Website reviews or Google reviews: which ones do AI shopping tools read?

Both, through different doors. Google reviews feed Google's AI. The reviews on your product pages feed ChatGPT and the other assistants, on one condition: the AI has to be able to read them, and there is a good chance it cannot.

Picture the review you were proud of last month. A customer loved the product, wrote five generous sentences on your product page, added a photo. You did everything the advice says to do.

Weeks later, a shopper asks ChatGPT to recommend the best version of exactly what you sell, and your store never comes up. Nothing is broken and the review is real. The machine that wrote the answer may never have seen it.

This post explains which reviews reach which AI tools, why the answer is different for Google reviews and website reviews, and how to check your own product pages in about five minutes.

Why do AI shoppers matter to a small online store?

Because they arrive ready to buy. Adobe measured traffic arriving at US retail sites from AI tools and found it grew 393% in the first quarter of 2026 compared with a year earlier. These are shoppers who asked a chat tool what to buy and clicked through to a store the answer named.

Those visitors behave differently. By March 2026 they were converting 42% better than other traffic, a full reversal from a year earlier when they converted worse. The comparing happened inside the chat, and whoever the answer names receives a shopper who has already decided.

The shopper an AI never sends you does not show up in any report. She never bounced, because she never arrived.

That is why this question deserves a sharper answer than "collect reviews everywhere." Your review effort is a limited resource. It should go where machines and humans both benefit.

Which reviews does ChatGPT use when it recommends products?

ChatGPT builds its shopping answers from structured product data, meaning organised facts about each product such as price, availability, and description. OpenAI states the tool weighs price, reviews, and ease of use when deciding what to show. Reviews are in the decision, straight from the company that builds it.

The review summaries ChatGPT shows shoppers are generated from reviews on public websites. Public is the load-bearing word: it means pages its crawler can read, not reviews on a famous platform. Your product pages qualify the moment they are readable.

There is a second door. OpenAI publishes a product feed standard, the file a store sends so ChatGPT holds its catalogue directly, and it treats ratings, review counts, and customer Q&A as data worth including. Your review data is turning into copy a machine reads.

Notice what is missing from both doors: neither needs your reviews to live on Google, or on any review platform. The reviews you already collect on your own site are exactly the raw material. The question is whether the machines can get at it.

Can AI tools read the reviews on your product pages?

Often, no. Most review apps display reviews with JavaScript, code that assembles part of the page in the shopper's browser after the page loads. Your customer sees the reviews appear; a machine that reads the page as first delivered sees the empty container they were meant to fill.

That difference decides everything here. Vercel, a company that hosts a large share of ecommerce sites, analysed roughly a billion requests from AI crawlers and found that none of the major ones run JavaScript. ChatGPT's crawler, Claude's, and Perplexity's all read what the server sends and nothing more.

A review widget can hold two hundred five-star reviews and still hand an AI crawler an empty box. The reviews exist for humans and not for machines.

Google is the one exception. Google's crawler does run JavaScript, which is how your reviews can appear in a Google search result while staying invisible to every chat tool. Ranking well on Google tells you nothing about what ChatGPT can see.

The problem lands hardest exactly where your reviews live. Adobe's testing found product pages are the worst-read page type on retail sites, with about a third of the average product page unreadable to AI tools. The pages that carry your reviews, your sizing answers, and your buy button read worst of all.

The five-minute check for your own store

You do not need a tool for this. Open one of your product pages in a browser, right-click, and choose View Page Source. You are looking at the page the way a crawler receives it.

Now search the source, with Ctrl+F or Cmd+F, for a distinctive phrase from a recent review. You are hunting for three things: the words of the review itself, the word aggregateRating, and the review dates. If the phrase is there, machines receive your reviews and you are in better shape than most stores.

If the phrase is missing, no AI tool has ever read a single review your customers wrote. The same goes for aggregateRating, part of schema, the hidden labelling that tells machines your average rating and review count. Absent, or added only by JavaScript, means the label problem matches the review problem.

If a sentence from your newest review is not in the page source, the fix is a setting or an app change, not a rebuild.

If the check fails, send your review app's support team two questions: are our review text and review schema included in the page HTML the server sends, and which plan or setting makes that happen? Most established apps have an answer. An app with no answer is telling you something useful before your next renewal.

Where do Google reviews fit for an online store?

Google reviews live on a Google Business Profile, the free listing that appears in Search and Maps, and Google has connected that data to its AI. Its Gemini models can check a live source before they answer, and that source covers more than 250 million places, including business details and user reviews. For the questions Google's AI answers, a strong review record is real fuel.

Here is the catch that never makes it into the advice. Google's guidelines require a business customers can visit in person, or one that travels to customers. A purely online store meets neither test, which rules it out of a Business Profile altogether.

