Review sites and AI recommendations: which platforms engines cite, and how to earn reviews by the rules

Short answer: review sites shape AI recommendations in two ways. Engines cite them as sources, and they also send buyers to them by name ("check Zocdoc reviews", "compare on Tripadvisor") instead of naming a business. In our September 2026 benchmark of five AI models, about a third of answers in hotel, restaurant and dentist categories pointed the buyer to a directory or review site, while almost no answers did in skincare or project management software. Which review sites matter therefore depends on your category, and the work is the same everywhere: be listed and accurate on the platforms engines actually cite for your buyers' questions, collect reviews from every customer in the way each platform allows, and never buy, filter or fake them. Each major platform publishes its own rules, and in the US the FTC's rule on reviews makes the worst tactics illegal.

This guide is for owners and marketers who want a practical answer to "which review sites, and how", grounded in what engines cite and in what the platforms themselves permit.

• Review sites and AI recommendations are tightly linked in local categories. Hotels, restaurants and dentists saw the highest share of answers pointing to directories; software and skincare the lowest.
• The cited platform differs by category. Tripadvisor led for hotels and restaurants, Zocdoc for dentists, Avvo for lawyers, Angi for home services, ClassPass and Yelp for gyms.
• Buyers check. In BrightLocal's 2026 survey, 88% of people who used AI for local recommendations said they fact-check the reviews AI cites; in Yext's survey only 5% went straight from an AI answer to a purchase.
• The rules differ by platform. Google and Trustpilot let you ask every customer but forbid incentives and selective asking; Yelp asks you not to request reviews at all; G2 allows small incentives if they are disclosed to G2, not tied to the opinion and capped at $100.
• Accuracy beats volume. A complete, consistent profile with a steady flow of genuine reviews is what engines and buyers can use.

How review sites feed AI recommendations

It is tempting to think of review sites as one input among many. In practice they reach AI answers through three distinct routes, and each route suggests different work.

Route 1: cited as a source

When an engine searches the web to answer "best hotel near South Beach for families", review and listing pages are natural matches: they contain the place name, the location, the price range and dozens of first-hand descriptions. The engine retrieves them, reads them and cites them. Your listing page on that platform is then part of the evidence, and what it says (category, location, photos, recent reviews, the owner's responses) is what the engine summarises.

Route 2: named as the place to check

Many answers do not name a business at all. They tell the buyer how to choose and where to look: "read recent Google and Zocdoc reviews", "filter on Tripadvisor for family rooms". Our benchmark records this as an answer that "points to directories", and in some local categories it was common. In those answers, the recommendation is delegated to the platform. If you are not listed, or your listing is thin, you are invisible at the exact moment the buyer follows the advice.

Route 3: absorbed into what the engine knows

Ratings, review snippets and listing details also reach engines through search indexes, maps data and licensing, without any visible citation. The route depends on the platform and the engine, and it is not fully documented. One clear signal comes from BrightLocal's research: its CEO Myles Anderson argues that ChatGPT and other AI tools cannot see inside Google's walled garden of reviews, so a reputation that exists only on Google is effectively invisible to them, and recommends a "Reviews Everywhere" strategy across the sites customers and AI models use.

What an engine can read on a review page

It helps to look at your own listing the way a retrieval system does: as a page of text to be matched against a question. The parts that carry meaning are predictable.

• The header block: business name, category, neighbourhood or service area, price band, hours. This is what decides whether the page matches "family-friendly hotel in Miami Beach" at all.
• The rating summary: the average and the number of reviews. A buyer reads this first; an engine may repeat it.
• The review text: the sentences that mention specific things, such as a dish, a treatment, a fault fixed or a wait time. These are what make a page match a specific question rather than a generic one.
• Owner responses: your words on the page, often the only place where corrections and current facts appear next to old complaints.
• Q&A and attributes: on platforms that support them, short structured facts such as "wheelchair accessible" or "accepts walk-ins".

Most of these you influence only indirectly, through service and through asking every customer. The header block, attributes and responses you control directly, and they are where most profiles fall short.

