Best E-Commerce Review APIs for AI Agents in 2026

"Review data" isn't one product. A review under an Amazon listing is about the item — does it work, was it accurately described. A review on a Yelp or Google Maps page is about a visit — the service, the wait, whether the staff got the order right. Picking a review API without asking which kind you actually need is how a team ends up pulling product-review sentiment when what they wanted was location-level service quality.

If you're an agent (or building one) reading this rather than a human, the full machine-readable schema for every endpoint below lives at mindcase.co/skills.md.

What should you actually check before picking a review API?

Four things, in order of how often they get skipped:

• What the review is actually about. A product review and a business review measure different things. Confusing them means analyzing the wrong signal.
• Whether there's a hard cap on volume. Some sources let you pull every review a listing has ever received; others cap you at a fixed number no matter what you ask for.
• Whether an owner can reply, and whether that reply is in the data. A business responding to a bad review is itself a signal. Not every source captures it.
• What identifies the reviewer. A verified-purchase flag, a local-guide badge, or nothing at all — the credibility signal available depends entirely on the platform.

How do you pull a review sample from any of the three?

Same shape across all three: one business or product URL per call, a review count, and a clean array back. This one pulls Google Maps reviews for a single location:

curl -X POST "https://api.mindcase.co/v1/data/google-maps/reviews/run?wait=true" \
-H "Authorization: Bearer $MINDCASE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
  "params": {
    "urls": "https://maps.google.com/?cid=1234567890",
    "maxReviews": 100,
    "sortBy": "newest"
  }
}'

# $0.00025 per review

Amazon Reviews and Yelp Reviews take the same one-URL-per-call shape — swap the endpoint and the input URL, and the rest of the call barely changes.

Why does the review cap matter more for Amazon than the others?

Amazon caps maxResults at 100 reviews per product — that's Amazon's own limit, not a Mindcase restriction, so no maxResults value above 100 returns more. Yelp and Google Maps both accept 0 for "all available reviews," with no fixed ceiling.

That difference changes what a pull can represent. A location with 400 Google reviews can be sampled at 100% if you're willing to pay for all of them. An Amazon listing with 4,000 reviews can never be fully represented by one product's data — 100 reviews is a fixed slice, not a sample size you control. Auditing Amazon reviews at real scale means pulling across many ASINs, not asking for more per product.

What can you learn from an owner's reply that you can't from the review itself?

Yelp returns a publicReply field and Google Maps returns ownerResponse plus ownerResponseDate — a business owner's public response to that specific review, when one exists. Amazon has no equivalent field. Sellers don't get a public reply thread under a review the way a restaurant or storefront does on Yelp or Google Maps.

That makes owner replies a real, if partial, service-recovery signal for local businesses — did anyone respond, how fast, what did they say — that simply doesn't exist as a data point for a product review. If reply behavior is what you're actually measuring, Amazon reviews can't answer that question no matter how many you pull.

Which endpoint should you use for which job?

What can't you get from any of these?

Two cases.

Cross-platform reviewer identity. None of the three tell you whether the same person left reviews on more than one platform. Each source only knows about its own reviewers.

A rating scale that means the same thing everywhere. A five-star Amazon rating, a five-star Yelp rating, and a five-star Google rating are not directly comparable — each platform's reviewer base and rating culture differ. Averaging star ratings across all three as if they were one scale will misrepresent all of them.

FAQ