Beta NoticeThe AI Visibility Agency is currently in beta. Scores, rankings, methodologies, evidence models, platform functionality, and reporting outputs may change prior to general availability. Read beta terms
Methodology · v1.2 · current as of June 2026
The AI Visibility Index measures how the six leading AI answer engines represent your company when buyers ask the questions that drive revenue. This page explains the categories we measure, the disciplines we apply, and the evidence we publish behind every figure.
Our scoring engine is proprietary. Our measurement principles, evidence model, and governance posture are not — and they are documented in full below.
Principles
Every score derives from a recorded AI answer event.
We do not infer what AI engines think. We record what they say. Each figure on your dashboard traces back to a specific prompt, a specific engine, a specific model snapshot, a specific timestamp, and the verbatim response — captured and stored. Nothing is modeled from opaque belief states.
Rules out
Synthetic data. Inferred sentiment. Opinion-driven scoring.
Every score is regenerable from stored inputs.
We pin engine versions, model snapshots, prompts, locale, and timestamps. If we recompute a score from the same inputs we land within tolerance of the original figure. When upstream engines change their models, we note it, version the calculation, and publish the diff in our change log.
Rules out
Ephemeral scores. Untraceable methodology drift. Quiet model swaps.
Every score ships with the evidence that produced it.
Drill down on any number and you reach the underlying AI answer events — the actual responses, the citations the engine offered, the language used about your brand and your competitors. We show our work at the evidence level. Customers can copy verbatim language into board decks because they are reading what the AI actually said.
Rules out
Black-box dashboards. Score-only reports. Unfalsifiable claims.
Confidence framing is published with every figure.
A score with a thin evidence base reads differently than a score with deep coverage. Every figure on your dashboard ships with a qualitative confidence label — Strong, Moderate, or Developing — derived from the breadth and depth of the underlying evidence. We tell you how much to trust the number, not just the number itself.
Rules out
False precision. Overstated certainty on thin data. Confidence laundering.
You are graded against your category, not an absolute curve.
Scores are normalized against a vetted cohort of comparable companies — same sector, similar buyer journey, similar revenue band, comparable geographic footprint. A 720 in commercial real estate and a 720 in retail banking represent comparable AI visibility relative to category peers. Cohorts are reviewed quarterly and published in your dashboard.
Rules out
Apples-to-oranges comparisons. League tables that flatter big brands. Sector-blind benchmarks.
Categories
Each category is observable, evidence-anchored, and cohort-calibrated. Drill any figure on a customer dashboard and you reach the underlying AI answer events that produced it.
01
How often AI engines surface you when buyers ask in-scope questions.
What we measure
For each engine, we test a vetted set of category-defining buyer questions and record whether your company appears in the answer at all. We track surface rate across question types — informational, comparative, transactional — and across engine generations. Presence is the floor of AI visibility: nothing else matters if you are not in the room.
Why it matters
Buyers who use AI for research treat the answer as the shortlist. Companies that do not appear are not considered. AI Presence is the leading indicator of pipeline exposure.
Example buyer question
"What are the best commercial property insurers for mid-market manufacturers in the Northeast?"
Evidence we capture
Failure mode
Appearing in two of six engines, absent from the rest — buyer sees you only if they happen to use the right tool.
02
When you are surfaced, the position and framing of the recommendation.
What we measure
We distinguish 'mentioned in passing' from 'recommended,' 'listed' from 'lead,' and 'caveated' from 'endorsed.' For each presence event, we capture position within the answer, surrounding language, and whether the engine frames you as a primary, secondary, or qualifying option. Strength is the qualitative weight of each appearance.
Why it matters
A buyer reading 'industry leader' acts differently than a buyer reading 'one option to consider.' Recommendation Strength translates raw presence into decision-grade signal.
Example buyer question
"Which law firms should a Series B SaaS company use for commercial contracts?"
Evidence we capture
Failure mode
Top-of-answer mentions for competitors; you appear in the 'also worth considering' tail.
03
The quality and recency of sources AI engines cite alongside your brand.
What we measure
Every AI answer carries an implicit or explicit citation graph. We capture the domains, publications, and document types AI engines reach for when discussing your category and your company. We then assess source quality — institutional authority, editorial rigor, recency — and the share of citations that route through your owned channels versus third parties.
