Question panel
A defined set of questions and prompts, tied to your intents and competitors. It stays fixed over the period so that two windows remain comparable.
Methodology
How Aive measures AI visibility.
Aive measures observed answers, not a universal truth. Every indicator is calculated within an agreed scope, with its formula, denominator, period and variance. Every raw answer is archived to serve as evidence. The limits and statuses of every offer are grouped on this page.
Last verified · 6 October 2026
QuestionEngineRaw answerIndicatorVariance
How does Aive measure visibility in AI answers?
Aive queries the agreed engines on a defined question panel, per market and language, at a set frequency and with several samples per question. Every raw answer is archived with its date, engine and access method. The indicators (mention, citation, AI share of voice, coverage) are calculated on these observed answers, always with their denominator, period and variance.
- Observed answer
- An answer actually collected from an engine, dated and archived with its question, access method and sources.
- Repeated sampling
- Asking the same engine the same question several times within one period, to measure how much the answers vary.
- Variance
- The spread of an indicator across samples and days. It is always displayed alongside the score.
The protocol
An agreed scope, verifiable deliverables.
A GEO Audit starts with a written scope: agreed questions, engines and markets. Nothing is measured outside it, and nothing is extrapolated beyond it.
01Agreed scope
- Brand or domain
- Questions and prompts
- Engines and channels
- Markets and languages
- Competitors
- Video assets and content
- Period and data sources
02Protocol
- Querying via provider API or observation protocol
- Repeated sampling at a set frequency
- Archiving raw answers
- Extracting mentions, citations and sources
- Calculating indicators and variance
03Deliverables
- Dated raw answers
- Mentions, citations and sources
- Indices with formula
- Confidence level
- Competitors and priorities
- Action plan and follow-up measurement
Engines and observation method
Provider API or observation: always kept apart.
Depending on the engine and configuration, Aive uses the provider’s API or an interface observation protocol. The method is stated on every line of the report. Measured: dated observation protocol. Monitored: regular observation, partial coverage.
| Environment | Status | Method shown | Evidence required |
|---|---|---|---|
| ChatGPT | Measured | Dated observation protocol | Model, date, parameters, region and cadence |
| Gemini | Measured | Dated observation protocol | Model, date, parameters, region and cadence |
| Claude | Measured | Dated observation protocol | Model, date, parameters, region and cadence |
| Perplexity | Measured | Dated observation protocol | Model, date, parameters, region and cadence |
| Google AI Overviews | Monitored | Observation, partial coverage | Country, interface, query, date and method |
An answer obtained through an API is never presented as the experience of every user of an interface. Same matrix as the Technology page. Grok, DeepSeek and Microsoft Copilot are under study and stay outside the public matrix; the engines observed in a GEO Audit are agreed at scoping.
Panels, markets and frequency
What makes two measurements comparable.
Markets and languages
Each question is asked in the agreed market and language. A result in French for France holds neither for another market nor for another language.
Frequency and repeated sampling
The same question is asked several times, at a set frequency. The spread across samples and days gives the variance.
Archived raw answers
Each answer is kept with its question, engine, access method, date and cited sources. It serves as evidence for every indicator.
Indicator formulas
Every rate shows its denominator.
The formulas are public. No target value is published: a GEO rate is read per engine, per period and with its variance.
| Indicator | Formula | How to read it |
|---|---|---|
| Mention rate | Answers mentioning the brand ÷ observed answers | Per engine and period. Directional, not universal. |
| Citation rate | Answers citing a brand URL or source ÷ observed answers | Per engine and period. A citation is not a click. |
| AI share of voice | Brand mentions ÷ mentions across the competitive set in the sample | Depends on the declared competitive set. |
| Prompt coverage | Active prompts with a valid observation ÷ defined prompts | The denominator is always shown. |
| Stability / variance | Spread across samples and days | Always displayed with the score. |
| Observed referred sessions | Sessions from identifiable AI referrers | Kept separate from crawling and citations. |
Definitions aligned with the Aive glossary. Assisted conversions are only reported with a declared attribution model.
Sample output
One report, read line by line.
