Why the Ratings Look Confusing in AI (And How to Read Them)

Search for reviews of any major education AI tool — Duolingo, Coursera, Khanmigo, Quizlet — and you'll hit something strange fast. The ratings seem to contradict each other depending on where you look.

Two different audiences rate the same product

G2 reviewers rated Duolingo 4.5 stars across 142 reviews. Trustpilot reviewers rated the same company 1.5 stars across nearly 8,000 reviews. Coursera and edX show a similar split. This isn't a data error; it reflects two different populations answering two different questions. G2 reviewers tend to evaluate instructional value and product design. Trustpilot reviews skew toward billing disputes, subscription cancellation friction, and refund complaints. A school administrator in Nairobi, Manila, or Toronto evaluating whether a tool actually teaches well should weight the G2-style reviews more heavily. A parent worried about auto-renewal practices before a rollout should weight the consumer complaint sites more heavily. Neither number alone tells the full story.

Institutional tools barely register on review platforms at all

Products sold directly to universities and libraries — HeinOnline, Turnitin's Feedback Studio line — often carry thin or nonexistent G2/Capterra review counts, despite serving millions of students across dozens of countries. This isn't a quality signal in either direction. Nobody signs up for an institutional library license the way they'd sign up for a consumer app, so there's rarely a natural moment for someone to leave a review.

"Free for education" terms vary more than people assume

Canva for Education offers a fully free tier for verified K-12 teachers and students worldwide, while Canva for Campus (higher education) runs on custom enterprise pricing. Khan Academy's Khanmigo follows a similar split: free for teachers, a low monthly fee for individual learners and parents. Eligibility and pricing for these programs differ by country and institution type, so it's worth checking the specific terms rather than assuming a "free for students" claim applies everywhere the same way.

Small review counts don't disqualify a tool

Several classroom tools that teachers actually rely on — MagicSchool AI, CoGrader, Coursebox AI — carry small review bases on the major platforms, sometimes fewer than 100 verified reviews total. That's typical for tools serving a specific niche like K-12 teachers or independent course creators, rather than a mass consumer market. It doesn't mean the product is unproven; it means the number carries less statistical weight than a rating built on thousands of reviews, so reading a handful of the actual reviews tells you more than the star average alone.

What this means for choosing a tool

Check which platform a rating came from, how many reviews support it, and whether any complaints concern the product itself or billing and support — those are different problems entirely. For classroom-facing tools, a small review base built from real educators often carries more useful signal than a large one where consumer frustration about billing outweighs feedback on how well the tool actually teaches.

Sources:

• Duolingo ratings data: G2 product reviews, Trustpilot
• Additional ratings referenced from G2 and Capterra vendor profile pages for Coursera, edX, Khanmigo, Turnitin, and related tools (accessed 2026)