Citation review : Time mismatch
D01 describes a planned, conditional pilot. Current availability is not established.
GUIDE 1701
SEO, GEO & AI visibility measurement
Measure what search engines show, what assistants answer, which sources they use and whether the entity remains accurately represented.
Operational briefPROMPTSstable panel
CITATIONSprovenance
ENTITYaccuracy
Operational brief
A screenshot is not an observatory.
Appearing in an AI answer, receiving a citation and receiving a citation that supports the claim are three different observations. He retains questions, test conditions, answers and cited passages. The analysis guides a documented correction; it produces neither a universal ranking nor a traffic promise.
AR / GUIDE 1701INDEPENDENT ANALYSIS
DOCUMENTED CASE
A correct link can still accompany a misleading answer
Three fictional answers use the same case. The check examines the relationship between a claim and its cited passage, rather than just the presence of a brand or link.
Entirely fictional teaching example: documents, answers and logs are invented. No client result or real platform test.
The claim under review
The service is already open in three countries.
Isolate
Record the exact claim made in the answer.
Open
Find the cited record and passage.
Compare
Check time, scope and conditions.
Qualify
Document the mismatch or agreement without generalising.
Open the review of each answer
Citation review : Scope mismatch
D04 records the organisation’s registration, not the countries in which its service can be used.
R03 · Fictional answer
A pilot is announced; access remains limited to invitees and one enabled region.
Citation review : Compatible wording
D01, D05 and D06 support this qualified wording within the fictional case.
Open the case documents
| Record | Declared origin | Fictional excerpt | Derivation |
|---|---|---|---|
| D01Launch announcement | Organisation documentation | A pilot is planned in three countries, subject to completion of testing. | Direct document |
| D02Editorial repost | Professional blog | The new service is open in three countries. | Copy of D01 |
| D03Aggregated repost | Aggregator | The service is available in three countries. | Copy of D02 |
| D04Register entry | Public register | The organisation is registered and its status is active. | Direct document |
| D05Pilot documentation | Organisation documentation | Access is restricted to invited participants. | Direct document |
| D06Service status | Organisation documentation | One pilot region is enabled; the others are still being prepared. | Direct document |
Read the reasoning
R01 cites the right record but turns a conditional plan into a present fact. The link exists; the meaning has changed.
R02 cites a genuine record within the scenario but gives it a scope it does not have. Administrative status does not establish commercial availability.
R03 retains the limits of the records. This supports a conclusion about this particular answer, not an assistant’s overall reliability or a change in traffic.
Method references : Google: AI features and your website · Content expanded on .
MEASUREMENT WORKSHOP
Being cited and being cited accurately are separate measures
Count answers citing the target domain separately from references whose contents support the claim. One answer can contain several links, so the two rates have different denominators.
AI-generated illustration
Fictional example · no measurement of this site or a client
| Indicator | Calculation or threshold | What the result establishes |
|---|---|---|
| Presence in answers | 12 / 40 × 100 = 30 % | 12 of 40 answers cite the domain: 30% presence in this sample. |
| Support among checked references | 15 / 20 × 100 = 75 % | 15 of 20 references support the claim: 75% of checked references. This says nothing about unchecked references. |
From observation to decision
Fix the panel
Keep questions, language, service, mode, dates and repetitions. Include answers with no citation.
Open each reference
Match the exact passage to the claim; record inaccessible links and citations that do not support it.
Compare a constant scope
A change requires the same panel and protocol. Publish counts and exclusions too.
An overall rise can hide two declines
Both panels contain 110 questions, but their mix of tasks A and B is reversed. The positive criterion is unchanged; each row retains its denominator.
| Observed unit | Fictional data |
|---|---|
| Task A: before → after | 9 / 10 = 90 % → 80 / 100 = 80 % |
| Task B: before → after | 40 / 100 = 40 % → 3 / 10 = 30 % |
| Total: before → after | 49 / 110 ≈ 44.55 % → 83 / 110 ≈ 75.45 % |
Does the overall rise establish improvement?
No: both A and B decline while the panel gives A more weight. Composition explains the reversal. Rerun a stable panel or report rates and counts by task.
References and method
- Google Search Central · AI features
- Google — Generative AI performance report (Search)
- Bing — AI Performance
Documentation consulted on .
