Loading…
Millions of buying decisions start with one question: can I trust this company? TrustRating answers it with evidence — a panel of independent AI models, verified reviews from real customers, and a scoring method we publish in full.
Anatomy of a TrustScore
AI panel + verified reviews, weighted by recency
TrustRating was built on a simple idea: ask the best available researchers the same careful questions, then show your work.
Trust online is broken in a specific way. Reviews are plentiful but easy to game, ratings get averaged without context, and the company being rated is often the one paying the platform that rates it. The result is a number that looks objective and means very little. We built TrustRating because we wanted a trust signal we would rely on ourselves — one where you can see exactly what went into the score, who said it, and when it was said.
Our approach rests on a simple observation: large language models are exceptional researchers. They have read the regulatory filings, the support forums, the news archives and the fine print that no individual has time to work through. Asked precise questions, they can summarise a company's public reputation in seconds. But no single model has a monopoly on judgment, and every model carries its own blind spots. So we never rely on one. We ask several, independently, and aggregate what they conclude.
Around that panel we built the parts that keep it honest: verified reviews from real customers, fraud screening on every submission, a methodology published in public, and a hard wall between the companies we rate and the revenue that funds us. A company can improve its score by treating its customers better. There is no other route, and there is no price list.
Three independent inputs, combined by a formula that anyone can read. No editorial thumb on the scale.
Each enabled model receives the same structured prompt about the same company, with the same evidence attached — including a live fetch of the company's own website. The models answer separately and never see each other's conclusions, so agreement between them is meaningful rather than an echo. Every published analysis names the model that produced it.
Anyone can write a review, but every review must clear a verified email address before it is published, and TrustGuard screens each submission for the patterns that betray coordinated or incentivised writing. Reviews are attributed, dated and permanent — businesses can reply to any of them, and can edit none of them.
The panel's verdict and the human reviews are blended into a single rating from one to five stars, weighted by recency so a company that turned things around last quarter is not judged forever on its worst year. Scores recompute automatically as new evidence arrives, and the full weighting is documented on our methodology page.
This is the live roster of models currently scoring companies on TrustRating. We add and retire models as the field moves, because a panel frozen in time slowly stops reflecting the state of the art. When a model is retired its past analyses stay on the record, still attributed to it.
A rating is only useful if everyone reads it the same way, so the bands are fixed and public. A four-star company is genuinely good; a two-star company has a pattern of problems, not a single bad day.
Four commitments that decide how the product gets built, including when they cost us money.
We do not sell scores, and we do not let companies edit or delete reviews about them. Our costs are covered by an opt-in business product — embeddable widgets, analytics and reputation tooling — none of which can move a public rating by a single decimal place.
Every AI analysis names the model that produced it. Every review is attributed and dated. Our scoring method is published rather than described, so you can check our arithmetic instead of taking our word for it.
Companies can claim their listing, prove they own their domain, and reply publicly to any review. Getting a review removed requires showing it breaks our policy — not knowing someone, and not paying us. When we get something wrong, we correct it in public.
Stated plainly, so you can hold us to them.
A strong reputation is an asset, and we think you should be able to use it. Claiming your listing is free: verify that you own the domain and you can reply to reviews, publish answers to the questions customers actually ask, invite recent buyers to leave feedback, and see how the AI panel is describing you.
Paid plans add the commercial tooling — widgets and badges for your own site, reputation analytics, competitor tracking, API access and integrations. What no plan adds is influence over your rating. The paid product sits entirely downstream of the score, and that separation is the whole reason the score is worth displaying.
We are a small, deliberately opinionated team. Decisions that affect a public score get made in the open and written down, because a trust platform that cannot explain itself has no business asking anyone to trust it. When we ship something that changes how ratings are calculated, it goes on the methodology page before it goes live.
We would rather be slow and right than fast and confidently wrong. That means shipping fewer features, saying no to revenue that would compromise the ratings, and treating every complaint about a score as a bug report worth investigating. If you think we have a company wrong, tell us — that feedback has changed our scoring more than once.
Look up a company, read what the panel and real customers found, and judge for yourself. If you have been a customer, add what you know.
A wrong score is worse than no score. That is why we run a panel instead of a single model, screen reviews for fraud before they publish, weight recent evidence more heavily than old evidence, and recompute continuously as the facts change.