Does your own website show up in AI shopping answers?

Brand owners often ask the same thing first: does my own website get used when an assistant answers a shopping question? The October 2026 source lists give a partial answer. They show that brand sites do appear next to the answers. They cannot show how likely your site is to appear, and this post explains why before it turns to what to keep current on the site.

What counts as an own-site match?

We used a deliberately strict rule. For each answer we took the brands it recommends, removed spaces and punctuation from each name, and compared that with the first label of every domain in the answer's source list. An answer has an own-site match when the two are equal: samsung.com for Samsung. Brand names shorter than three characters are skipped.

That rule is simple to repeat and it misses real cases:

• a store on a subdomain, such as shop.example.com, whose first label is "shop";
• a country domain that does not start with the brand's name;
• a parent company's domain, when the brand sits under another name;
• the brand's own page on a retailer's site, which counts as the retailer's domain.

Two more limits matter. Only brands the answer recommends are counted, not brands it merely mentions. And the denominator is every answer, not every answer that names your brand.

How often did a recommended brand's site appear?

The October 2026 edition collected 800 answers from OpenAI's API with web search (gpt-6-luna) and 787 answers from Perplexity's API, where the Sonar model (perplexity/sonar) ran with web search, across 160 categories in three markets. For OpenAI the source list is the pages the answer links. For Perplexity it is the pages the search retrieved, which the answer may not use.

That is about 373 OpenAI answers and about 488 Perplexity answers. Three things stop these figures from being read as "your odds".

1. They are per answer, not per brand. The average answer recommends 2.2 brands on OpenAI and 2.5 on Perplexity (excluding the alternative-to-a-brand question, October 2026 edition), so each answer gives several brands a chance to match. A given brand's own rate over all answers would be lower than these figures, because each answer can match through any one of its brands. Among only the answers that recommend that brand, the rate could be higher or lower. We have not computed either. The two percentages above use every answer, including that question, whose answers list more brands.
2. The list lengths differ. OpenAI's answers linked a mean of 2.0 distinct domains, so each link is a larger share of the list. Perplexity's retrieved lists held a mean of 15.2 distinct domains, which makes a match easier to find. Do not rank the engines on the two percentages.
3. They exist only where a brand was recommended. We do not know how often a brand that no answer recommended had its site linked, because the rule needs a recommendation to match against.

Which store and maker sites show up in OpenAI's links?

Perplexity's retrieved lists lean on editorial review sites and retailers: a Wirecutter page was in 333 of its 787 lists (42.3%). The Wirecutter post covers that. In the October 2026 edition, several store and maker sites reach the 30 most frequent domains in OpenAI's linked pages (797 answers with a link). The table shows all of them, labelled by kind, as examples only. The counts are answers that link the domain, not own-site matches, and it is not a ranking of brands.

The denominator is the 797 OpenAI answers that link at least one source. None of the maker domains is among Perplexity's 30 most frequent, while four of the store domains are: rei.com, target.com, liverpool.com.mx and walmart.com.mx. The Kind column is our reading of each domain, not a measured field. Our output lists only each engine's top 30 domains, so we cannot show how the other matches are spread across brand sites.

Can an assistant read your site at all?

Before any content question, check access. The following are the engines' own statements as of October 2026, not our measurements.

• OpenAI says OAI-SearchBot is used to surface websites in search results in ChatGPT's search features, and recommends allowing it in your site's robots.txt file. It adds that its systems can need roughly 24 hours to adjust after you update that file.
• Perplexity says PerplexityBot is designed to surface and link websites in search results on Perplexity, and recommends allowing it in robots.txt.

A page that returns a normal response is not thereby indexed or used. The crawler guide covers each bot and the trade-offs of allowing or blocking.

What on-site facts should you keep current?

Our data shows that brand domains appear in source lists. It does not show which facts an assistant reads or whether a page changed a recommendation. The checklist below is reasoning from how answers are built: the engines search the web and read what the search returns, so a page that states a fact plainly can supply it, and a page that is stale or vague can supply the wrong one.

The product page audit goes through these fields one by one, and the structured data post covers markup. No public evidence we know of shows that markup alone changes what an assistant recommends.

How can you check your own site in the source lists?

1. Write four questions for your category that name no brand: best overall, a budget with a price ceiling, premium, and a specific use.
2. Ask each on each engine you can reach, in clean sessions, at least three times, and save every source list.
3. Search each list for your domain and its variants: your main domain, a store subdomain, country domains and any parent company domain.
4. Count answers, not links, with your domain in the list, and write the total next to the number of answers you collected.
5. Look at your server logs for the user-agent names OpenAI documents (OAI-SearchBot) and Perplexity documents (PerplexityBot) to see whether the bots reach you.

With few answers, one more or one fewer changes the share a lot. The answers needed post shows how wide the range is at small numbers of answers. The manual method gives a spreadsheet layout.

What these numbers cannot tell you

• A domain in a source list does not show that it shaped the recommendation. Perplexity's list is what its search retrieved.
• The data is the API, not the app. The apps can differ in personalization, memory, location and interface.
• In the October 2026 edition we sampled answers to prompts we wrote, once per prompt per engine. We did not observe shoppers.
• The figures are the October 2026 edition across the US, Mexico and Canada, with about 10 answers per category. See the methodology.

Frequently asked questions

If my site blocks these bots, will I still be recommended?

We did not test it. Our data records which pages appeared in source lists, not what the engines could have read. OpenAI recommends allowing OAI-SearchBot to help your site appear in its search results, and Perplexity recommends allowing PerplexityBot in robots.txt.

Does a high own-site share mean brand sites matter more than review sites?

Not on this evidence. The rule only looks for a brand's own domain, and a list can hold many other domains. Perplexity's retrieved lists held a mean of 15.2 distinct domains, and editorial review sites were common among them. A domain's presence in a list does not show how much it mattered.