Surfd

Methodology

The Discoverability Score is deterministic and explainable. Same footprint, same score. An LLM polishes recommendation wording, it never touches a number. Here is exactly how it works and where the numbers come from.

Seven pillars, weighted

The total is a weighted average of seven measured pillars. Weights shift per target role. Defaults:

Semantic match24% weight

How closely your public text matches the language a recruiter or agent uses for the role. Both games.

Evidence density16% weight

The share of your claims backed by a real number. Quantified, dated claims earn citations and model trust. Agent discovery.

Coverage14% weight

How many of the surfaces machines read you actually appear on. One profile is one point of failure. Both games.

Crawlability12% weight

Whether the AI retrieval bots can read a site you control, and whether it is structured and fast. Agent discovery.

Authority12% weight

Corroboration: the same true claim on more than one surface, so a model trusts and repeats it. Agent discovery.

Recency12% weight

How dated and fresh your footprint looks. Staleness quietly decays everything else. Both games.

Structure10% weight

Whether your footprint is machine-readable: clear sections, schema, consistent formatting. Agent discovery.

Two games

Discoverability is really two games with opposite optimizations. The same pillars feed both with different pull.

Get found by databases

How well LinkedIn Recruiter, Apollo, and ContactOut can index you and return a match. Won by presence, exact keywords, and a findable email.

Examples: LinkedIn Recruiter, Apollo, ContactOut, SeekOut.
Get ranked by AI agents

How well a sourcing agent or an LLM surfaces and trusts you from the open web. Won by corroboration, quantified evidence, recency, and structure.

Examples: Noon, Talentium, Juicebox, ChatGPT, Claude.

Measured versus modeled

  • Measured means we ran the real thing: the assistants queried live in the AI mirror, your verified facts read from connected surfaces, enrichment records checked against the finder databases. Labeled measured.
  • Modeledmeans we simulate with the same embedding model that powers sourcing tools: the sourcing rank and the peer benchmark. No one's private data is read. Labeled modeled.

What we do not do

Surfd improves real signal. It never fabricates a credential or a number, never writes hidden text or prompt-injection payloads, and never games detection. The widely repeated claim that most resumes are auto-rejected by an ATS before a human sees them is a debunked myth: the real risk is mis-parsing, so we help you parse correctly and get surfaced, not trick a machine.