Inside Authority Signals
The Evidence Ladder
Proof is not pass or fail. AI systems do not trust every signal equally, so Surfaced grades the evidence on your site against five tiers inside the Authority Signals pillar. A vague claim is weak. A named customer story is stronger. Original data is strongest.
The five tiers
Stronger, harder to fake
Original research, benchmarks, or numbers only you hold. An assistant using this fact has to attribute it to you, because it exists nowhere else.
Surfaced check: Original data or research present?
Independent reviews, press, interviews, and unprompted mentions. Weighted highly precisely because you did not write them.
Surfaced check: Independent third-party mentions present?
A real, named customer with a specific, checkable outcome. Still first-party, but concrete and attributable.
Surfaced check: Named case studies present?
A named person vouching for you. Better than anonymous praise, but you still chose which quotes appear.
Surfaced check: Named testimonials present?
"Loved by thousands", "the leading platform" — nothing behind them that a system can check. Largely ignored.
Surfaced check: Unsupported claims detected?
Weaker, easy to fake
Why tiering matters
Between two similarly well-structured sites, an assistant leans toward the one whose claims can be corroborated. That makes the kind of proof more important than the amount. Ten anonymous testimonials do not outweigh one independent review.
The tiering applies regardless of how the markup was produced. Review schema generated by a tool is treated exactly like equivalent hand-written markup — same detectors, same weight. What is being graded is the substance of the evidence, not its authoring method.
How to climb it
Most sites sit at tier 4 or 5. The fastest realistic move is tier 3: ask one happy customer for a short, specific outcome you can publish with their name. That single page outperforms a wall of adjectives.
Tier 1 is more achievable than founders assume. You do not need a large study — patterns across your own users, a small benchmark, or a survey of a few hundred people is genuinely new data. Publish it with a clear headline number, a stated method, and a date.
See your evidence tierRelated guides