Read that as clarity, not bad news. If you run a showroom, a market stall, or a staffed pickup counter, you qualify, and sharing your profile's review link after each sale is allowed, provided nothing is offered in return. If you are online only, the standard advice was never available to you, and you can stop feeling behind on it.

Google still offers online stores two programs through Merchant Center, the free system that feeds your products into Google Shopping. Google Customer Reviews surveys your buyers after checkout and builds a seller rating, while Product Ratings can put stars on your Shopping listings once you have at least 50 reviews across your products, refreshed by a monthly feed. Both run on reviews you are already collecting.

Which reviews should you focus on?

Product reviews on your own website are the foundation for an online store. They carry the detail shopping tools are built to ingest: fit, sizing, durability, the photo of the product in a real kitchen. Your first job is not collecting more of them, it is making the ones you have readable.

Humans complicate the picture in a useful way. Consumer research from Clutch found fewer than half of shoppers trust the reviews on a brand's own website, while about two in three trust what they read on Google. Machines ingest your owned reviews happily; humans want one proof point you do not control.

If you qualify for a Business Profile, that covers it. If not, a seller rating or a modest presence on one independent platform your customers check does the same work.

Do not chase review platforms for AI citations, either. When Semrush analysed more than 100 million citations across AI tools over three months, review platforms were absent from the 25 most-cited domains. Reviews sway AI answers as trust signals attached to your brand and products, not as pages the AI quotes.

Our own early data points the same direction. In the GetRecommended.io AI Visibility Index sample from Q2 2026, drawn from local and service businesses rather than online stores, no business was invisible to AI, whatever its review count. Businesses with no Google reviews were still named by about four of the six engines we test.

The difference showed up in depth. Businesses with a healthy review base were named by almost five engines and scored around 90% on our authority measure, against roughly 65% for those with none.

Reviews rarely decide whether AI knows you exist. They decide how many engines name you, and how confidently.

One more pattern from that sample is worth keeping. A meaningful share of businesses sat on a perfect 5.0 rating, which sounds ideal and is not: Northwestern's Spiegel Research Center found purchase likelihood peaks between 4.2 and 4.7 stars, because a flawless record reads as filtered. Publish the honest three-star review, it makes the rest believable to humans and machines alike.

What are the rules on collecting reviews?

Keep the collection clean, because the legal edges are real. In Australia, the ACCC requires reviews to reflect genuine experiences. Writing fake ones, editing criticism out, or quietly removing negative reviews can breach consumer law, and incentives must be disclosed and offered equally whether the review turns out glowing or grumpy.

The penalties are not theoretical. HealthEngine paid $2.9 million in Federal Court penalties after suppressing negative reviews. Similar rules apply through the FTC in the United States and the CMA in the United Kingdom, so a neutral ask is the safe ask everywhere: every buyer invited, no reward for stars.

What can you do about it this week?

1. Run the page-source check, 5 minutes. View source on your best-selling product page. Search for a phrase from a recent review, then for aggregateRating. Both present means machines get your reviews. Either missing means they do not, and the next step is your fix.
2. Send your review app one email, 10 minutes. Copy this: "Two questions about our store. Is our review text included in the page HTML your app serves, or loaded by JavaScript afterwards? And is our review schema, including aggregateRating, in that same server-delivered HTML? If not, which setting or plan changes that?" Most established apps answer within a day, and the answer tells you whether to change a setting, a plan, or an app.
3. If you run on Shopify, connect your catalogue to ChatGPT, 15 minutes. OpenAI accepts product data straight from Shopify's catalogue, so check that the connection is switched on in your sales channels and that titles, prices, and stock are current. That file is the version of your store ChatGPT reads first.
4. Pick your Google lane, 10 minutes. Customers visit you in person? Claim your Business Profile and start sharing the review link after each sale. Online only? Open Merchant Center and switch on Google Customer Reviews, then work toward the Product Ratings threshold.
5. Fix the ask, this month. Invite every buyer after delivery rather than at checkout, send one reminder, and never attach a reward to stars. Our guide on how to ask customers for reviews covers the timing and wording that work.
6. Set a baseline. Run a free AEO scan to see how the six major engines treat your business today, make one change, and retest in a month.

One caution before you draw conclusions from any AI tool: answers vary between runs. Check twice before you celebrate or panic. What shows up twice is a pattern worth acting on.

The bottom line

This was never a choice between your website and Google. The two review piles feed different machines: Google reviews power Google's AI, if you qualify to collect them at all, while the reviews on your product pages are what the shopping assistants want. Most stores have never checked whether a machine can read them.

That check takes five minutes and costs nothing. Run it before you spend another dollar or another hour on review collection anywhere.

Then keep collecting where you own the ground. The store whose reviews are readable, honest, and steadily growing is the store the machines can vouch for. If you want to see where you stand across all six engines in one pass, run the free scan or read the common scan questions first.