact, OpenAI’s merchant page says a feed is not required if ChatGPT already crawls your site. Feeds offer more control over product information. As checked on 25 September 2026, that page lists shopping availability in the US and says Shopify and Etsy catalogues are already integrated.
Check the current options for your platform and your customers’ market before investing in an integration. Your business address alone does not tell you which shopping experience your customers can use.
My recommendation is to include catalogue quality in your visibility work. Start with the products that matter commercially. Check that their names, variants, prices and stock status agree across your pages and any feeds you supply.
A recommendation is of little use if it leads to the wrong size or an unavailable product.
Run a small test you can repeat
Start with one commercially important category. Choose something you can fulfil reliably and that contributes useful margin.
You can build an initial picture with 20 questions. Treat the result as a diagnostic sample; it will not estimate your share of every AI conversation.
Choose questions from actual buying decisions
Use customer service emails, on-site searches and questions your sales team hears. Include five prompts from each group:
Finding a suitable product
“Which waterproof hiking boots suit wide feet under £150?”
Comparing options
“How do these two boot models compare for long walks in wet weather?”
Choosing a seller
“Where can I buy wide-fit hiking boots with delivery to Manchester and easy returns?”
Solving a problem
“Why do my hiking boots let water in around the seams?”
Replace the examples with your products and your customers’ language. For comparison questions, use real models.
Keep your store name out of this initial discovery test. Asking an assistant about your business by name measures something different from checking whether it introduces you to a buyer.
Branded questions can be useful later for finding inaccurate information about your shop.
Keep the conditions consistent
Run each question twice in ChatGPT and twice in Perplexity, starting a new conversation each time.
That produces 80 answers: 40 per platform. Reading and checking them takes time. If that is too much for your first pass, use eight questions, two from each group, and expand once you have a useful process.
Record the date, platform, model or mode where visible, language and target location. Use search-enabled modes for a citation audit, and note whether a run actually produced web sources. Minimise memory and personalisation where the product allows it, but do not assume a fresh chat reproduces every customer’s experience.
Keep the wording and settings consistent when you repeat the test. Save the answers so you can compare what changed.
Record where the customer could go next
For each answer, note:
Whether your store is mentioned, and whether the mention is positive, neutral or negative.
Whether a URL on your domain is cited, and which page it is.
Whether your store is recommended as somewhere to buy.
Which other sellers are recommended and which sources support the answer.
Whether product cards appear, and which merchant links they show.
Open the important citations. Check whether the linked page supports the claim and whether it actually mentions the recommended business. A nearby citation is not proof that a particular page caused a recommendation.
Calculate rates using answers, not unique questions
If your shop appears in 8 of the 40 ChatGPT answers, its store mention rate in that sample is 20%. Count each answer once, even if it mentions you several times.
Use the same denominator for your own-site citation rate.
For a seller recommendation rate, use the answers to prompts you classified as seller-seeking before testing. In this example, the five “choosing a seller” questions run twice give you ten answers per platform. If two recommend your shop, that rate is 20%.
Keep informational questions separate. An answer about leaking seams may help a buyer without recommending any seller.
Report ChatGPT and Perplexity separately, and keep the intent groups visible. A single average can hide the fact that your guides get cited while your shop disappears when someone asks where to buy.
You can also record first position when an answer provides an ordered list. Do not treat first position as evidence that you received a click or a sale.
Find the gap you can actually fix
Group the sources into marketplaces, large retailers, manufacturers, independent shops, editorial sites, community or video platforms, and reference sources. Keep community and video separate if they suggest different actions.
The goal is to identify a recurring pattern, then inspect the pages behind it.
When independent shops appear repeatedly, compare the cited pages with yours. Do they answer a sizing question you leave unresolved? Show compatibility across brands? Explain delivery restrictions clearly? Improve the missing information where customers need it. Do not assume copying the page format will reproduce the result.
When editorial reviews shape the answers, check whether their comparisons cover products you stock and whether they mention your store. A useful outreach approach might involve supplying a product for testing or sharing original information. The editorial decision belongs to the publisher; an inclusion or citation is never guaranteed.
When manufacturers dominate, look for questions you can answer from a retailer’s perspective. A shop stocking several brands can explain differences in fit, compare alternatives and offer buying advice across the range. Use actual experience or data you can substantiate.
When marketplaces and large retailers take the purchase links, inspect the offer as well as the content. Can you compete on stock, delivery or specialist service while protecting your margin? There is little value in becoming more visible for an order you cannot fulfil profitably.
When forums or videos recur, read the discussions and watch the relevant sections. Identify the unresolved questions. A demonstration may be more useful than another article for something a customer needs to see. Participate transparently and make your connection to the business clear.
Reference sources may simply be supporting background facts. You do not need to compete with every cited domain to become a relevant seller.
For your first round, choose one change with a clear commercial reason. Record it, repeat the same test after the information has had time to be discovered, and look for a sustained pattern across later runs. A single different answer is weak evidence of improvement.
Keep the commercial result in view
There are useful limits to what this exercise can tell you.
A citation shows that your page appeared as a source in an observed answer. It does not reveal an internal “trust score”. An increase in mentions does not establish an increase in demand.
Likewise, Google rankings and AI citations are different measurements. Ahrefs’ study of ChatGPT’s 1,000 most cited pages in September 2025 found that roughly 28% had no organic keywords in its data. That does not mean those pages were absent from Google’s index.
Check technical accessibility and search indexing as part of the diagnosis, without treating Google index status as a universal rule for every AI system. A crawl block, missing product information or broken purchase link deserves attention regardless of the citation score.
Then connect visibility work to the business. Track identifiable AI referrals, orders and revenue. Where possible, assess the margin those orders leave after fulfilment and returns.
Attribution will be incomplete. Someone may discover your shop in an assistant and later search for your name. A post-purchase “How did you hear about us?” question can add context, although it will not capture everything either.
Use the prompt test to decide what to investigate and improve. Use business outcomes to decide how much to invest.
The goal is to make your shop a credible buying option for the customers you can serve profitably.
Pick one category this week. Check a handful of buying questions. Follow the links, identify the sellers, and look for the most useful improvement you can make.
That will give you a better starting point than another industry-wide citation percentage.
Want help finding your store’s biggest visibility gap?
I help ecommerce businesses improve how customers find them through Google and AI search.
If you want help deciding where to start, send me your store URL and the category that matters most to your business. Include your main customer market so I can put the findings in context.
Kind regards,
Neven Lovrekovic