Access is a platform decision, not yours

Review platforms set their own crawler rules, and they differ sharply. As read on 2026-10-23:

Two lessons follow. First, a robots.txt rule does not settle whether a platform gets cited: in our benchmark, yelp.com was still cited 19 times in the fitness category and 6 times for real estate agents, mostly by Perplexity, despite its restrictive file, so engines evidently reach such pages by other means, such as search indexes. Second, because access differs, spreading your reputation across several platforms is a hedge against any one of them being invisible to a given engine. These files change often; check the current version before drawing conclusions, as our guide to local GEO recommends for any crawler question.

Which categories send AI recommendations to review sites

Our benchmark gives a category-by-category picture. On 2026-09-07 we asked 5 AI models 20 buyer questions in each of 12 categories, local categories with a named sample city, and recorded whether each answer named specific providers, pointed the buyer to a directory or review site, or both. Directories, marketplaces and review sites were never counted as providers.

The split is clear. Six of the seven local categories had at least 17 answers in 100 pointing to a directory or review site; of the five product and software categories, only furniture and pet supplies reached that level, often through retailers or Consumer Reports. In hotels, 32 of 99 answers did so, in restaurants 30 of 98, and in dentists 29 of 100. In project management software, none of the 100 answers did, and in skincare one.

Which platforms were cited, by category

The second view is citations: which review and directory domains the engines linked to. In local categories one platform usually dominated.

A few caveats matter. These are the top 30 cited domains per category, so smaller platforms are not listed. The same query in another city may favour different platforms. And not one cited URL in this data was a Google Maps listing or Google review page, which fits the "walled garden" point above: the engines did not link to Google's reviews, even though buyers rely on them heavily.

Engines differ, too

Grok pointed buyers to directories in 71 of its 240 answers, ChatGPT in 45, Claude in 39, Gemini in 29 of 228 and Perplexity in 28. Perplexity named directories least often in its text but cited them heavily as sources: in hotels, 28 of Tripadvisor's 49 citations came from Perplexity. So an engine can rely on a review site without telling the buyer to go there, and another can send the buyer there without citing it. Both behaviours reward the same thing: a complete, accurate, well-reviewed profile on that platform. The full per-category tables are on each industry page, for example hotels and dentists.

Reading these numbers for a category we did not test

If your category is not one of the twelve, the pattern still gives you a starting hypothesis. Categories where buyers choose a provider they will meet in person, where quality is hard to judge in advance and where an established review platform exists (clinics, trades, hospitality, professional services) behaved like our local categories: many answers delegated the choice to a review site. Categories where buyers compare products on specifications, or where vendors publish detailed comparison content (software, cosmetics, many consumer goods), behaved like our product categories: answers named brands or pointed to publishers instead. Furniture and pet supplies sat in between, with 27 of 98 and 23 of 100 answers pointing to a directory or review site, often a retailer or Consumer Reports. Treat the hypothesis as something to test with your own prompts, not as a conclusion.

What buyers do with AI recommendations and reviews

An AI answer is rarely the last step. Two independent 2026 surveys describe what happens next, and both point back to reviews.

BrightLocal: AI is now a mainstream local tool, and people verify

BrightLocal's representative panel of 1,002 US adults found the share of consumers using AI tools for local business recommendations rose from 6% in 2025 to 45%, making AI the third most used tool for local recommendations behind Google and Facebook, while use of Google reviews slipped from 83% to 71%. Among the 455 respondents who had used AI for local recommendations, 88% said they fact-check reviews cited by AI tools, and 97% said they sometimes double-check AI recommendations against real reviews. ChatGPT was used by 31% for business recommendations in the past year and Google AI Mode by 23%.

Yext: very few people buy straight from the answer

Yext's 2026 survey of 3,848 adults who search for local businesses, which Yext says should be treated as directional because it is self-reported, found that 42.7% had used an AI tool for local search in the past month and 28% had tried a new local business because of an AI recommendation in the past six months. Its most useful finding is what people did after the recommendation.

Read together, the surveys describe a loop. AI narrows the options, then buyers check: they search, visit the business's site, open the cited sources and read reviews. Review sites therefore matter twice, once as an input to the answer and once as the place where the buyer confirms it. A business with a strong AI mention and a weak or stale review profile loses at the second step.

What this means for the page the buyer lands on

Combine the two surveys with the benchmark and a simple sequence emerges. The engine names you or sends the buyer to a platform; the buyer then searches, visits your site, opens the cited sources or reads reviews. Each of those landing points needs to confirm the same story: the same name, the same location, the same specialities, recent reviews that mention what the buyer asked about, and responses showing that problems get fixed. A mismatch at any of them, such as different opening hours on your site and your listing, or a top review complaining about something you fixed a year ago without a reply, costs you the buyer at the moment of verification. That is why the profile audit and the responses below come before any push for more reviews.

