Methodology

ADAM·CITE Audit Methodology

USAT-MTH-CITE-001 · Version 1.0 · Effective 2 September 2026

1. What an ADAM·CITE audit is

ADAM·CITE is an evidence-based AI citation audit. The automated scorer is one instrument inside the audit, not the audit itself. Every audit defines the entity and the query set, evaluates real content at the URL level, and measures the pillars that are in production. Crawlability is scored today. Identity, Trust and Extractability are pending and are marked as such until they ship. A complete audit also observes what answer engines actually do, benchmarks comparable peers, produces evidence-backed findings, prioritizes the corrective actions, and records a baseline that can be measured again.

What each product tier buys is set out in Terms of Service §4. That is the single published statement of tier deliverables; this page describes the method those tiers draw on.

2. Three measurements, not one

One number cannot carry three meanings, so we publish three.

Three rules govern them, and we hold ourselves to them in writing:

• Readiness is never described as citation performance. It is a readiness measure and our language says so.
• Peer Position leads in what we tell you. An absolute 61 means little on its own. “61 of 100, 72nd percentile among 14 comparable regional groups, category leader is 79” means something and is defensible.
• Only Observed AI Visibility may claim anything about actual citation behaviour — and only for the queries actually tested, on the dates actually tested.

We never compare across categories. This product does not compare a law firm to an encyclopedia, and the interface makes that comparison impossible rather than merely discouraged.

3. The four pillars, and their limits

Crawlability, Identity, Trust, Extractability.

Structured data is one check family inside Extractability, not the definition of it. A scorer that keys heavily on FAQ or Product markup measures commercial page construction — which is how a brochure page can out-score an encyclopedia. It also rewards a tactic with a shrinking payoff: search engines have been narrowing structured-data rich results for years. And markup that exists but contradicts the visible page is worse than no markup at all. The defining test of Extractability is therefore the machine extraction test in §4, Phase 6 — whether facts actually come out — not whether markup is present.

Current coverage, stated plainly. Crawlability is fully scored in every audit today. Identity, Trust and Extractability are rolling out with the v0.2 engine and are marked pending in delivered reports until they ship. We would rather show you the gap than average over it.

4. The fifteen phases

Not every tier runs every phase; see Terms §4 for what your tier includes.

Phase 1 · Scope and entity definition

The audit records its control record — organisation, canonical domain, entity type, location, offerings, market, audiences, audit date, engine version, methodology version, auditor. The query set is then defined and frozen before any engine is queried, and pre-registered with a timestamp. A query set assembled after seeing results is marketing, not measurement.

Phase 2 · URL discovery and content inventory

We never score only the homepage and imply the organisation was audited. Meaningful URLs are discovered and classified by page class, and sampled by a documented rule with the sample size recorded. Every tested page retains its submitted URL, resolved URL, HTTP status, redirect chain, declared canonical, page class, crawl timestamp and a content hash.

The URL contract: the engine either scores the URL as submitted or fails loudly with a stated reason. Silently substituting a different URL — and then citing it as evidence — is never acceptable output.

Phase 3 · Crawlability

Not whether a page returns 200, but whether retrieval systems can reliably discover, access, follow and interpret your important content: discovery and sitemaps, robot directives, HTTP integrity, rendering, canonicalisation, and site architecture.

AI crawler policy is a check, not a pass or fail. These are distinct agents with distinct effects, and you may rationally allow one and block another, so we report them separately:

Agent names and behaviour change. We read each operator’s published documentation at audit time and record which version we checked against, rather than hardcoding a list and letting it rot.

The output is never just “Crawlability 78.” It is “17 of 23 priority pages are discoverable within three internal links; six rely on PDF navigation and have weak HTML discovery paths.”

Phase 4 · Identity

Can a machine confidently determine who you are, and connect your pages, people, locations, products and external profiles to the same entity? We check on-site signals, structured relationships, and independent corroboration across directories, registries and profiles.

The output is an Entity Confidence Map with five states: confirmed, probable, ambiguous, missing, contradictory. Contradictory is the most valuable of the five and the one conventional SEO tools never report — two different addresses for one location actively suppresses confidence in the entity.

Phase 5 · Trust

What independent evidence would give an answer engine confidence in your claims? The distinction that carries the pillar: self-assertion is not evidence. “We are the best provider in Florida” is a claim; an accreditor’s public record is evidence.

We report corroboration depth per claim: zero independent sources is self-assertion only, one is weak, two or more credible independent sources is stronger. Each source is recorded with its URL and observation date.

Phase 6 · Extractability

Can a machine reliably turn your content into discrete, accurate facts? We check content structure, structured data as one family among several, document accessibility (facts trapped in PDFs or images), and semantic clarity.

The machine extraction test is the defining measure. For each priority page we attempt to extract a defined fact set and record the result:

Expected fact:  Tuition
Extracted:      $24,500
Confidence:     High
Source:         /programs/cybersecurity  (resolved URL, 2026-09-02T10:42:03Z)

Expected fact:  Program duration
Extracted:      Ambiguous
Evidence:       Page states both "18 months" and "two years", no qualifying context
Confidence:     n/a

Two standing rules: we never recommend FAQ markup for content that is not genuinely a FAQ, and never recommend schema that contradicts the visible page. Both would create a quality risk for you.

