Define before testing
Agree the question, target, provider, surface, market, language and sample plan before the scan begins.
VideoGEO methodology
AI visibility measurement should show its workings.
AdCyte records the question, target, provider, sample, answer and available sources behind every observation.
No mystery score. No hidden denominator. No claim that a page edit caused what happened next.
Current methodology · Version and update record pending publication
Measurement principles
Observe carefully. Report precisely.
External AI answers are variable. A useful methodology does not remove that uncertainty. It records enough context for the result to be understood, compared and challenged.
Keep the test independent
Send the approved question without secretly supplying the desired answer, transcript, target facts or preferred sources as extra context.
Retain the answer
Store the returned text, available citations, date and relevant settings behind every classification.
Separate different events
A mention, recommendation and citation are different observations. One does not imply another.
Keep missing data visible
A failed request or unsupported response is not the same as an eligible answer in which the target was absent.
Compare without claiming causation
Repeat scans can show that sampled answers changed. They cannot prove why an external system changed them.
The complete workflow
From a campaign question to an inspectable observation.
The method connects what the video supports, what the brand publishes and what an external AI-answer service returns. Each stage produces a record that can be reopened.
Fictional methodology demonstration
Stage 01 · Define
Set the benchmark
Approve the exact question, named targets, question intent, market, language, provider, surface and planned samples.
Record created
A stable benchmark revision.
Meridian GT observation record
Question: Which premium electric GTs are suitable for long-distance driving? Target: Meridian GT · unprompted Plan: 3 samples
Test definition
Decide what counts before running the scan.
Approved question
Use the exact question agreed for the benchmark. Editing its wording creates a new revision.
Question intent
Record whether it is factual, instructional, evaluative, comparative, navigational, entertainment-led or seeking a specific video.
Named target
Define the brand, product, person, character, video or tracked page being observed.
Prompted or unprompted
A target named in the question must not be presented as spontaneous discovery.
Provider and surface
Record the actual service and surface. An API response is not proof of what every consumer interface displayed.
Market and language
Retain the declared language and relevant market where supported.
Sample plan
Keep planned, attempted, eligible and failed counts separate.
Fictional methodology demonstration
Build a benchmark record
Benchmark revision
Which premium electric GTs are suitable for long-distance driving?
- Target
- Meridian GT · unprompted
- Provider surface
- Fictional configured service
- Plan
- 3 samples
- Live request
- Not run
Connected coverage
Know which gap you are looking at.
A missed AI observation does not automatically mean the public page needs more copy. Evidence, writing, publication and observation are independently assessed conditions, not an additive funnel.
Fictional methodology demonstration
Evidence · Supported
A separate range-and-charging film supports the long-distance answer.
Source moment and approved reference retained.
Evidence states
Supported · Partial · Conflicting · Not supported · Not assessed
Publication states
Published · Published revision differs · Unable to check
Observation states
Observed · Not observed · Failed · Unsupported
External sampling
Ask the approved question without feeding the answer.
The visibility test should measure the external answer environment, not AdCyte’s ability to prompt it towards the desired response.
For each sample, AdCyte sends only the approved question and neutral settings required by the configured provider. It does not add the source transcript, selected evidence, approved facts, target name, preferred answer or official domains as hidden persuasive context.
Declared official domains may be used later to classify returned citations. They are not supplied unless already part of the approved question.
Several authorised samples can reveal variability. They do not turn the result into a market-wide visibility measure.
What each sample retains
- Exact approved question and revision
- Confirmed target or targets
- Provider and surface
- Available model and retrieval settings
- Start and completion time
- Returned answer
- Genuine citation metadata made available by the provider
- Eligible, failed or unsupported status
- Evaluation revision and supporting answer passages
Denominators
Keep failures and absences separate.
Every rate should show its numerator and eligible denominator. Attempted requests that failed or could not be evaluated remain visible but do not become negative observations.
| Result | Included in eligible denominator? | Meaning | Treatment |
|---|---|---|---|
| Eligible | Yes | A usable sampled answer was returned | Classify and retain |
| Not observed | Yes | An eligible answer contained none of the defined target events | Retain as an eligible negative observation |
| Failed | No | The provider or request failed | Record separately with the failure state |
| Unsupported | No | The response could not be evaluated reliably | Record separately with the reason |
| Cancelled | No | The planned sample did not complete | Record separately where supported |
Reporting example
2 mentions from 3 eligible answers · 4 samples attempted · 1 failed
Never simplify this to 50% visibility or treat the failed sample as an absence.
Change the sample record
Eligible
Mentions
Failed
2 mentions from 3 eligible answers · 4 samples attempted · 1 failed
Educational demonstration only. No data is saved.
Observation classes
Mention, recommendation and citation are different things.
Presence and recommendation
Neutral mention
The confirmed target appears without being actively recommended or discouraged.
Positive recommendation
The answer actively suggests the confirmed target.
Conditional recommendation
The answer suggests the target for a stated need, audience or condition.
Negative advice
The answer advises against the target or identifies a material reason not to choose it.
Ambiguous identity
The wording may refer to the target, but identity cannot be confirmed reliably.
Not observed
The eligible answer contains no confirmed mention of the target.
Citation classification
Citation is assessed separately from presence and recommendation.
Owned-domain citation
A page controlled by the target organisation.
Exact tracked-page citation
The specific page or video URL registered for the benchmark.
Third-party citation
An independent source.
Unresolved citation
A source that cannot be classified reliably from available metadata.
Fictional methodology demonstration
Fictional sampled answer
Why this classification
The confirmed target is present. This records a mention, not an endorsement.