Which review sites to prioritise for your category

Few businesses can be excellent on every platform. A short, evidence-based way to choose:

Step 1: see which platforms engines cite for your questions

Run 20 to 30 of your buyers' questions, without your brand name, in the engines your buyers use. Export the cited URLs and list the review and directory domains, with how often each appears. Note also which platforms the answers tell buyers to check. This is your evidence list, and it usually differs from the list of platforms you happen to be on.

Step 2: add the platforms your buyers use

Engines are not the only audience. Buyers verify on the platforms they trust, and surveys consistently put Google reviews at the top of that list for local businesses. Even though engines outside Google did not cite Google's review pages in our data, the BrightLocal finding that most people double-check AI recommendations against real reviews makes your Google profile essential.

Step 3: check each platform's rules before planning

A platform where you are not allowed to ask for reviews (Yelp) needs a different plan from one where you may invite every customer (Google, Trustpilot). The rules are in the next section.

A starting map by category type

Auditing a profile in thirty minutes

Before collecting a single new review, audit what is already there on each priority platform. A quick, repeatable checklist:

1. Claim and verify. Is the listing claimed by you, with working owner access, and is there only one listing for each location? Duplicates split reviews and confuse both buyers and engines.
2. Header facts. Name exactly as on your site and signage; correct category; address or service area; phone; website link to the right page; hours including holidays.
3. Attributes and services. Every option the platform offers that is true of you, such as accessibility, payment types, services, specialities, insurance accepted or languages spoken.
4. Photos. Current exterior, interior and product or service photos, so the page reads as active.
5. Review health. Date of the most recent review, the number in the last 90 days, and any unanswered reviews from the last six months.
6. Consistency. Compare the header facts across platforms and with your own site. Every mismatch is a reason for an engine to hesitate or for a buyer to doubt.

Record the results in one sheet with a row per platform and location. The gaps are your first month of work, and they usually matter more than new reviews.

What each platform allows when you ask for reviews

Every major review platform publishes rules about how businesses may collect reviews. They are not identical, and following one platform's rules on another can get your reviews filtered or your profile flagged.

Sources: Google's Maps user-generated content policy, Yelp's "Don't Ask for Reviews" page, Trustpilot's guidelines for businesses and G2's community guidelines, all read on 2026-10-23.

The legal floor in the US

Above the platform rules sits the law. The FTC's Rule on the Use of Consumer Reviews and Testimonials (16 CFR Part 465) makes it a violation for a business to write, create or sell fake reviews; to give compensation or incentives conditioned on a review expressing a particular sentiment, positive or negative; for officers or managers to post reviews without disclosing their relationship; and to suppress reviews with groundless legal threats, intimidation or false accusations. It also covers your own site: showing only favourable reviews while implying they are all of them is a violation when reviews are suppressed for their rating or negative sentiment. Generalized requests to all purchasers to post reviews are exempt from the provisions about procuring reviews, which is why "ask everyone, the same way" is both the platform rule and the safe legal pattern.

How platforms enforce

Enforcement is automated and large-scale. Trustpilot's Trust Report 2025 says it removed 4.5 million fake reviews in 2024, 7.4% of all reviews submitted, with 90% removed automatically, out of more than 61 million reviews written that year. It also says the vast majority of fake reviews it removes are positive. Yelp's support page says its recommendation software may not recommend reviews that look prompted by the business. A review that is filtered or removed cannot help you with buyers or engines.

A compliant review programme, step by step

The programme below works on Google, Trustpilot and similar platforms that allow invitations. For Yelp, skip the asking steps and keep the profile and response steps.

1. Choose one trigger and apply it to everyone

Pick a single point in the customer journey, such as two days after delivery, after an appointment or at checkout, and invite every customer who reaches it. Trustpilot's guidelines describe exactly this: invite everyone in the same way and at the same point, whether their experience was good or bad. Automating the trigger from your order or booking system is the simplest way to prove it is not selective.

2. Write a neutral invitation

The wording must not hint at the rating you hope for. Two templates:

Subject: How was your visit to [Business]?

