Fact-Checking Policy

What we verify

Before publication we check the claims a reader could act on: AI model names, versions and capabilities; software features; API availability and pricing; subscription and free-plan limits; cybersecurity recommendations; technical specifications; and regulatory or industry news. Colour, opinion and commentary are clearly written as such — they are not presented as fact.

Primary sources first

Product claims are checked against the provider's own documentation, announcements, changelogs, model cards and pricing pages; regulatory and research claims against the primary publication. Reputable secondary sources are used for context or when no primary source is available — and articles link their sources so you can check them too. We do not cite a source unless it actually supports the claim it is attached to, and we do not fabricate sources, URLs, dates or quotations.

How AI-assisted drafts are checked

Much of our coverage is researched and drafted with AI assistance in our newsroom. That assistance never makes a claim true. Every draft is fact-checked against its research record before publication, quotations are restricted to what sources actually say, and an article is published only after a human editor approves that exact version — AI output is never auto-published. Drafts that fail verification are corrected or rejected rather than softened. The audit trail of who approved what, and when, is kept in our systems.

Prices, plans and availability

Prices and plan limits in our tools directory and articles come from the provider's own pages, carry a "last verified" date, and link to the official source. We distinguish free plans from time-limited trials, and app subscriptions from API access — they are billed separately almost everywhere. Where a figure could not be verified, we leave it blank or label it rather than estimating. Taxes, regional pricing and promotions can change what you actually pay; the official pricing page always wins.

Security and technical claims

Security recommendations favour advice that is stable and widely agreed (unique passwords, phishing-resistant multi-factor authentication, prompt patching) over vendor marketing. Vulnerability and incident reporting relies on official advisories or the affected vendor's own statements. When security information goes out of date in a way that could put readers at risk, correcting it takes priority — see our Corrections Policy.

How we label confidence

Verified facts are stated plainly and, where it matters, linked to their source. Estimates, projections and our own analysis are written as such. Reviews distinguish hands-on testing (documented in the article's "How we tested" section) from research-based comparisons, and we never publish benchmark scores, ratings or test results we did not obtain or verify from a named source.

When we cannot verify

If a claim cannot be verified, it is either left out, attributed explicitly ("the company says…"), or flagged for editorial review. Rumours are labelled as rumours or not covered. Where our research system detects a claim it could not confirm, that limitation travels with the draft instead of being quietly dropped.

Related policies

This policy is part of our Editorial Policy. See also the Corrections Policy and the Review Methodology.