Demo data: anonymised brand, fictitious panel and engines. The report shows the reporting format, not a customer result.
Sample report · anonymised brandGEO baseline · home appliances · UK
Demo data Period2 windows of 14 days
Panel40 questions · 3 engines
Repeated sampling5 samples per question
DateExample, undated
| Engine and method | Observed answers | Mentions | Citations | Variance |
|---|---|---|---|---|
| Engine A · provider API | 40 | 18 | 11 | ± 6 pts |
| Engine B · provider API | 40 | 15 | 7 | ± 8 pts |
| Engine C · observation | 40 | 12 | 9 | ± 9 pts |
Archived raw answer
Which quiet washing machine should I pick for a flat?
“… models tested under 48 dB, such as the X range, are often recommended [2] …”
[2]example-brand.co.uk/quiet-guide · video moment 01:12
Reading and limits
Citation rate = answers citing a brand URL ÷ observed answers, per engine and period. An API answer does not represent every user’s experience. Citation ≠ click; no causality is established without a dedicated protocol.
Evidence levels
From observation to causality, six distinct levels.
Every statement in an Aive report is tied to one level. No level ever licenses the conclusion of the next.
- 01ObservationWhat Aive queried, collected or received: questions, engines, raw answers, cited sources and dates.
- 02InterpretationWhat scores and analyses indicate, within a given scope and with their variance.
- 03RecommendationWhat your team could change or produce. A recommendation is not an action.
- 04ActionWhat is actually approved, generated, published or connected, with its log.
- 05Observed outcomeThe change in mentions, citations, traffic, engagement or conversions between two comparable windows.
- 06CausalityPut forward only when a sufficiently robust protocol supports it.
Two audit levels
An exploratory preview, or a real measurement.
Level 1 · Free check
A first reading with no meeting and no account, from a domain or URL. The result is an exploratory estimate: not a real observation of the engines, not a competitive benchmark, and no promise of citation.
Test my site for free
Level 2 · GEO Audit
A scoped engagement on agreed questions, engines and markets: dated raw answers, mentions, citations and sources, indices with formula, confidence level, competitors, priorities and an action plan. No fixed turnaround is promised.
Request a GEO Audit
Video GEO and moving from SEO to GEO
What is measured, what is controlled, and what you need to start.
This block gathers what the Video GEO and From SEO to GEO pages point to here: measurement, controls and prerequisites.
What is measured
- Mentions, citations and sources per tracked question
- Internal linking: orphan pages, depth, approved links
- Crawl, publishing and first-party engagement
- Period, denominator and coverage of every rate
Controls
- Every statement tied to a verifiable source: timestamped moment, document or page
- Review and approval before publishing
- Versions and publishing log
- Separate roles: writing, approval, publishing
Prerequisites
- Video as a file, URL or YouTube; formats, size and duration specified before import
- Analysis and reuse rights on every video
- CMS access, confirmed per connector
- Agreed question panel, and analytics to link results
Limits and statuses
Limits and statuses, in one place.
Product pages and journeys stick to the essentials; their limits and statuses are gathered here, once. Each line applies to every offer concerned.
Measurement and results
- No citation, ranking, visibility, audience or revenue is promised.
- Citation ≠ click; referral ≠ causality: a correlated lead or session proves nothing without a robust protocol.
- Causal ROI requires declared instrumentation and protocol; a linked conversion alone does not establish incremental effect.
- Aive does not control AI engines: it structures content that is easier to understand, verify and reuse.
- Free check ≠ real measurement: only a GEO Audit, scoped on agreed questions, engines and markets, sets a dated baseline.
- No delivery time is promised beyond what scoping sets.
- Video GEO prepares content; it does not guarantee that an engine will cite it. Valid JSON-LD guarantees neither rich display nor citation.
Content, publishing and rights
- Generating is neither publishing nor proving: every piece is reviewed and approved by your team.
- Publishing only targets activated destinations, with a log.
- EDGE deploys nothing on its own: every change is previewed, approved, reversible and subject to your authorisations.
- Semantic internal linking does not run on its own: every suggested link is approved.
- Video GEO output quality depends on audio, video and metadata quality; extracted statements are proposals to verify, not established truths.