Why it matters
Citation Authority is the underlying evidence base AI uses to form its picture of you. Improving the sources that surface for your category is the most durable lever in AI visibility.
Example buyer question
"What clinical evidence supports [therapy class] for [condition]?"
Evidence we capture
Failure mode
Engines cite outdated press releases and a single Wikipedia paragraph instead of your current technical documentation.
04
The polarity and stance of language AI uses to describe you.
What we measure
We capture the descriptive language AI engines use when discussing your company — the adjectives, framings, hedges, and historical references. Sentiment is tracked not as a single number but as a language profile: how the engine talks about your reliability, your innovation, your leadership, your incidents, your trajectory.
Why it matters
Narrative drift is invisible until it surfaces in a buyer's screen. Brand Sentiment catches the slow rewriting of your story before it compounds into pipeline drag.
Example buyer question
"Tell me about [company] — their strengths and weaknesses."
Evidence we capture
Failure mode
Engine repeatedly references a four-year-old outage when describing your reliability story.
05
How accurately AI engines resolve your company as a real entity.
What we measure
We test whether engines correctly identify your legal entity, your products, your leadership, your footprint, your subsidiaries, your acquisitions, and your relationships. Entity Authority captures both confidence (does the engine commit to facts) and accuracy (are those facts right).
Why it matters
Buyers using AI as a fact source absorb whatever the engine asserts. Entity errors propagate into procurement memos, RFP responses, and analyst notes. Grounding AI in correct facts is a board-level concern.
Example buyer question
"Who is the CEO of [company] and where are they headquartered?"
Evidence we capture
Failure mode
Engines list a former executive as current, or confuse you with a similarly named competitor.
06
Your win rate against named peers in head-to-head buyer questions.
What we measure
We construct head-to-head comparison prompts — your company against each named peer in your cohort — across the full set of buyer decision criteria. We then track the share of those head-to-heads where AI engines frame you as the preferred or recommended option, and we capture the language used to justify each verdict.
Why it matters
A buyer's final shortlist is built from comparison questions. Competitive Position is the closest proxy to AI-driven win rate available in the public answer layer.
Example buyer question
"Should a regional hospital system choose [you] or [competitor] for [solution]?"
Evidence we capture
Failure mode
Engines consistently route 'enterprise' framing to a competitor while you win 'startup' framing.
07
How directly your owned content answers the questions buyers actually ask AI.
What we measure
We map the buyer-question taxonomy for your category against your owned content footprint — site, docs, knowledge base, blog, press. Content Answerability captures whether your content is structured in a way AI engines can extract, attribute, and use as a citation. Coverage is one dimension; extractability is the other.
Why it matters
Engines cite content they can parse. Content Answerability is the single largest lever a company controls directly: the gap between 'we have an answer somewhere' and 'AI engines find and quote it' is where most AI visibility budgets are wasted.
Example buyer question
"What's the implementation timeline for [your product] in a 200-bed hospital?"
Evidence we capture
Failure mode
You have a great FAQ buried in a PDF that no engine ever reaches.
08
The breadth of buyer questions where you have any presence at all.
What we measure
We maintain a category-specific question library — typically several hundred prompts spanning the buyer journey from awareness through procurement. Prompt Coverage measures the share of that library where you appear in at least one engine's answer. It is the breadth dimension of AI Presence.
Why it matters
Concentrated presence in a handful of prompts is fragile; broad presence is durable. Prompt Coverage exposes the blind spots where competitors are quietly compounding their lead.
Example buyer question
"[Full library of category prompts, refreshed quarterly]"
Evidence we capture
Failure mode
Strong presence on five flagship prompts; absent on the eighty buyers actually ask.
09
Your standing among the institutions AI engines rely on.
What we measure
AI engines lean heavily on a small set of curated sources for category-level authority — analyst reports, regulator publications, industry rankings, trade-association lists, peer-reviewed publications. We map which of those sources reference your company, in what context, and how recently. Industry Authority captures whether the institutions that matter to your category are vouching for you.
Why it matters
Authority is sticky. A single inclusion in a respected analyst report can shape AI framing for years. Industry Authority shows whether the structural sources of credibility in your category are working for you or against you.