Phase 7 · Answer engine observation

A separate observational layer, never silently folded into pillar scores. For each frozen query, on each supported engine, we record the engine, the date and time, the exact query, the captured answer, and whether you were mentioned, identified correctly, cited, cited from your own site, and whether any stated facts were wrong.

Non-determinism is handled explicitly. Each query runs a minimum of three times per engine per observation window. We report a mention rate and a citation rate as fractions, never as a single yes or no — a result seen once in three runs is reported as one in three. Personalisation, geography and session state affect answers, so we record the conditions of observation and hold them constant between baseline and re-test, or the delta means nothing.

The governing sentence for this whole product: a technical score and an observed citation are not the same thing. A site can be technically excellent and not cited. A site can be technically poor and cited on authority alone.

Phase 8 · Peer benchmarking

The peer group is defined by industry, geography, business model, size and page class, and recorded before scoring. We run the same method against the peers and report your absolute score, peer percentile, peer median and gap to the leader. Client reports use real competitors you approve; our marketing samples use fictional organisations only. We do not publish a named competitor’s score in our own marketing.

Phase 9 · Scoring and confidence

Overall Citation Readiness = (Crawlability + Identity + Trust + Extractability) ÷ 4, unweighted. We publish that it is unweighted: it is a stated simplification, not a validated model, and it stays that way until we have the evidence to justify weighting it.

Published beside every score, always, as a first-class field:

A score without a confidence level is not a deliverable. Six pages crawled out of a hundred and a clean 72 on the cover is false precision, and false precision is the thing this product exists to cure. Where we are blind, we show you the blindness.

The check families under each pillar are published. Exact detector weights are not, to stop the score being gamed — and we state that boundary here rather than hiding it.

Phases 10–11 · Findings and prioritization

Each finding carries an ID, title, severity, evidence, why it matters, the action, the effort and the tier. No finding says “AI cannot cite this” unless we observed exactly that. The defensible form is “this content is difficult to extract reliably because…” followed by the evidence.

Findings are not the deliverable; actions are. Recommendations are prioritized by expected impact, confidence, effort, dependency and breadth, then sorted into Now, Next and Later, each with a pillar and an owner.

Phase 12 · Human QA

Required on every paid audit before it leaves the system, and the reviewer’s name goes in the report. They check score arithmetic, entity matching, false positives and negatives, redirect and canonical interpretation, competitor identity, claim wording, that every evidence URL resolves to what the finding says it shows, and that recommendations are feasible.

The free audit is automated and carries no human QA. We say so here because the named-reviewer promise applies to paid engagements and you should not have to infer the boundary.

Phases 13–15 · Report, verification, baseline

The report represents the audit; it is not the audit. Every figure in the front of the report traces to an entry in the evidence appendix — no orphan numbers.

After you implement changes we verify each action individually before re-scoring, which separates “did the work happen” from “did the number move.” Each audit is a baseline, and we track readiness, pillar scores, audit confidence, peer percentile, mention rate, citation rate, source diversity, corroboration depth, coverage, and open, resolved and new findings over time.

5. How we test the method itself

A method is not validated because the arithmetic works. It is validated when a high score means what we tell you it means.

A validation panel of at least 40 organisations — across reference, government, health, higher education, major SaaS, local services, professional services, regional healthcare and our own properties — is the gate we have written for the method. That panel has not been run yet. Until it has, we do not claim the scorer is validated against observed AI visibility. When the panel runs, it will record both the readiness score and the observed AI visibility for each organisation. Before the scorer may be described as validated:

• Within a peer group, readiness rank and observed visibility rank must correlate positively — the relationship must not be flat or inverted.
• No organisation that is cited in the majority of its own relevant queries may score in the Risk band. If a major health system is cited constantly and our scorer calls it 27, either the score is wrong or the label on it is wrong. Both are fixable; shipping neither is not.
• Cross-category comparison must be impossible in the product, so cross-category absurdity cannot be produced.
• Every pillar must have at least one check whose removal measurably changes the score, and no single check may account for more than a stated share of a pillar. A pillar that is one detector wearing a pillar’s name gets renamed.

When the panel exists, it will be re-run whenever the engine version changes, and quarterly regardless, because answer engines change under us. We will publish the result, including where we score poorly — running this audit on our own site and publishing the weaknesses is the strongest thing we can show you.

6. What we do not claim

Out of scope for v1, said plainly so nobody has to guess: weighted pillar models, predictive scoring (“your score will reach X”), guaranteed citation outcomes, engine-specific ranking claims, and any statement about why a specific engine did or did not cite a page. We observe behaviour; we do not explain the engines’ internals, and the report language reflects that.

7. Versioning

Every audit records the engine version and the methodology version, and both appear in the report footer and the evidence appendix. Methodology changes get a version bump and a dated changelog entry on this page. Reproducibility without versioning is not a promise, it is a hope.

Version 1.0 — 2 September 2026. First published version.

8. Questions

USA Telecom Consulting LLC · info@usatelecom.us · 888-989-4872. A human replies.

ADAM·CITE · cite.adampulse.us · USAT-MTH-CITE-001 v1.0 · USA Telecom Consulting LLC · SDVOSB CAGE 9QJS2 · UEI NJ7FKBV9X6L1