Wirecutter in AI shopping answers: Perplexity's sources vs OpenAI's links

One domain shows a stark gap between the two engines in our October 2026 source lists: nytimes.com, where every URL we saw was a Wirecutter page. This post gives the counts, explains why the two engines' numbers cannot be compared like for like, and says what a brand can and cannot take from them.

How often does Wirecutter appear?

In Perplexity's Agent API with the Sonar model (perplexity/sonar), a Wirecutter page was among the retrieved sources of 333 of 787 answers, which is 42.3%. With only 787 answers, the true rate for this kind of prompt could plausibly sit anywhere from about 38.9% to 45.8% (a 95% Wilson interval). It was in the lists for 130 of the 160 categories, and the 333 answers held 558 Wirecutter URLs, about 1.7 per answer that had one. Those 558 URLs are about 2.6% of the roughly 21,500 URLs retrieved across the 787 answers (a mean of 27.3 per answer), so the domain is present in many answers while being a small share of all the URLs.

In OpenAI's API with web search (gpt-6-luna), none of the 800 answers linked nytimes.com. A count of zero in 800 still leaves room: the Wilson interval puts the true rate for these prompts anywhere up to about 0.5%.

For scale, within Perplexity's lists the next most frequent domain is amazon.com, in 265 answers (33.7%). Within OpenAI's links, the most frequent domain is rtings.com, in 47 answers (5.9%).

Why are the two numbers not the same kind of count?

They are recorded differently.

For OpenAI, we record the pages its answer links. The API attaches an annotation to each linked page, with its URL, title and location in the text. As of October 2026, OpenAI's API can also return the full list of URLs the search consulted, and OpenAI's documentation says that list is often longer than the list of linked pages. Our October sample records the linked pages, so a zero means no Wirecutter page was linked in an answer. It says nothing about whether OpenAI's search looked at one.

For Perplexity, we record every page its search retrieved, with rare exceptions. The API returns a list of search results alongside the answer, and it does not say which of them the answer relies on. Perplexity's documentation also says that whether the answer text carries numbered markers depends on the prompt. Our prompts are plain shopper questions that do not ask for markers; when an answer carries them anyway, we keep only the pages they point to, which is why the methodology says "almost always".

So the 42.3% means a Wirecutter page was among the pages retrieved for the question. It does not mean the answer used it or quoted it. And 0 of 800 means none was linked. Putting them side by side shows an engine difference in what the records contain as much as in what the engines do, and we do not know how much of each.

Where does it appear, by market?

The Perplexity counts in the October 2026 edition, split by market, look like this. OpenAI contributes no Wirecutter answers in any market, so these are Perplexity's counts.

This is descriptive. The three markets have different categories, Mexico's prompts are in Spanish and the mix of categories differs, so the gaps should not be put down to language or location. The numbers are what they are: in this sample, Wirecutter pages turned up in more than half of the US Perplexity answers and in about one in nine of Mexico's.

What is Wirecutter, and what does it say about how it picks products?

Wirecutter's recommendations live at nytimes.com/wirecutter. The New York Times Help Center, which covers Wirecutter, says that Wirecutter independently tests and reviews thousands of products, and that it decides which products to feature on editorial and product-testing standards, not on affiliate deals or advertising relationships. It also says that Wirecutter may earn a commission when a reader buys through a link, and that its writers and editors are never made aware of any business relationships with retailers.

Taken at its word, that points away from a paid route to a Wirecutter recommendation. If your category has a Wirecutter guide, it is worth knowing which products it names and why, but not worth looking for a shortcut.

What can a brand take from this?

• In Perplexity's retrieval, a Wirecutter page appears in 42.3% of answers. Editorial review sites can be present in many answers even though one domain is a small share of the URLs retrieved, and Wirecutter was the most frequent domain across Perplexity's 787 answers. If you sell in a category it covers, find out what the guide says about products like yours.
• Do not generalize to OpenAI. The pages OpenAI linked came from other domains: rtings.com, rei.com, tomsguide.com and wired.com led its list, each in 3.3% to 5.9% of answers. A plan built on one engine's source pattern may not transfer.
• A source pattern is not a recommendation rule. We did not test whether a Wirecutter page in the list changes which brands an answer names. We would not read the gap between 42.3% and 0% as a cause of anything.
• Our advice is dull, and untested. Wirecutter says it independently tests and reviews products. Making your product easy to buy and test, with accurate public specifications and normal retail availability, is our own advice, not something Wirecutter has said it requires, and this data does not show that it leads to a Wirecutter review or an AI recommendation.

How can you check your own category?

1. Find the Wirecutter guide for your category and note which brands it names.
2. Write five prompts a shopper in your category might type: the overall best, a pick under a price cap, a premium pick, a question about one particular need, and a request for buying advice.
3. Ask Perplexity each question several times on different days and keep the list of sources for every answer.
4. Count the answers where a nytimes.com/wirecutter URL appears, and divide by the answers asked. That is your Wirecutter share, with its denominator.
5. Compare the brands in those answers with the guide. This is a check, not a test of cause: you are looking at whether the answers resemble the guide.
6. Repeat with OpenAI and look at the pages it links, which will come from a different set of domains. The free manual method covers the wording and the sample size.

What this data cannot tell you

• It is the API, not the app. The consumer apps can use different search tools and show different sources. See what API sampling can and cannot tell you.
• About 5 Perplexity answers per category. In the October 2026 edition each category has about 10 answers across both engines, so about 5 from Perplexity, and every Wirecutter count here is Perplexity's. Wirecutter's presence in a given category is an example, not a measure. Across all 160 categories it is a pattern.
• One month, fixed prompts. Each prompt was asked once per engine in October 2026.
• No cause. We did not measure why OpenAI links no Wirecutter page or why Perplexity retrieves so many, and we do not guess.

For all domains side by side, see the websites in AI shopping answers. The edition's other headline numbers are in the October 2026 pillar post.

Frequently asked questions

Can a brand pay to be in Wirecutter?

The New York Times Help Center says Wirecutter chooses products on editorial and product-testing standards, not through affiliate deals or advertising relationships. It also says Wirecutter may receive a commission when a reader buys a recommended product through its links. Check Wirecutter's own pages for how it works with brands.

Does a Wirecutter page in Perplexity's sources mean the answer used it?

Not necessarily. Perplexity's API returns the pages its search retrieved without saying which of them the answer relies on, and we almost always record that whole list, so a Wirecutter page in the list shows the search found it, not that the answer drew on it.