The Salienci Score - Measurement Methodology.
Salienci measures Recommendation Authority - whether AI answer engines independently select, cite, and recommend a business when buyers ask.
This document publishes what we measure, how we measure it, how often, and where the known limits are. The composite weights and probe construction are proprietary; everything else is stated here in the open.
What We Measure - and Why Visibility Is Not Enough
Being visible to AI is not the same as being recommended by AI. An answer engine must complete three movements before it recommends a business: it must reach and identify the entity, extract a citable answer from its content, and find the wider ecosystem corroborating that answer. The Salienci Score is built on exactly that causal chain - three driver pillars, each auditing one movement.
| Pillar | Question it answers | What is audited |
|---|---|---|
| Agent Readiness | Can a machine reach, render, and verify you? | Crawler access, rendering, structured identity, entity verification, business-profile integrity |
| Answer Quality (AEO) | Can a machine extract a citable answer? | Content structure, answer targeting, citation readiness, factual anchoring, content quality |
| Entity Visibility (EVS) | Does the ecosystem vouch - and does AI cite you? | Trust and review signals, authority signals, community presence, and the measured citation outcome |
A fourth band - Foundation (conversion, usability, measurement instrumentation, accessibility and governance) - is audited and repaired but reported as a separate Digital Presence Score. It is never blended into the authority result, because fixing a contact form does not make an AI more likely to recommend a brand. Keeping the two apart keeps the headline score honest.
How the Outcome Is Measured
The pillars audit inputs. The outcome - who AI actually recommends - is measured directly, on a fixed monthly cadence, using a structured probe program:
Buyer-intent prompts
ProbeA bank of real category-buyer questions spanning five intent classes (discovery, capability, trust, comparison, and natural phrasing), localized to the client's service area and, where relevant, language.
Four leading AI assistants
ProbeEach probe cycle runs across ChatGPT, Gemini, Perplexity, and Claude, with video and community ecosystems (YouTube and Reddit) measured through their own channels - because answer engines demonstrably draw on both.
Seven citation-outcome dimensions
ProbeEach cycle records how often the business is mentioned, where in the answer it appears, whether the stated facts are accurate, how many distinct platforms cite it, how fast citation frequency is changing, and its visibility in the video and community sources engines draw on.
Fixed re-score gates
ProbeEvery engagement is re-scored on the same probe bank at set intervals, so movement is measured against an identical instrument - not a moving target.
The Content Quality Gate
No content produced through Salienci OS reaches publication without passing a 475-rule validation framework, applied in three tiers: deterministic structural checks, linguistic analysis, and AI-judgment evaluation for the dimensions that require reasoning. Content that fails is returned for correction - it is not published and hoped for. Published content is then re-measured through the same probe program that scored the baseline, closing the loop between diagnosis, production, and outcome.
Attribution Confidence - Never Guessed
When Salienci reports that AI visibility produced a lead, a call, or a booking, the claim carries one of five confidence tiers. We publish the taxonomy because attribution claims without a confidence label are marketing, not measurement.
| Tier | Meaning |
|---|---|
| CONFIRMED | Verifiable proof - an advertising click identifier, a matched customer record, or a verbatim call statement. |
| HIGH | Strong signal - an AI assistant or search engine detected as the referring source. |
| MODERATE | Declared by link tagging or aggregate estimate; not independently verified. |
| CONTRIBUTING | The channel helped but did not solely cause the lead. Partial credit only. |
| UNKNOWN | Evidence insufficient. Reported honestly - never guessed. |
Data Integrity Commitments
Provenance on every number
Every emitted score is labeled LIVE, CACHED, ESTIMATE, or UNAVAILABLE. Missing data is reported as UNAVAILABLE - it is never invented.
Ranges, not point promises
Score targets are always expressed as ranges. A single-digit promise about a probabilistic system is a sales device, not a measurement.
Same instrument, every cycle
Baselines and re-scores run on the identical probe bank, so improvement is attributable to the work - not to a changed yardstick.
Known Limitations
A methodology that hides its limits should not be trusted with your budget. Ours are these:
- AI answer engines are non-deterministic. The probe program is a structured, repeated sample of real buyer demand - a rigorous sample, not a census of every possible question.
- Citation outcomes converge over roughly 60–90 days after access and content fixes land. Anyone promising instant citation movement is describing a system that does not exist.
- Benchmark comparisons are drawn from our measurement base, which is growing through our beta program; benchmark figures are labeled as such until the base supports broader claims.
- Third-party AI platforms change behavior without notice. Cadenced re-measurement is the control for this; immunity to it is impossible.
What We Do Not Publish - and Why
The composite weighting, the probe construction method, and the scoring mathematics are the audited core of the framework and are not published. They are available under NDA to clients and to independent auditors. The distinction we hold to: the framework is the intellectual property; the score - and everything in this document - is the artifact you are entitled to understand before you rely on it.
Written methodology questions are answered publicly at salienci.ai. If we cannot defend an element of this method in writing, we will change the method - not the answer.