How we keep claims honest
Every claim earns the confidence it carries.
Splendorize recommends changes based on evidence from your own visitors, so the rules below decide how confident any statement is allowed to sound. They are enforced by software before you read a word — not by good intentions.
The claim ladder
Evidence is graded, and every statement carries a ceiling set by that grade. A statement may sound more cautious than its evidence, never more confident.
- 01
Descriptive
“We observed this pattern.”
The data shows something happening on your page. It does not say why, and it is not yet a reason to act.
- 02
Hypothesis
“Worth testing — not proven.”
The evidence supports an idea worth testing and measuring. It is never presented as a proven recommendation.
- 03
Recommendation
“Make this change.”
The evidence cleared our strongest bar: enough visitors, enough conversions, balanced audiences, and recent data.
Grades come from frozen, calibrated statistical gates that keep expected false discoveries at or below 5% of findings. The AI never grades its own evidence.
Causal language is banned
Below the recommendation ceiling, Splendorize never claims one thing caused another. Wording like “proven winner” or “will improve conversions” is rejected by an automated check, and before-and-after readouts of a shipped change say plainly that they are observational.
Numbers are quoted, never invented
Every AI answer cites the exact stored facts it read, and every number in the answer is checked against those facts, character for character, before it reaches you. The AI never computes your numbers; it can only quote them. If a figure cannot be verified, the answer is withheld.
Small samples stay hidden
Aggregate patterns appear only once enough visitors sit behind them. Below our minimum sample-size floors, a segment is suppressed rather than dressed up as insight — tiny samples mislead, and we protect the few people behind them.
And we never record your visitors
Splendorize photographs your own page, not the people on it, and works from aggregate signals. Nobody gets recorded to teach you why visitors leave.