Example buyer question
"What does [respected analyst] say about [category]?"
Evidence we capture
Failure mode
Engines cite a competitor's analyst placement from last quarter; your last mention is from 2022.
10
Third-party verification AI engines weigh heavily.
What we measure
AI engines reach for verifiable trust markers — compliance posture (SOC 2, ISO, HIPAA, PCI), schema completeness, certifications, awards, customer logos, public case studies, and editorial freshness. Trust Signals measures the surface area and consistency of those markers across the sources AI engines actually crawl.
Why it matters
The credibility gap is the most common reason a technically capable vendor loses an AI-driven shortlist. Trust Signals identifies the missing markers that explain why engines hedge when describing you.
Example buyer question
"Is [company] enterprise-ready and compliant for [regulated use case]?"
Evidence we capture
Failure mode
Your SOC 2 report exists, but AI engines cite a stale compliance page that omits it.
Each category contributes to your composite AI Visibility Score (AIVS) on a 0–1000 scale. The contribution model is proprietary; the categories, evidence, and outcomes are documented above.
Evidence
Every figure on your dashboard reduces to a population of recorded AI answer events. Each event is a single transaction with a single engine, and we store enough about it to reconstruct it, attribute it, and audit it.
Prompt
The exact question text, the prompt variant ID, and the buyer-journey stage it maps to.
Engine
Which of the six engines responded, plus the model snapshot identifier — pinned so the answer is reproducible.
Response
The verbatim text the engine returned. Stored in full, not summarized.
Citations
Every URL, document, and source the engine surfaced — captured with domain, publication date, and authority tier.
Context
Locale, session settings, account state, and any other parameter that materially influenced the answer.
Timestamp
When the event occurred, to the second. Engines drift; timestamps let us locate that drift.
Entities
Every named company, product, person, and place the answer references — yours and your competitors'.
Language profile
Adjectives, hedges, qualifiers, and framing markers extracted from the response.
We do not store any personally identifiable information about the buyers whose questions our prompts emulate. All prompts are synthetic representations of category buyer behavior, validated against anonymized aggregate query patterns.
Cohort calibration
A cohort is a vetted set of comparable companies — same sector, similar buyer journey, similar revenue band, comparable geographic footprint. Cohorts are not "all companies that share a NAICS code." They are hand-curated by our research team, reviewed quarterly, and published in your dashboard with full peer-list visibility.
For most clients, cohort size sits between 12 and 40 peers. Cohorts are intentionally narrow: too broad and the comparison loses decision value; too narrow and statistical signal weakens.
AI visibility varies enormously by category. A 700 in a category where the leader scores 740 is excellent; a 700 in a category where the leader scores 920 is a warning. Calibrating to cohort makes the number actionable instead of theatrical.
Every figure on your dashboard is dual-labeled — your absolute score, and your cohort-relative position (e.g., "top quartile, above peer median"). The relative position is the figure that should drive decisions.
Cohort governance
Confidence framing
A score backed by hundreds of recent answer events reads differently than a score backed by a thin or stale evidence base. We surface that difference directly, using a three-tier qualitative label that appears next to every figure on every dashboard.
Strong
Dense, recent, multi-engine evidence base. Use this figure directly in board materials, procurement responses, and strategic planning.
Moderate
Adequate evidence for trend interpretation and prioritization; treat point estimates with caution and prefer comparisons over absolutes.
Developing
Limited or recently established evidence base. Useful as an early signal; not yet reliable enough to drive strategic commitments on its own.
We deliberately publish confidence as a label rather than a numeric interval. Numeric uncertainty implies a precision the underlying phenomenon — AI engine behavior — does not yet support. Labels force the right conversation: is this figure good enough for the decision in front of you?
Refresh
Continuous
AI answer events are recorded continuously as engines emit them across our prompt library.
Daily
Scores recompute on a daily cadence for Growth tier and above. Starter tier refreshes weekly.
Weekly
A two-page PDF summarizing material movements, hallucinations, and recommended actions.
Quarterly
Cohort lists, prompt libraries, and category definitions are reviewed and versioned each quarter.
Every figure on your dashboard carries a "last computed" timestamp. When upstream engines ship new model versions, recomputation is triggered automatically and the affected figures are flagged in your next executive brief.
