A 10% increase in G2 reviews correlates with a 2% increase in AI citation rates for B2B software, per She Innovates AI's 2026 structured data analysis. Product comparison queries — "best CRM for a 10-person sales team," "top running shoes under £100," "highest-rated project management tool" — pull directly from AggregateRating schema when generating shortlists. AI engines do not read the star rating on your page and infer a number. They read the machine-declared number in your schema and use it as a filter criterion.
Review schema and AggregateRating schema serve different functions in AI retrieval and should be implemented separately rather than as a single block. AggregateRating declares the overall score and review count — the number the AI filters and ranks by. Review schema declares individual review content — the specific text AI engines extract when a buyer asks "what do users say about X?" Both matter, and both need to be in the server-rendered HTML where AI crawlers can find them.
For most brands, the bigger problem is not implementation quality — it is the self-review trap and the stale review floor. Both cause AI engines to either ignore or down-weight review schema that was technically well-constructed.
What Is the Difference Between AggregateRating and Review Schema?
AggregateRating is a summary: it declares the average rating value, the number of reviews that produced it, the best possible rating, and the worst possible rating. It is a statistical summary for machine parsing. Review is an individual review: it declares the reviewer, the text they wrote, the date they wrote it, and the score they gave.
For AI citation purposes, AggregateRating earns citations in comparison and ranking queries. Review earns citations when a buyer asks for user opinions or reported experiences. The cleanest implementation nests both inside a Product, LocalBusiness, or SoftwareApplication schema:
{
"@context": "https://schema.org",
"@type": "SoftwareApplication",
"name": "NotioncCue",
"description": "AEO tracking platform that monitors AI citations across ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews.",
"applicationCategory": "BusinessApplication",
"operatingSystem": "Web",
"offers": {
"@type": "Offer",
"price": "49",
"priceCurrency": "USD"
},
"aggregateRating": {
"@type": "AggregateRating",
"ratingValue": "4.8",
"bestRating": "5",
"worstRating": "1",
"ratingCount": "247",
"reviewCount": "247"
},
"review": [
{
"@type": "Review",
"author": {
"@type": "Person",
"name": "Sarah Chen"
},
"datePublished": "2026-06-15",
"reviewBody": "Set up took 20 minutes. Within the first week we could see exactly which prompts we were winning and which competitors were beating us on. The Perplexity tracking is the most useful — it moves fastest and gives you the earliest signal on whether content changes are working.",
"reviewRating": {
"@type": "Rating",
"ratingValue": "5",
"bestRating": "5",
"worstRating": "1"
}
}
]
}Three field requirements that differ from how most brands implement review schema:
bestRating and worstRating are required, not optional. Most implementations declare only ratingValue. Without bestRating and worstRating, AI engines cannot normalise the score — a 4.8 on a 10-point scale looks the same as a 4.8 on a 5-point scale without the context fields. Always declare all three: ratingValue, bestRating, worstRating.
ratingCount and reviewCount should match. ratingCount includes all ratings (star-only votes without written text). reviewCount counts only written reviews. For AI citation purposes, reviewCount is more important than ratingCount, because AI engines extracting "what do users say about X" need written text, not just scores. If your schema declares 247 ratings but only 12 written reviews, the discrepancy signals to AI engines that most of your rating signal comes from unverifiable votes rather than substantive reviews.
Use a period as the decimal separator, always. "ratingValue": "4.8" is correct. "ratingValue": "4,8" is invalid — Google's Rich Results Test fails it, and AI crawlers that respect Google's schema guidelines ignore it. This error appears most often on sites localised for European markets where the comma is the standard decimal separator.
What Is the Self-Review Trap and How Do You Avoid It?
Google's review schema guidelines state that pages where "the entity being reviewed controls the reviews about itself" are ineligible for star-rating rich results. In practice: a business publishing testimonials about itself on its own website in AggregateRating schema will not earn star-rating rich results. The same restriction affects AI citation confidence — AI engines applying the same guideline logic treat self-review schema as lower-trust than third-party review schema.
Two legitimate paths around the self-review restriction. First, embed third-party review platform widgets with the reviews sourced externally. If you display your G2 reviews on your website using G2's embed, the reviews are authored on G2's platform — your website is a display surface, not the review origin. The AggregateRating schema on your product page declaring these reviews is reflecting data from an independent platform, which is legitimate under Google's schema guidelines.
Second, use AggregateRating on product pages for customer product reviews (not company reviews). An ecommerce store where customers review individual products is the canonical correct use case — the store is not reviewing itself, customers are reviewing products. This is fully eligible for rich results and AI citation. The same AggregateRating schema pattern on a "testimonials about us" page is the self-review pattern that triggers the ineligibility flag.
