Editorial
How UseRightAI sources its data, scores models, decides what gets published, and fixes what it gets wrong. If you want to know whether to trust a recommendation on this site, this page is the honest answer.
UseRightAI is a comparison directory for AI models. Every page exists to answer one buying question — which model to use for a task, what it costs, and what you give up by choosing it. We publish capability scores, verified API pricing, context windows, speed ratings, and editorial verdicts.
We do not publish news, rumours, or unverified specifications. A model does not enter the ranked directory until a provider or a host has published real per-token pricing, because a comparison page without a price is a page that cannot answer the question it was written for. Models we are tracking but cannot yet price are noted in the changelog rather than ranked.
Pricing is checked daily against provider APIs and pricing pages, and shown as standard, non-batch, per-million-token rates so that models are comparable. Where we quote a monthly cost, we state the workload it assumes and exclude batch and caching discounts, which makes every such figure a ceiling rather than an estimate.
Benchmark figures are taken from published results — SWE-bench, MMLU, ARC-AGI-2, HumanEval — sourced from provider announcements or third-party leaderboards, and cross-referenced against community evaluations such as Chatbot Arena where a provider's own claim is the only source. Context windows and capabilities come from official API documentation.
Our capability scores out of 100 are our own weighting, not a benchmark result. They combine published scores with production signals like which models teams keep using after the first month. They are an editorial judgement and we label them as one.
Detection is automated; publishing is not. When a provider ships a model, our tooling notices it within about a day, pulls the pricing, and drafts an entry. That draft is written to a review queue and is not visible anywhere on the site.
A person then checks the draft against the provider's own pricing page and model card before it is promoted. Until that happens the model does not appear in the directory, in comparisons, in rankings, or in the changelog. Nothing machine-drafted reaches readers unreviewed.
The same rule applies to existing pages. Our tooling recalculates rankings automatically when model data changes — that is arithmetic on verified numbers, and letting a stale recommendation stand would be the worse outcome. But rewritten page copy is staged for review rather than published, for the same reason a new model entry is.
We will get things wrong. Prices change without announcement, providers restate benchmark numbers, and a verdict that was right in March can be wrong by June. If you find an error — a stale price, a wrong context window, a benchmark figure that does not match the source, or a recommendation that no longer holds — tell us and we will fix it.
Email hello@userightai.com with the page URL and what is wrong. Factual errors in pricing or specifications are corrected as soon as we can verify the right figure, usually the same day. Where a correction changes a recommendation rather than a number, we update the page and note it in the changelog so the change is visible rather than silent.
Every comparison page shows when its underlying model data was last verified. If that date looks old relative to a model you know has changed, treat the page with suspicion and tell us.
No provider pays to appear in the directory, to rank higher, or to be recommended. No provider sees a ranking before it is published or has any mechanism to influence a score. We have no commercial relationship with the AI providers we cover.
The site carries automated display advertising. Those ads are selected by the ad network, not by us; we do not know which advertisers appear on a given page, and an advertiser appearing next to a comparison has no bearing on what that comparison says. Ad slots are visually and structurally separate from editorial content.
If affiliate links are added in future they will be disclosed on the page that carries them, and they will not affect rankings, scores, or which models are included. A model is never added or excluded on commercial grounds.
We use language models as a drafting and data-processing tool: extracting specifications, drafting a first pass of a model entry, and generating the comparison text that is assembled from verified figures. It would be dishonest to describe a directory of this size as entirely hand-typed.
What we do not do is publish that output unreviewed. Machine-drafted entries sit behind the review gate described above. Comparison pages are assembled from figures we verify, not from a model's recollection of what a price used to be — the numbers on those pages come from our database, not from a generated paragraph.
If you spot text on this site that reads as though nobody checked it, that is a bug and we would like to hear about it at the correction address above.