Signature deliverable
A time-stamped record of every material misrepresentation the six leading AI engines emit about your company — captured verbatim, attributed to the responsible engine, classified by severity, and tracked from first observation to resolution.
What gets logged
How it ships
The ledger lives in your dashboard, exports to PDF for legal and comms review, and feeds your weekly executive brief. Each entry carries the verbatim AI response, the engine, the timestamp, and a recommended remediation path — from owned-content correction to formal engine-vendor escalation.
Why this matters
Hallucinations propagate. A misstatement absorbed by one engine and quoted by a buyer becomes the buyer's belief. The ledger is how you find them before they reach procurement.
Coverage
AI visibility is not one market. Each engine has a different user base, citation behavior, and reasoning posture. We measure all six independently, then surface both per-engine and composite readouts.
ChatGPT
Highest consumer reach; broad buyer-journey coverage from research through procurement. Citation behavior varies sharply by model version.
Claude
Strong enterprise and technical buyer adoption; tends to ground answers in cited sources more aggressively than peers.
Gemini
Deepest integration with workplace tools; surfaces in answers buyers see inside the apps they already use.
Perplexity
Citation-native by design; the most direct measurement of source-to-answer attribution for any engine.
Copilot
Embedded across enterprise software; high weight in regulated-industry and large-account buyer journeys.
Grok
Newer evidence base; rising signal especially in consumer and creator-economy categories.
We add engines when they reach material buyer adoption, and we version engine coverage in our change log. Removal of an engine — should one occur — would also be versioned and announced.
Governance
Enterprise customers receive a per-tenant audit lake — a full, immutable record of every AI answer event, score recomputation, cohort change, and dashboard interaction. The audit lake is exportable and queryable, designed for regulated industries where every figure must be defensible.
Every change to the measurement framework — new categories, cohort revisions, prompt-library updates, engine additions — is timestamped, versioned, and published in the methodology log below. Customers are notified of material changes before they take effect.
Version log
Every change to the measurement framework is timestamped and published here. Nothing changes quietly.
v1.2
June 2026
Cohort review (Q2 2026); engine snapshot refresh across all six providers; expanded Industry Authority source list for Healthcare and Legal sectors.
v1.1
March 2026
Confidence framing system promoted to a published deliverable; refresh cadence formalized per tier.
v1.0
January 2026
Initial methodology release. Ten categories, six engines, six sector portals.
Limitations
The AI Visibility Index is a rigorous measurement of how the six leading AI answer engines represent your company. It is not a complete model of buyer behavior, and we are direct about that.
What we measure well
What we don't claim
Our customers use the AIVS as one of three or four strategic signals — not as the single source of truth. We design the methodology, the deliverables, and the language on this page with that posture in mind.
Validation
A standing council of category-leading researchers, former regulators, and senior practitioners reviews the methodology quarterly. Council membership is published in your client dashboard and rotates on a three-year cadence.
An independent audit of our measurement framework is scheduled for Q4 2026. Scope, auditor, and publication terms will be announced in the methodology log before the engagement begins.
We collaborate with academic researchers studying retrieval, attribution, and AI-mediated information ecosystems. Selected findings are published as research notes on this site; primary customer evidence is never shared without explicit permission.
FAQ
Two reasons. First, the contribution model is the core intellectual property of the platform — publishing it would invite gaming and degrade signal quality for every customer. Second, naive weight publication implies a precision the underlying phenomenon does not yet warrant. We publish the categories, the evidence behind every score, and the qualitative confidence framing — which together give you a more honest picture than a number with a false-precision decimal point.
SEO measures retrieval from search indexes. Share-of-voice measures media mention volume. Neither measures what an AI engine actually says when a buyer asks a category question. AI engines synthesize, recommend, hedge, and exclude — behaviors that don't map to keyword rankings or press-clip counts. The AI Visibility Index measures the output layer buyers now read instead of search results.
Evidence capture is continuous. Dashboard scores refresh daily for Growth tier and above, and weekly for Starter. Every figure carries a 'last computed' timestamp so you always know how fresh the number is.
We re-pin the snapshot, trigger a recomputation, and flag the affected figures in your next executive brief. Engine model changes are logged in our public methodology changelog so the diff is auditable.