- Discovering content grants no rights; publishing triggers no payment.
- YouTube import depends on the media type and the rights on the channel.
Statuses by plan and contract
- CMS, e-commerce and destination connectors: no universal connection, each one is confirmed individually, tested and documented.
- Enterprise API and Agency offer: on quote, with scopes, limits and support defined in the contract; webhooks and SSO depending on plan.
- AI shopping (360 product pages, recommendations): subject to catalogue connectors and workspace.
- Aive Expert: subject to the authorised corpus and plan; every answer is checked against its source.
- Video production (clips, voice-over, variants): subject to credits, rights and contract.
- Licensing and revenue sharing: eligible, active, contracted programmes only; no rate or threshold published, no payment triggered by a citation.
- No auction, wallet, top-up or automatic payout.
GEO Measurement
Rules of evidence and reading limits.
The GEO Measurement page keeps to the essentials; its rules of evidence, its two audit levels and its limits are gathered here.
What each signal establishes, and nothing more
- An AI answer is a signal observed in a panel and on a date, not the experience of every user.
- A sentiment qualifies how an answer is framed within a protocol, not a psychological truth.
- A citation is neither a click nor a conversion.
- A first-party referral establishes an observed visit, not the question that caused it.
- A linked conversion does not, on its own, establish an incremental effect.
- A crawl is not a citation; a missing log is not a missing visit.
Free check or GEO Audit
- Free check: a first reading, no meeting needed; one domain or URL; reading the page, the HTML rendering and structured data; indicative priorities.
- GEO Audit: a verifiable, dated baseline; brand or domain, questions, engines, markets, languages, competitors, video assets and period; repeated sampling at an agreed frequency, score with formula and confidence level; dated report with raw answers, mentions, citations, sources, competitors and action plan.
- The free check does not query the engines: it remains an exploratory estimate, neither a real observation, nor a benchmark, nor a citation guarantee.
Limits
- Measurement covers the agreed panel, not every possible question nor every engine.
- An answer obtained through an API may differ from what a signed-in user sees.
- Share of voice is relative to the chosen competitive set.
- Each score shows its method, date, inputs and coverage; no universal success score is published.
- No measurement guarantees a citation, a ranking or revenue.
Editorial approval
AI prepares. Your team decides.
A human review between every prepared piece of content and its publication: gates, corrections, approvals, separation of duties, snapshots and log. By default, nothing is published without approval; steps and roles are configured per workspace.
Roles
- The roles allowed to approve are defined in your workspace.
- Separation of duties: the person who writes is not the one who approves the same content.
- Disabling a gate is an explicit decision by your organisation, recorded in the log.
Approval steps
- Drafting: the prepared content arrives with its statements and their sources, tied to the video moment.
- Review: the reviewer corrects, comments or sends it back to drafting.
- Approval: a distinct role approves; a snapshot freezes the approved version, kept for audit and rollback.
- Publishing: only an approved version goes to an enabled destination.
Log
- Every decision is timestamped and attributed: you know who approved what, when, and on which version.
- The log covers actions taken in Aive, not those taken outside the platform; export options depend on the plan.
- An approval attests a human decision, not the absolute accuracy of a piece of content; the quality of the review depends on the reviewers and the rules defined.
Media Factory
Publishing modes and guardrails.
Rights checked per video, editorial approval before any go-live, full publishing log.
Publishing modes, one log
- Draft: the content is placed in the destination without being visible; your team finalises it there.
- Scheduled: the content goes out on the chosen date; the send can be cancelled before the deadline.
- Published: the approved content is published; the log records the author, time and version.
- Every action is attributed, timestamped and, depending on the destination, reversible; available modes vary by destination.
Guardrails
- A demo is not a publication: nothing goes live without your approval.
- Media Factory guarantees neither audience, nor monetisation, nor rankings.
- A public video is not rights-free: music, images and speakers still need checking.
- The Expert answers from the authorised corpus; an answer is verified in its source.
- Domain control limits Player embedding; it does not prevent every capture.
Value levers, under conditions
- Loyalty: collections, progression, editorial recommendations and reserved experiences, within the scope of your plan.
