Evidence
before opinion.
We do not begin with a preferred tactic, platform or predetermined answer. We begin by identifying the business question, determining what evidence is available, and separating observable signals from assumptions.
What can be shown comes before what can be claimed.
what the methodology protects against
Better data does not automatically produce a better decision.
Problems often begin when incomplete information is treated as a complete picture, weak signals are given too much weight or a tactic is recommended before the underlying issue is understood.
what we examine
What we examine depends on the decision.
An engagement may focus on one of these areas or connect several of them. The work is shaped by the question, not by a fixed package.
AI visibility and recommendation
- Question
- Can AI systems understand, verify and confidently recommend the organization?
- Evidence
- Answer-engine outputs, citations, entity clarity and third-party corroboration.
- Decision
- Where to strengthen the evidence AI systems rely on.
Organizational trust and authority
- Question
- Do the organization’s people, expertise, platforms and claims align into one credible narrative?
- Evidence
- Positioning, executive and subject-matter authority, digital touchpoints and proof points.
- Decision
- How to align the narrative so it holds together across audiences.
Brand and market evidence
- Question
- What evidence about the organization exists, and where is it fragmented or missing?
- Evidence
- Signals across executives, employees, customers, franchisees, partners, media and platforms.
- Decision
- Which evidence to build, connect or correct across the network.
Paid-media signals and intent
- Question
- Are automated advertising platforms optimizing toward qualified, in-market demand?
- Evidence
- Conversion definitions, first-party data, audience signals, placement and delivery data.
- Decision
- Which signals and boundaries to give the platform.
how the work progresses
From evidence to action.
Not every engagement follows an identical path, but most move through the same progression, from the decision at stake to whether the change actually improved things.
Establish the business question, scope and success criteria before any data is gathered.
Collect, normalize and assess the available data, signals and research.
Document what the evidence supports, what cannot be evaluated and where limitations exist.
Connect findings to business consequence, practical recommendations and accountable next steps.
Support changes where appropriate and determine whether performance, visibility or measurement quality improved.
The system can automate collection, comparison, classification and repeatable checks. The analyst remains responsible for context, materiality, interpretation and recommendation.
The tools automate the evidence.
The analyst owns the verdict.
what the client receives
What you receive.
Depending on scope, an engagement produces a documented, defensible record you can act on, not a slide of reassurance. A typical engagement can include:
The starting position, captured so later change can be measured against it.
Each finding traceable to the data or source that supports it.
What could not be evaluated, stated plainly rather than papered over.
Ranked by business consequence and confidence, not volume.
Enough specificity for your team, or ours, to make the change.
Who does what next, so findings do not stall after the report.
How and when to check whether the change worked, when validation is in scope.
Bnarrativ principles
Every recommendation is a trust decision. Every trust decision is built on evidence.™
Observable evidence and analyst interpretation are identified separately.
Material findings remain traceable to their supporting evidence.
Uncertainty is disclosed rather than silently filled with assumptions.
Automation supports repeatability; it does not decide what matters.
Recommendations account for organizational, market and location context.
Measurement continues after implementation when validation is part of the scope.
How Cited measures.
Cited reports what AI assistants actually say — never invented numbers. This is the method behind that claim: what is tested, how questions are built, what each status means, and what the method cannot establish.
Cited reads Perplexity, OpenAI (web-grounded) and Gemini, with Google AI Overviews and additional surfaces on the roadmap. These are the assistants most people currently use to research businesses and shortlists. Each reads a different set of sources, so they are recorded separately rather than blended into one figure.
Prompts are buyer questions assembled for a brand’s exact category and market — the way customers actually ask, not generic keywords. Questions surface a shortlist without naming the brand, so a mention has to be earned by the assistant rather than prompted by us. The free scan reads a focused set live; a full assessment widens coverage across more questions and engines.
Market is an explicit, user-visible scope choice — national or online, or a specific city, county or state. Local answers can differ sharply from national ones, so the market a scan represents is always recorded with the result and never assumed.
A mention means the answer names the brand. A citation means a source supports the answer — which may or may not be the brand’s own site. A recommendation means the brand is put forward as the answer. Competitor appearances are logged the same way. These are separate observations, reported separately, and never combined into a single score.
Every status is read from the surface’s actual answer, with the sources shaping it and verbatim quotes retained. Because answers vary, findings are described as counts and patterns across a defined prompt set, market and date range — not as a single ranking number.
Each observation is captured with its system, prompt, market and date, so it can be repeated. AI systems are non-deterministic: the same prompt can return different wording or sources between runs. Repetition and date-stamping are treated as part of the method, and running the same prompts over time is what turns single answers into a baseline.
What the assistant returned — the wording, the named brands, the cited sources — is recorded as observed fact. What it means for a brand’s priorities is analyst interpretation, labeled as such. The two are kept separate throughout.
Where an answer surface labels a result as sponsored or shopping, it is recorded as such and not counted as an organic mention, citation or recommendation.
Cited observes how a brand currently appears; it does not reveal any platform’s internal ranking logic, and it cannot guarantee future placement or recommendation. A baseline describes patterns for a defined set of questions, systems, markets and dates — a strong, repeatable signal, not a promise.
start an investigation
Start with the decision you need to make.
Tell us what is changing, what remains unclear or where the available information is not giving you confidence. We’ll determine what needs to be examined before recommending the next step.
Start an Investigation →