How we report

Every Ground Truth piece is the result of three things: hands-on use of the product over a meaningful period, conversations with the people who build and use it, and an editorial perspective on what matters.

We do not write coverage from a press kit. We do not write coverage from a single demo session. If we haven't used the product the way a real user would use it — over weeks, on real briefs, on the device the product is actually used on — we don't publish the piece.

Minimum testing windows by piece type:

• Reviews — minimum two weeks of daily use across at least two real production scenarios.
• Deep dives — minimum three weeks of product use, plus four to six interviews with current and former team members, customers, and where appropriate, investors.
• Comparisons — minimum one month of paired daily use, with at least two paid tiers tested on each side.
• Best-of listicles — minimum five days of testing per tool across four production scenarios, with the top three tools getting an additional round of focused testing before final ranking.

Tools are scored on five dimensions, each rated 0 to 5. The composite Stack Score is a weighted average — features and value weighted more heavily than support and ease-of-use, because features and value are harder to change.

• Overall. The composite score, weighted toward what actually matters in production.
• Ease of use. How quickly a new user gets to their first useful output. Measured by handing the tool to a non-power-user and timing.
• Features. Depth and breadth relative to the category. Weighted toward the features that are actually used, not the ones on the marketing page.
• Value. Output quality per dollar across all tested pricing tiers, normalized against the category median.
• Support. Response time and quality on real product issues raised during testing, plus the depth of the product's documentation.

Scores are revisited every six months — or sooner if a major product release warrants a re-test. The last-reviewed date appears on every tool page.

We do not write about every AI product. The following will keep something off the site, regardless of buyer interest:

• Products without a real, usable free tier or trial. We cannot review what we cannot use.
• Products from companies that have demonstrably trained on copyrighted material without licensing or opt-out, where the practice is currently subject to active litigation.
• Products that have shipped fewer than two meaningful releases in the past six months. New launches need time before they are coverable; if velocity drops to zero, so does our interest.
• Products that condition press access on coverage approval. We do not show drafts before publication. We do not negotiate phrasing.

Every Ground Truth contributor must declare in writing any commercial relationship — past, present, or anticipated — with a company they intend to cover. This includes prior employment, consulting work, advisor positions, equity holdings, family ties, and personal relationships.

Declared conflicts are listed at the bottom of every piece, in the methodology box. If a conflict is significant enough to affect coverage, the piece is reassigned.

We correct errors fast and visibly. Every correction is appended at the bottom of the original piece, dated, and explained — what we got wrong, what the correct information is, and how we became aware of the error.

Significant corrections (those changing the substance of an argument or recommendation) are surfaced separately in the next edition of The Edition. We have an obligation to make sure our corrections reach the people who read the original.

To flag a possible error, write to corrections@groundtruth.example.com.

For the avoidance of doubt: Ground Truth does not, and will not, do any of the following.

• Accept payment for favorable coverage. Sponsored placements are clearly labelled and never affect editorial scores or judgment.
• Show pre-publication drafts to subjects of coverage. Fact-checking exchanges are limited to verifying specific claims.
• Embargo journalism in exchange for early access. If we accept embargoed material, the embargo is purely about timing, never about content.
• Use generative AI to write articles. AI is used in our research workflow (extracting quotes from interview transcripts, summarising long documents); it does not write our pieces.