Hi [first name],
Thanks for choosing [Business] on [date]. If you have a minute, we'd
value an honest review of your experience, good or bad:
[link to your profile on the platform]
It helps other people decide and helps us improve.
[Name], [role]

---
SMS: Thanks for visiting [Business]. We'd appreciate an honest review
of your experience: [short link]. Reply STOP to opt out.

What to avoid in the wording: "if you loved us", "5-star", any mention of a discount, a prize draw or a thank-you gift, and any routing that sends unhappy customers to a private form and happy ones to the public platform. That last pattern, often called review gating, is exactly what Google, Trustpilot and G2 prohibit.

3. Make the review easy, not guided

Link directly to the review form on the platform. Do not supply suggested text, star ratings or talking points. Reviews in customers' own words, with the details that mattered to them, are more useful to the next buyer and carry the specifics an engine can extract: the dish, the treatment, the room, the response time.

4. Respond to every review, especially the bad ones

Responses are public, permanent and read by both buyers and engines. Thank, address the specific point, explain any change, and offer a direct contact for unresolved problems. Never argue, never reveal a customer's personal details, and never threaten a reviewer: the FTC rule specifically lists groundless legal threats and intimidation used to remove or prevent reviews as a violation.

A negative review, worked through

Suppose a review on your profile says a booked appointment was cancelled at the last minute and nobody called. A response that works for the reviewer, the next buyer and an engine reading the page has four parts: acknowledgement, the specific fact, what changed, and a direct route to resolution.

Thank you for telling us, and we're sorry. Your appointment on [date]
was cancelled because [brief, true reason], and we should have called
you. Since then we've [specific change, e.g. "added same-day text
alerts for any schedule change"]. If you'd like to rebook, please
contact [name] directly at [phone or email] and we'll prioritise it.
[Name], [role]

What it avoids matters as much: no argument about whether the review is fair, no mention of the customer's private details, no offer of compensation in exchange for changing the review (Google's policy specifically lists incentives for revising or removing a negative review as fake engagement), and no copy-paste text used on every review. If the review is genuinely fake, for example from someone who was never a customer, use the platform's reporting process rather than the public reply, and say nothing in public that accuses a real person of lying.

5. Keep the profile itself accurate

Engines read the profile as much as the reviews: name, category, address, service area, hours, prices, photos and the business description. Update these whenever something changes, and keep them identical across platforms. A recommendation that sends a buyer to a closed location is worse than no recommendation.

6. For G2 and similar B2B platforms

Use the platform's own campaign tools. If you use incentives, follow G2's guidelines: the incentive must not depend on the opinion, must be disclosed to G2 so the review is labelled, and must not exceed $100 in value. Invite all active users of the relevant product, not a hand-picked list of champions.

Multi-location businesses

For chains and franchises, the unit that matters is the location. Engines answer "dentist near Zilker Park", not "dentist brand", and the cited page is usually a location's listing. Give each location its own claimed, complete profile; trigger invitations per location so each builds its own review history; and make sure the location pages on your site match each listing exactly. A strong brand-level rating does little for a location whose own profile has a handful of old reviews.

B2B software: G2, Capterra and similar

Software buyers in our data were rarely sent to directories, and the cited sources were mostly vendor sites and comparison publishers, with capterra.com the only review platform among the most-cited domains. Review platforms still matter for verification and for the comparison articles that cite them. The practical points: keep your category placement and feature data accurate, because these drive which comparisons you appear in; run campaigns through the platform's own tools; invite all active users, not only champions; and if you offer incentives on G2, follow its rules on disclosure, labelling and the $100 cap.

Making reviews useful to AI engines

Within the rules, a few choices make your review presence more useful for AI recommendations without changing what customers say.

Recency and steady flow

A profile with a burst of reviews two years ago and silence since looks stale to buyers who verify. A steady flow from an automatic, non-selective trigger keeps the profile current. It also produces the pattern platforms' detection systems expect from genuine reviews, rather than the spikes associated with campaigns.

Specific reviews come from specific questions

You may not tell customers what to write, but a neutral prompt can invite detail: "What did you come in for, and how did it go?" is neutral and tends to produce reviews that mention the service, which is what engines match to buyer questions such as "dentist for a cracked tooth in Austin".