Similar titles, shared-platform domains or matching snippets do not establish that the exact tracked page was cited.
How the target is represented
Being present is not the same as being represented accurately.
Where approved reference facts or expected answer points exist, AdCyte can compare them with statements in the sampled answer.
01
Agrees with the approved reference
02
Potential discrepancy
03
Important qualification omitted
04
Relevant point absent
05
Insufficient reference information
06
Not assessed
Show both the answer passage and the approved reference behind the assessment. General model knowledge is not used as silent proof that either statement is correct.
From observation to action
Fix the part of the chain you control.
AdCyte should not respond to every missed observation by generating more copy. The correct action depends on where the gap sits.
| Diagnosed condition | Appropriate next action |
|---|---|
| The source does not support the answer | Record the evidence gap or brief new content |
| The evidence exists but the writing omits it | Prepare a targeted written update |
| The writing is ready but not public | Publish or export the approved revision |
| The live destination differs | Update or recheck the tracked page |
| The answer is public but was not observed | Retain the observation and consider a later comparable scan |
| The sampled answer conflicts with approved facts | Inspect the answer, sources and reference facts before proposing a response |
| The request failed or was unsupported | Resolve the measurement issue rather than treating it as a content gap |
Every proposed action retains the question, target, supporting evidence, affected revision and observation that led to it.
Comparison over time
Compare like with like.
A repeat scan is useful only when the underlying benchmark remains sufficiently stable. AdCyte compares the exact question, confirmed target, provider, surface, language, market and relevant settings before presenting a change.
Comparable
The material benchmark conditions match closely enough for the observations to be viewed together.
Partly comparable
One or more conditions changed and the limitation is shown beside the comparison.
Not comparable
The question, target, provider or surface changed materially. Show the observations separately.
Fictional methodology demonstration
01 · Initial observation
1 of 2 eligible answers mentioned Meridian GT
02 · Public-page update
Approved long-distance answer published
Intervention recorded separately03 · Later observation
2 of 3 eligible answers mentioned Meridian GT
Change observed after publication update. Causation not established.
Worked example
One question, shown from source to observation.
Fictional methodology demonstration
Question
Which premium electric GTs are suitable for long-distance driving?
Tracked target
Meridian GT · unprompted
Evidence
The launch film supports the hand-finished interior and 0 to 60 claim. A separate range-and-charging film supports the long-distance question.
Written
The selected campaign copy answers the long-distance question using both reviewed sources.
Published
The approved answer appears on the fictional Meridian GT campaign page.
Observed
Three attempted · two eligible · one failed. One conditional recommendation; third-party citation; tracked page not cited.
Classification
- Unprompted target mention: 1 of 2 eligible answers
- Conditional recommendation: 1 of 2 eligible answers
- Exact tracked-page citation: 0 of 2 eligible answers
- Third-party citation: 1 of 2 eligible answers
- Failed samples: 1, excluded from the eligible denominator
Next action
Retain the observation, inspect the cited third-party source and repeat the stable benchmark later. Do not claim that another page edit will cause a citation.
Limits of the method
What an observation cannot establish.
The limits are part of the result, not a footnote added after it.
- External AI answers are variable and may change without notice.
- A sample does not represent every user, location, model, provider or future response.
- An API response does not necessarily reproduce a consumer application interface.
- Provider or retrieval changes can reduce comparability over time.
- A mention does not establish preference, endorsement or commercial impact.
- A citation does not prove that the cited page caused the answer.
- Publication does not guarantee indexing, ranking, recommendation or citation.
- Suggested questions indicate plausible discovery intent, not measured search demand.
- Returned competitors and sources describe the sampled answers, not market-wide share of voice.
- Repeat scans can show a changed observation but not prove causal uplift.
The method is designed to make a bounded observation useful, not make uncertainty disappear.
Methodology governance
Keep the method as reviewable as the result.
The current status and last material update belong with the method. If classification rules, supported providers or denominator treatment change, AdCyte retains a concise change note rather than silently rewriting the method behind historical observations.
Current record
Current methodology · Version and update record pending publication
No decorative version history is shown. A version identifier will appear only when a maintained methodology version is configured.
Methodology questions
Questions about the measurement.
What does AdCyte mean by AI visibility?
AI visibility describes what AdCyte observed when an approved question was submitted to a supported external answer provider under recorded conditions. It is not a universal measure of how visible the target is to every AI system or user.
Why does AdCyte use approved questions?
Stable wording makes the benchmark reviewable and helps later observations remain comparable. Users can revise the question, but the revised wording becomes a new benchmark version.
What is an eligible sample?
An eligible sample is a returned answer that can be evaluated against the defined target and observation rules. Failed or unsupported responses are recorded separately.
Does a mention count as a recommendation?
No. A mention records target presence. A recommendation requires language that actively or conditionally suggests the target.
Does an owned-domain citation mean the tracked page was cited?
No. AdCyte distinguishes any owned-domain link from a citation to the exact page registered for the benchmark.
Why does AdCyte not provide one GEO score?
A single score would combine different events and hide their denominators. AdCyte reports mentions, recommendations, citation types, failures and question coverage separately.
Can repeat scans prove that an optimisation worked?
No. They can show that comparable sampled answers changed after an intervention. They cannot establish that the intervention caused the change.
How many samples are enough?
The answer depends on the purpose, provider support and authorised usage. AdCyte shows the planned, attempted and eligible sample counts so the reader can judge the scope of the observation.
See the method applied to your own video.
Start with one important advert, define the questions that matter and keep the evidence behind every result.