Methodology

How We Audit

Every claim is dated, sourced, and classified. No vendor input, no paid placement. Here is exactly how we measure what pricing pages leave out.

Transparency Score Formula
Transparency Score =
(Real Headroom × 0.30)
+ (Response Speed × 0.20)
+ (Limit Transparency × 0.15)
+ (Data & Privacy × 0.20)
+ (Reliability × 0.15)

// each dimension 0–100 · weights sum to 1.00 · displayed as %
The Five Dimensions
Real Headroom
30%
How much you can use before hitting a wall — stated caps, reset cycles, and whether those caps have been quietly tightened.
Response Speed
20%
Time-to-first-token from independent latency benchmarks. Never vendor-claimed numbers.
Limit Transparency
15%
Does the product show a usage meter before you hit the wall, or only after via an error?
Data & Privacy
20%
Default training data posture, opt-out availability, and how prominently it's communicated.
Reliability
15%
Incident count and uptime estimate from the official public status page over the last 30 days.
Source Classification
OFFICIAL
Sourced directly from the company's own published pages — pricing, help centers, privacy policies, status pages, official release notes. Verifiable by anyone at the linked URL.
OBSERVED (community)
Independent third-party benchmarks or widely corroborated user reports cited in tech press. Not from the company itself, but independently verifiable.
ESTIMATED (not verified)
Inferred or community-estimated figures where official data is not published. Always marked with an inline disclaimer. Do not treat as confirmed facts.
What "Shrinkflation" Means Here

The subscription price stays the same, but the actual usage shrinks — through a credit-system switch, a feature-specific cap cut, or a confusing new quota model. Every confirmed event is documented with a date and source.

Our Pledge

Found an error? Email data@aitransparency.com with a source link. Verified corrections go live within 48 hours.