How Our AI Works
RatingMantra turns a plain-English question into a clear, side-by-side comparison. This page explains exactly how — the public sources we analyze, how we average ratings and read sentiment, how fresh the data is, and, just as importantly, where our limitations lie.
Last reviewed: August 2026
The short version
When you ask a question, RatingMantra searches publicly available web sources relevant to your topic, uses AI to read and extract ratings, prices, and review sentiment, and then organizes everything into a ranked comparison with links back to the sources. Nothing is stored in a fixed, hand-curated database — each comparison is assembled from current public information at the moment you ask.
What data sources the AI analyzes
RatingMantra draws only on publicly available information, and it chooses sources based on what you're comparing. It does not use private data or scrape behind logins. Typical sources by category include:
Software & tools
G2, Capterra, and similar review platforms.
Restaurants & places
Google, Yelp, TripAdvisor, and Zomato.
Products
Amazon and major retailers, plus expert and buyer reviews.
Employers
Glassdoor and AmbitionBox.
Insurance & finance
NerdWallet, Forbes Advisor, ValuePenguin, Trustpilot, and ConsumerAffairs.
General discussion
Public forums and articles where people share real experiences.
Every comparison links back to the pages it drew from, so you can open the original sources and check them yourself.
How ratings are averaged
Different sites use different scales and audiences, so a single star rating rarely tells the whole story. The AI reads the ratings it finds across relevant sources and produces a representative average on a consistent 5-point scale. Where a product is rated highly on one site but poorly on another, that spread is reflected in the sentiment breakdown rather than hidden behind one number.
How sentiment is aggregated
Beyond the star rating, RatingMantra reads a representative sample of public reviews and estimates the balance of opinion — the share that is positive, neutral, and negative. This is shown as a simple sentiment bar next to each option, so you can see not just how highly something is rated but why: the recurring praise and the recurring complaints. Sentiment is an AI estimate of overall tone, not a precise count of every review ever written.
How prices are handled
When price information is publicly available, RatingMantra includes it and converts it into your local currency where needed (based on your approximate region). Prices move constantly and may be approximate, converted, or out of date — treat them as a signal, and always confirm the current price with the seller before buying.
How often the data is refreshed
RatingMantra does not maintain a static, periodically-updated database. Instead, each comparison is generated on demand from current web results at the moment you search. To keep things fast, popular queries may be cached briefly, so the information you see reflects recent public data rather than a fixed snapshot. In practice this means results stay current with what's publicly available, without a fixed refresh schedule.
Honest limitations
We'd rather be trusted than perfect. RatingMantra is a fast, helpful starting point — not the final word. Please keep these limits in mind:
- AI can be wrong. Results are generated automatically and may be incomplete, outdated, or occasionally incorrect.
- Coverage varies. For niche or very new items, there may be little public review data — you'll sometimes see "no review data yet."
- Prices are approximate. They may be converted or out of date; verify before purchasing.
- Not professional advice. For insurance, financial, legal, medical, or safety decisions, treat results as background only and consult a qualified professional and official sources.
- Not exhaustive. A comparison reflects the sources available at the time, not every option or every review in existence.
Our commitment to transparency
Because results come from public sources and AI analysis, we always show our work: source links on every comparison, a clear note that information is AI-generated and may contain errors, and this methodology page you're reading now. If something looks wrong, the Support button on any page lets you flag it and we'll take a look.
See the methodology in action.
Run a comparison →Want the simpler overview instead? Read how RatingMantra works in four steps, or dig into review sentiment analysis.