- Discovery: structured pages, VideoObject, chapters and internal links; no ranking or citation promised.
- Licensing: a catalogue prepared with provenance, metadata and restrictions; explicit rights, consent and contract.
- Revenue: media programmes and contextual partnerships, only under an eligible programme and an active contract. No revenue is automatic.
Automation
From template to a logged run.
What a workflow run makes readable, and what it does not prove.
One run
- Inputs: URL, video, catalogue or queries to track; template nodes and instructions; approval rules and schedule.
- Flow: nodes run in order, Yes / No conditions based on outputs, stops at approval points.
- Outputs: assets prepared, approved and delivered; history available per run; failure alerts, depending on configuration.
- When a node fails, the run stops and the error is logged in the history.
What every action makes readable
- Who ran what, when, with which inputs and which outputs.
- Every figure from the agent links to its data: tracked engines, questions, period and sources.
- Roles, workspace isolation, logs, secrets and budget frame every action.
Limits
- Nothing is published without approval from an authorised role.
- A successful workflow proves the steps ran, not that the content produced will be cited.
- The agent answers from your workspace data and the sources it cites.
- Available nodes and actions depend on the enabled modules and destinations.
- A listed connector is not enabled by default; availability is confirmed connector by connector.
EDGE
Mechanism, authorisations, and what EDGE does not do.
EDGE applies page recommendations without changing your site all at once; the integration mode and scope depend on the authorisations granted on your infrastructure.
Mechanism
- Inputs: approved recommendations, chosen page scope, granted technical authorisations.
- Flow: previewing the change, progressive rollout across the scope, observation and human decision.
- Outputs: pages updated within the approved scope, rollout log, rollback available.
What EDGE does not do
- EDGE is not an autopilot: no change is applied without human intervention.
- A change observed after rollout does not establish causality without a suitable protocol.
- The scope depends on the authorisations granted and on your architecture.
- EDGE does not replace your release processes for deeper changes.
- EDGE applies recommendations; it ensures neither citation nor ranking.
GEA
A governed framework, and what GEA does not promise.
GEA pairs a brand message, labelled as sponsored, with the AI answers of Aive media: Aive Expert, answer pages and network content.
Mechanism
- Inputs: questions asked to the Aive Expert and network answer pages; topics, intents and contexts identified by MGT; brand rules, excluded contexts and budgets.
- Flow: pairing a relevant message with the answer context, approval by the brand and by the media, labelled and logged delivery.
- Outputs: sponsored messages in answers; measurement of impressions, interactions and contexts; revenue share with carrier media, under an eligible programme and an active contract.
What GEA does not promise
- GEA delivers in the answers of Aive media; Aive does not sell space in ChatGPT, Gemini or other third-party assistants, and does not control them.
- No delivery volume and no ROI are guaranteed.
- A sponsored message is always labelled and never changes the content of the answer.
- No personal user data is used for targeting.
- An observed change does not prove causality without a suitable protocol.
Frequently asked questions
Does Aive query the engines directly?
Depending on the engine and configuration, Aive uses the provider’s API or an interface observation protocol. The method is stated for each engine in the report, with the model or surface, date, region and cadence. The two methods are never presented as equivalent.
Why ask the same question several times?
Generative answers vary from one sample to the next and from one day to the next. Repeated sampling measures that spread: a rate is read with its variance. A change smaller than the observed variance should not be interpreted as a trend.
Does a higher citation rate mean more traffic?
Not necessarily. A citation is not a click, and a session referred by an AI does not establish causality. Aive measures citations, referred sessions and conversions separately, and only links an outcome to an action under a declared protocol.
What is the difference between the free check and the GEO Audit?
The free check is an exploratory estimate from a domain or URL, with no account. The GEO Audit rests on agreed questions, engines and markets, dated raw answers, published formulas and a confidence level. Only the latter establishes a verifiable baseline.
Are raw answers delivered?
Within the scope of a GEO Audit, dated raw answers are part of the deliverables. Each is archived with its question, engine, access method, date and cited sources, and serves as evidence for the report’s indicators.