MEASUREMENT / VISIBILITY SOURCES
Impression, citation, visit: three different observations
Appearing in an AI answer and receiving a website visit happen at different stages. Choose the signal that answers your question before comparing periods.
- Be accessible
Crawling may make a page available for search.
- Be shown or cited
A link can appear in an answer without receiving a click.
- Receive a visit
The site records an arrival within the limits of its measurement setup.
| Observation source | What is measured | Interpretation limit |
|---|---|---|
| Google · Generative AI in Search Console Official documentation | Link impressions in AI Overviews and AI Mode, with dimensions including page, country and device. | Days use Pacific Time (PT). Property totals can differ from the sum of page rows because of aggregation. |
| Bing Webmaster Tools · AI Performance Official documentation | Visible citations and cited pages in supported AI experiences; grounding queries are grouped. | A citation is neither a click, a ranking nor a measure of page importance. Grouped queries do not reproduce individual prompts. |
| Website measurement | Arrivals and actions recorded under your measurement configuration. | An attributed visit does not represent every impression. Missing referrers or blocked measurement limit attribution. |
| Documented question panel | Citation presence and fidelity across a set of observed answers. | The result depends on questions, mode, language and date. It does not describe all users. |
Google AI report impressions are included in the Search Console Web scope: do not add them to the Web total. Compare visibility, citations and visits separately.
Compare the scope of two observations
Enter the metadata of two observations. This checks basic compatibility without asking for your counts or accessing your accounts.
Documentation accessed on .
02 / Intervention areas
Intervention areas
Question panel
Brand, expertise, comparison, recommendation, definition, risk and decision questions tested through controlled wording.
Answer share
Appearance frequency, position in the answer, descriptive depth and presence of competitors or alternatives.
Entity accuracy
Name, role, location, expertise, relationships and limits compared with the canonical public reference.
Citation analysis
Used sources, diversity, authority, freshness, reused passages and recoverability of evidence.
Zero-click journey
Recall, next action, qualified visit, branded search and conversion after answer consumption.
Variation log
Differences across models, dates, languages, accounts, wording and significant changes documented.
03 / FIELD NOTES
What the observatory reveals
Measurement supports editorial and technical decisions rather than creating another decorative score.
01ABSENCE
The entity does not appear
Sources may be weak, scattered or insufficiently connected to important questions.
02CONFUSION
Right name, wrong role
The public graph mixes namesakes, outdated descriptions or implicit relationships.
03SOURCE
A weak reference dominates
The easiest source to retrieve is not always the most reliable or recent.
04JOURNEY
The answer satisfies without a visit
The brand must remain identifiable, memorable and connected to a useful action even at zero click.
04 / OUTPUTS
Generative visibility board
Results connect observations, likely causes, evidence and action priorities.
Versioned corpus
Questions, variants, languages, intents, models and test dates.
Cross-channel baseline
Classical search, generative interfaces and assistants compared through one reference.
Citation map
Domains, pages, passages, frequency, freshness and source diversity.
Error register
Omissions, confusion, outdated information, contradictions and severity.
Reinforcement plan
Pages, evidence, structured data, relationships and public sources to improve.
Periodic review
Evolution, new errors, observed corrections, volatility and next decisions.
WORKING RECORDS
Make the dossier verifiable
Three practical records connecting diagnosis to a traceable decision.
Reference sample
Minimum content : Fix questions, phrasings, languages, interfaces, declared location and measurement dates.
Check : Does the sample describe a useful audience rather than easy queries?
Observation sheet
Minimum content : For each answer: interface version, mention, citation, link, accuracy and dated capture.
Check : Is an unlinked mention being mistaken for a usable citation?
Interpretation table
Minimum content : Distinguish normal variation, repeated gap, wrong source and possible editorial effect.
Check : Is the decision based on several comparable observations?
05 / Method
Compare before interpreting
Every observation is repeated, dated and placed in a series. Significant gaps are verified before triggering public action.
- 01Define entities
- 02Version questions
- 03Fix test conditions
- 04Capture answers and citations
- 05Qualify errors
- 06Compare models
- 07Connect SEO data
- 08Decide corrections
DECISION CHECKS
Decisions to document
Two checks that change the next decision.
The question panel changes too often
Keep a stable core for comparing series and place new questions in a separate set. Otherwise a protocol change may be mistaken for a visibility change.
