Getting into the review sites that appear in AI shopping answers

The question behind "how do I get reviewed" is really "which sites, and how do they choose". Most advice skips both steps and goes straight to a pitch template. Here the order is reversed: find the sites your category's AI answers actually draw on, per engine, read how each one says it picks products, and only then decide what to send. The data below comes from the October 2026 edition; the policies were read on October 28, 2026, and they change.

Which review domains appear, and in which engine?

The data is 800 answers from OpenAI's API with web search (gpt-6-luna) and 787 from Perplexity's Agent API with the Sonar model (perplexity/sonar), across 160 categories in three markets. The lists below record different things: for OpenAI, pages linked in the answer; for Perplexity, pages its search retrieved, some of which the answer may not use. Presence in a list does not show influence, so the two engines are read separately.

Among OpenAI's 30 most frequent domains, 13 are publishers. Among Perplexity's 30, 19 are. The reviews and ratings post explains how we sorted publishers by hand. The tables show ten publisher domains for each engine, from the most frequent downward (outsideonline.com ties nbcnews.com at 8 answers in OpenAI's list and is not shown), with the number of categories in which the domain appeared at all.

OpenAI, pages linked in answers (797 answers with at least one link)

Perplexity, pages its search retrieved (787 answers)

October 2026 edition. The lists span the US, Mexico and Canada. Categories are the number of the 160 in which the domain appeared in at least one answer's list. Each engine on its own denominator.

Two patterns are worth taking into planning, both descriptive. The first is breadth. Within Perplexity's retrieved sources, nytimes.com appeared in answers in 130 of 160 categories, while outdoorgearlab.com appeared in 34 categories, treelinereview.com in 31 and rtings.com in 30. OpenAI's links are far fewer per answer (a mean of 2.2, against 27.3 retrieved URLs for Perplexity), so category counts are not comparable across the two engines. A specialist site covers a narrower set of categories, so for a niche product it may matter more than the broad name. We cannot say that from this data, only that it is plausible and worth checking in your own category.

The second is that the engines share names but not weight. Seven publisher domains are in both top 30 lists, and rtings.com is first among OpenAI's publisher domains at 5.9% of answers but seventh among Perplexity's at 13.9%. Targeting a site because it is "in AI answers" needs a statement of which engine.

How do these sites say they choose products?

We read the published statements of five sites that appear in these lists. They are the sites' own words about themselves, not our verification of their practice.

The pattern is consistent. For three of the five (RTINGS, Consumer Reports, OutdoorGearLab), the stated route is that the site buys the product, and for Wirecutter and TechRadar the pages we read do not describe a sample or pitch process. None of the statements we read says payment or affiliate terms get a product in. That narrows what a brand can do without crossing a line:

• Do not pay for editorial coverage presented as independent. Four of the five (Wirecutter, Consumer Reports, TechRadar and OutdoorGearLab) say in the statements above that payment, affiliate deals, advertising, sponsorships or paid endorsement do not decide what they cover or are not accepted. The RTINGS statements we read do not address it.
• Do not count on unsolicited samples. At three of the five, the site says it buys its own. A free unit is not the route there, and sending one may do nothing.
• Make the product easy to buy and test. If a site buys at retail, the product has to be in stock at a retailer it can use, with the model name, size and configuration stated clearly, and with a clear warranty and return policy.
• Make verification easy. Publish specifications, test data, certifications and manuals on your own site, so that a reviewer or an assistant checking a claim finds one consistent answer. The product pages audit covers the page side.
• Know the routes a site publishes. RTINGS says it chooses the TV it buys every 60 days based on reader votes, so for TVs, owners who like a product can vote like any other reader. Do not organize votes or offer anything in return for them. Other sites in the table do not describe a suggestion route in the statements we read.

What is the step-by-step plan?

1. Pick three to five prompts for your category. Use best overall, budget, premium and a specific use case, written the way a shopper would ask. Each category page in the index lists the prompts the index uses.
2. Run them in each engine, several times, in clean sessions. Record the source URLs, per engine, and tally the domains. Keep the ones that publish product tests or roundups.
3. Read each site's policy. Look for three facts: how it gets products (buys, borrows, accepts samples), how it chooses what to cover, and where it takes suggestions or corrections. Put them in a sheet beside the domain's tally.
4. Sort the list into routes. Buys its own: work on retail availability and verifiable facts. Accepts samples or loans, where a site says so: follow its submission rules exactly. Roundups and gift guides: find the writer and the page's last update date, and pitch before the next update, not after.
5. Build one press kit. Product facts and specifications, price and availability by market, test results and certifications, images that can be used, warranty and return terms, a named contact, and what you can supply on request.
6. Pitch with relevance, not volume. Say in two or three sentences which question the product answers better than the page's current picks, and give a verifiable reason. Do not offer payment or links in return for inclusion, and do not make a free or discounted unit conditional on a favorable review.
7. Disclose what you give. If you provide a free or discounted product, the FTC says the reviewer's connection to the company should be disclosed everywhere they endorse it. Put that in your outreach. A site that says it buys everything may decline the unit, and that is fine.
8. Do not build a fake independent site. The FTC's consumer reviews rule prohibits a business from misrepresenting that a website or organization it controls gives independent reviews or opinions about a category of products or services that includes its own.
9. Resample. Ask the same prompts again after a placement appears, with the same number of answers, and read the change with an interval. The share of answer post shows how. A new placement and a changed answer can coincide without one causing the other.

What does this not tell you?

Being on these sites will not, on this evidence, get your brand recommended. We have not measured the effect of a review on an answer, and the tables show where sources appear, not what they did. We have no data on how long a site takes to review a product or how often a roundup is updated, and we left those out because we could not verify them. The apps can also differ from the APIs in what they show. Treat a review as one input you can earn, alongside the retailer pages in the retail listings post and your own site.