The SEO and AEO signal from third-party review platforms — G2, Capterra, Trustpilot, G2's January 2026 acquisition of Capterra means those two review pools will increasingly be treated as a combined entity — is more valuable than any on-site review schema. Per Topify's 2026 analysis, G2 holds 23.1% citation share across B2B and SaaS queries. Five review domains account for 88% of all review-platform links cited by AI engines for software categories. Your on-site schema is the supporting layer, not the primary review signal.
What Review Content Earns AI Citations?
Not all review text is equally citable by AI engines. Three patterns produce the most extractable review content:
Outcome-specific reviews with named metrics. "Great product!" earns no citation. "After switching from Competitor X to NotioncCue, our Perplexity citation rate increased from 8% to 31% over six weeks. The setup took one afternoon." earns citations for queries about AEO tool effectiveness because it is specific, measurable, and independently verifiable. Encourage reviewers on G2 and Trustpilot to document outcomes, timelines, and specific metrics rather than general satisfaction. Your review request template makes this happen — ask specifically: "What changed after you started using us? Can you mention a specific number or outcome?"
Use-case specific reviews. A review that says "perfect for a 15-person SaaS team tracking Perplexity citations" is more citable for the query "best citation tracking tool for SaaS teams" than a review saying "great for any business." Use-case specificity is what makes a review extractable for the precise query a buyer is running. Request reviews that describe the reviewer's company type and team size alongside the outcome.
Dated reviews. AI engines weight review recency. A product with 300 reviews from 2023 and no reviews in 2026 signals a product that has lost user traction, regardless of the overall rating. A product with 50 reviews where 30 were posted in the last three months signals active, current adoption. Build a review collection programme that generates a consistent stream of new reviews each month rather than a single review-collection campaign followed by dormancy.
How Do You Validate That Review Schema Is AI-Crawlable?
The same validation process that applies to all schema for AEO applies to review schema. The specific check: confirm the AggregateRating block appears in the server-rendered HTML response rather than only in JavaScript-rendered DOM.
Run curl -A "Googlebot" https://yourproductpage.com | grep -A 20 "AggregateRating" in a terminal. If the AggregateRating JSON-LD block appears in the output, AI crawlers can read it. If it does not appear, your review schema is JavaScript-rendered and invisible to most AI crawlers — the same issue documented for all schema types in the schema errors guide.
Also validate through Google's Rich Results Test. Review and AggregateRating schema is eligible for star-rating rich results on product and local business pages — passing the Rich Results Test confirms the schema is valid and schema-eligible for both SERP features and AI citation signals simultaneously.
The NotioncCue AI Crawler Audit checks whether your review and rating schema appears in the initial HTML response accessible to PerplexityBot, GPTBot, ClaudeBot, and Googlebot. It also flags the most common review schema errors — missing bestRating and worstRating fields, comma decimal separators, and AggregateRating blocks that declare no written reviews — before you submit for re-crawl and wait three weeks to discover the schema was ignored. Run it before making schema changes to confirm the baseline, then run it again after to confirm the fix is in the crawlable HTML response.
Frequently Asked Questions
How many reviews does AggregateRating schema need before it affects AI citations?
Five is the practical minimum for safe implementation according to most schema practitioners' 2026 guidance. Below five, AggregateRating data is statistically unreliable and AI engines may ignore it. Above 50, the statistical weight becomes meaningful enough to influence comparison query results. Google's March 2026 schema update specifically flagged AggregateRating with unverifiable or very low review counts as facing enhanced scrutiny. Build to 25 to 50 reviews on your primary platform before investing heavily in on-site AggregateRating schema implementation.
Can you mark up reviews from Trustpilot or G2 in your own site's schema?
You can mark up the aggregate data sourced from third-party platforms — the average rating and review count — in your own site's AggregateRating schema, provided you attribute the source. Include a source field or a note in the description: "Based on 247 verified reviews on G2 as of July 2026." Do not reproduce the individual review text in your own site's Review schema if the original review was authored on a third-party platform — the original is owned by the review platform and reproduction raises copyright concerns. The aggregate data (score, count) is factual and markupable; individual review bodies should stay on the originating platform.
How does SoftwareApplication schema differ from Product schema for AEO?
SoftwareApplication schema includes fields specific to software products — applicationCategory, operatingSystem, softwareVersion — that help AI engines correctly classify your product for software-specific queries. AI engines answering "what is the best AEO tracking software" are looking for SoftwareApplication typed entities, not generic Product entities. Use SoftwareApplication for any web-based or desktop application, and nest AggregateRating and Review within it rather than using the generic Product type. The category-specific schema type produces more precise AI citation targeting for software comparison queries.
Does AggregateRating on a homepage improve citation rates for brand queries?
Limited. AggregateRating on a homepage declaring reviews of the company (not a specific product) runs into the self-review concern. AggregateRating on specific product pages or feature pages earns stronger citation signals because the reviewed entity is a concrete offering, not the company itself. For brand-level social proof that AI engines can cite, third-party platform presence — G2 company profile, Trustpilot page — earns more citation weight than any on-site declaration.