The category prompt libraries are versioned and reviewed with your account team during cohort review. We do not publish the full prompt libraries openly — that would invite content engineering targeted at specific prompts rather than at the underlying buyer questions they represent.
Our research team selects a peer set based on sector, buyer-journey similarity, revenue band, and geographic footprint. We share the proposed peer list with you, you can request additions or argue inclusions, and the final cohort is logged and dated. Cohorts are reviewed every quarter and adversarial peers stay — we won't remove a peer simply because it flatters your score.
Strong means the evidence base behind a figure is dense, recent, and multi-engine — use the number in decisions. Moderate means evidence is adequate for direction and prioritization but treat point estimates carefully. Developing means evidence is limited or newly accumulated — useful as an early signal, not yet a basis for strategic commitments. Labels are published next to every figure on every dashboard.
Because, together, they account for the overwhelming majority of buyer-facing AI answer traffic in the categories we serve. Coverage expands as new engines reach material adoption — and additions are versioned in the changelog.
Every material misrepresentation is captured in your hallucination ledger — verbatim, attributed, timestamped, classified by severity, and routed with a recommended remediation path. The ledger is one of the platform's signature deliverables.
No. Your evidence base, scores, and reports are never used to train any third-party AI model. Our prompts are synthetic representations of category buyer behavior and contain no customer-identifying information.
By default, in the US. EU-only residency is available on Pro tier and above. No customer evidence data crosses configured residency boundaries. The sub-processor list is published and versioned.
Yes. We routinely deliver methodology briefings to procurement, risk, legal, and audit teams as part of a standard onboarding. Request one through your account team.
Glossary
AIVS
The AI Visibility Score — a 0–1000 composite figure summarizing how the six leading AI engines represent your company. Always paired with category readouts and a confidence label.
AI answer event
A single recorded transaction with one AI engine: a prompt, the verbatim response, the citations, and the metadata required to reproduce the event.
Cohort
A vetted set of comparable companies against which your scores are normalized. Reviewed quarterly with the client and visible in the dashboard.
Confidence label
A qualitative tag — Strong, Moderate, or Developing — published next to every figure to communicate how much weight a customer should place on it.
Citation graph
The structure of sources an AI engine reaches for when answering a category question. Surfaced in Citation Authority.
Composite score
The AIVS itself — a single number aggregating performance across the ten measurement categories.
Engine snapshot
A specific pinned model version for a given AI engine, used to make scores reproducible across time.
Evidence base
The full population of AI answer events that underlies any given figure. The thicker the evidence base, the stronger the confidence label.
Hallucination
A material misrepresentation by an AI engine — factual error, outdated reference, misattribution, or confused entity. Logged in the hallucination ledger.
Hallucination ledger
A time-stamped, classified record of every material AI misrepresentation observed about your company, delivered as a continuous deliverable.
Head-to-head prompt
A buyer-question variant pitting your company directly against a named peer in your cohort. Used to compute Competitive Position.
Industry authority source
A curated reference institution — regulator, analyst, ranking body, peer-reviewed publication — that AI engines treat as a high-credibility source for the category.
Methodology log
The public, timestamped changelog of every modification to the measurement framework.
Owned content footprint
The set of content assets you control — site, docs, knowledge base, press — that AI engines can use as citation material.
Peer-relative position
Your standing within your cohort on a given category, surfaced as quartile and median-relative language rather than as a percentile decimal.
Prompt library
The versioned, sector-specific set of buyer questions used to elicit AI answers for measurement. Reviewed quarterly.
Refresh cadence
How often a given dashboard figure is recomputed — continuously for evidence capture, daily or weekly for dashboard scores depending on tier.
Source quality tier
A classification applied to each cited domain — regulator, analyst, trade press, owned, social — used to assess Citation Authority.
Strength bar
The qualitative visual on category cards — Strong, Moderate, Developing — that replaces numeric category-level weighting in customer-facing reports.
Trend arrow
The directional indicator (↑ rising, → steady, ↓ declining) attached to every figure to communicate movement without exposing point-to-point math.
Talk to us
Book a working session with our research team. We'll walk your category, show you live evidence from the six engines, and tell you — honestly — whether the AI Visibility Index is the right instrument for the decisions you're trying to make.