Answers appear to improve
Check that the improvement repeats across chosen wordings, languages and times. Link it to accuracy and cited sources before calling it useful progress.
Two approaches to distinguish
Two approaches to distinguish
The right method depends on the question and the decision it must inform.
Answer snapshot
Shows what an interface displayed for one question, language and moment. Useful for a specific error, not a standalone frequency estimate.
Repeatable panel
Keeps questions, settings, periods, results, sources and coding rules stable. Interpret changes alongside content changes and sample limits.
Proposed work plan
Adapt this outline to the agreed scope. It describes a proposed method, not a completed mission, measured result or guaranteed schedule. Record your observations and decisions as checks proceed.
AI-generated illustration
Prepare before starting
Questions, interfaces, languages, dates and response classification criteria. Define expected references and conditions for an inconclusive case.
Scope the work
Freeze a panel version and reading rules. Identify interface and context, leaving unobservable settings unknown.
Apply the method
Retain complete answers and open useful references. Classify mention, link and supporting passage without removing unfavorable trials.
Verify and hand over
Report count, denominator, dates and exclusions with reasons. Compare only series with compatible questions and conditions.
Pause or revision conditions
A small panel describes retained trials, not all uses. No percentage is interpretable without its denominator and exclusion rules.
06 / FAQ
Frequently asked questions
Clear answers on scope, limits, methods and engagement conditions.
Can a ChatGPT ranking be measured?
There is no universal stable ranking comparable to an SEO position. A panel of answers is measured under documented conditions.
How many questions should be tracked?
Enough to cover decisive intents without creating an unmaintainable corpus. Stability matters more than volume.
Can an answer without links be measured?
Yes. Presence, accuracy, narrative position, recall and branded search remain observable.
Why do answers change?
Models, indexes, sources, personalisation, wording and product policies evolve. Every conclusion must be dated.
Should several languages be tested?
Yes for international activity because available sources and representations may differ greatly.
How can false information be corrected?
First strengthen canonical, coherent and accessible sources, then document evolution without promising immediate model updates.
Do citations guarantee traffic?
No. They improve traceability and authority, while users may consume answers without visiting the source.
Can monitoring be automated?
Partly, with control of cost, terms of use, reproducibility and human validation.
Which metric matters most?
The one tied to the decision: accuracy for identity, citation for expertise or useful action for an offer.
Can a citation to the right site still be misleading?
Yes. The passage may refer to another date, a narrower scope or a plan that remains conditional. Review each claim against the passage meant to support it.
Methodological references
Official documents explaining methods, review criteria and their limits. These references do not certify the service or guarantee search rankings.
08 / CONTACT
Open a private channel
State the context and objective. You will receive a clear initial reading.
DEEPER READING / PRACTICE
Define what an AI observation measures
A team wants to quantify its visibility in generated answers. Before calculating a rate, it must define the questions, platforms, languages and conditions included in the observation.
Updated on .
Priority checks
Distinguish the appearance of the name, a URL quotation, the recommendation and the accuracy of the answer. Fix a counting rule and keep the cases excluded or impossible to test.
Interpretation pitfalls
Adding observations made under different conditions produces an indicator that is difficult to interpret. Session or formulation variations may change the observed result.
Working record to retain
The measurement plan documentes targeted population, sample, repetitions and definition of categories. The report presents the gross numbers with their denominator and the comparison limits.
Choose the measurement unit and weights
Decide whether the indicator concerns distinct questions or all generated answers. A question repeated ten times should not inadvertently receive ten times the weight.
Show results by language and question type before pooling them. State missing answers and the aggregation rule: averaging rates and calculating a rate across all answers do not necessarily weight groups equally.
EXAMPLE / READING THE NUMBERS
Presence does not always mean accurate representation
Fictional data illustrating the method; not client results or measurements of this website.
| Observation | Count | Interpretation |
|---|---|---|
| Panel questions | 40 | Fixed comparison base. |
| Answers mentioning the brand | 12 | 12 / 40 = 30% of the panel. |
| Answers describing the offer correctly | 8 | 8 / 40 = 20% of the panel; 8 / 12 ≈ 66.7% of mentions. |
Show the interpretation limit
What these numbers do not prove
These observations measure neither market share nor sales. A mention may be inaccurate. Report the panel, date, language and repetitions alongside the result.