Owner responses as structured information

Your responses are one of the few places on a review platform where you can state facts in your own words: that you now open on Sundays, that parking is free, that the issue a reviewer raised was fixed. Keep them factual, short and consistent with your own site.

Connect the profiles to your site, not the other way round

Link to your main review profiles from your site, and include them in your Organization or LocalBusiness markup sameAs list so engines can tie them to the same entity. Do not copy third-party reviews onto your own pages and mark them up as if they were yours; showing a selection of reviews on your site while implying it is all of them can fall foul of the FTC rule on review suppression described above. Our guide to third-party corroboration covers the wider picture of sources you do not control.

When a platform shows a fact you cannot edit

Some details on review platforms come from the platform, from users' suggested edits or from data partners, not from you. When a wrong fact persists, use the platform's suggest-an-edit or support route with evidence, such as a utility bill or business registration for an address, and keep a record of the request. Meanwhile, state the correct fact clearly on your own site and in your owner responses where relevant, so that an engine comparing sources finds more agreement on the right version than on the wrong one.

What not to do with review sites

The incentives to cheat are obvious and the market for cheating is large. Every tactic below is prohibited by at least one platform, and most by the FTC rule.

There is also an AI-specific reason to avoid all of this. Engines and buyers increasingly read review profiles critically, and BrightLocal's finding that 88% of AI users fact-check cited reviews means a manipulated profile is likely to be examined at exactly the moment you need it to hold up. A Trustpilot consumer warning on your profile, or a cluster of removed reviews, is the kind of detail that ends up in an answer.

If you inherited a bad history

Some businesses discover that a previous agency or employee bought reviews, ran a gated survey or posted reviews from staff accounts. The instinct is to quietly delete or replace them. A safer sequence:

1. Stop the practice immediately and document what was done, when and by whom.
2. Do not try to bury the problem with a sudden wave of new reviews; spikes are exactly what detection systems look for.
3. Where the platform offers a way to report or withdraw reviews you know to be inauthentic, use it.
4. Restart collection with a single, non-selective trigger and let the profile rebuild at a natural pace.
5. Take legal advice if the practice may fall under the FTC rule or similar rules where you operate.

A profile with fewer, genuine reviews is a better asset than one that may be flagged or carry a consumer warning, which on Trustpilot hides the TrustScore and closes the profile to new reviews.

How to measure what review sites do for your AI recommendations

Review metrics such as star rating and review count describe the platform, not the answers. Measure both, and connect them through your prompt set.

On the platforms

• Profile completeness and accuracy on each priority platform, checked monthly against a fact sheet.
• Reviews per month and the share of customers invited who left a review, which shows whether the trigger is working for everyone.
• Response rate and median response time to reviews.
• Reviews filtered, removed or flagged, as an early warning.

In the answers

• For your buyer prompts: how often each review platform is cited, and whether your own listing on it is among the cited pages.
• How often answers point to directories, and whether you appear when a buyer follows that advice (check the platform's own search for the same query).
• Mention rate and the accuracy of what answers say about you: rating, location, specialities.

A monthly dashboard

Expect slow, cumulative change

Reviews accumulate slowly by design, and engines pick up changes on their own schedules. Re-run the same prompt set monthly and look at the trend over a quarter, not at single runs. If the cited platforms for your category change (for example, a booking site starts appearing where a review site used to), adjust your priority list rather than spreading effort evenly. For background on how AI discovery is changing small-business traffic overall, see how AI recommendations are reshaping small-business traffic; for tracking which review platforms engines cite for your own prompts over time, our citation sources report lists them by prompt and engine.

Sources

• BrightLocal, Local Consumer Review Survey 2026: AI and trust
• Yext, 2026 Consumer Search Behaviors Report
• Google, Maps user-generated content policy: prohibited and restricted content
• Yelp, Don't Ask for Reviews
• Yelp, Content Guidelines
• Trustpilot, Guidelines for businesses
• Trustpilot, Trust Report 2025
• G2, Community Guidelines
• FTC, 16 CFR Part 465, Rule on the Use of Consumer Reviews and Testimonials
• Yelp robots.txt, Tripadvisor robots.txt, Trustpilot robots.txt, G2 robots.txt (all read 2026-10-23)
• InsightWonder industry benchmark, 12 categories, 5 models, 20 questions each, collected 2026-09-07 (first-party measurement; method on each industry page)