Updated on
There is no universal GEO ranking that represents all users and systems. A useful measurement therefore works with a fixed prompt library, documented test conditions and trends over multiple measurement moments. The outcome is not a separate visibility percentage, but a report that shows which questions you appear for, which sources are used and which pages or facts need improvement.
Baseline measurement or periodic follow-up | tailored to the market, questions and channels
What does AI visibility mean?
AI visibility describes whether and how an organization, person, service, product, or proprietary resource appears in responses from generative search and response engines.
Visibility can take different forms. A brand can only be mentioned, recommended as a possible supplier, described with a specific message or cited as a source via its own URL. These forms should not be combined into one unclear metric. The GEO main page explains how this visibility is built.
What types of visibility do you measure?
A reliable report tracks eight visibility types separately, so a mention is not automatically counted as a recommendation or self-citation.
Branding
The brand or person is mentioned, without any necessary recommendation or source link.
Recommendation
The brand is presented as a possible choice for the question or situation.
Source citation
A URL is visibly used as a source.
Quotation from own source
The quoted source comes from the company's own website.
External source reference
An external publication, directory, or profile supports the listing.
Message and positioning
The content and context with which the brand is described.
Factual correctness
The extent to which key facts are presented correctly and up to date.
AI referral
A visit that comes to the website recognizably from an AI tool.
Why is there no fixed AI ranking?
AI answers are curated per question and session and may vary by system, model, language, location, account context, current events, and available web resources.
A position in one answer is therefore not comparable to a fixed organic ranking. You can measure how often a brand appears, how prominent the answer is, which sources return and whether the outcome improves over several comparable tests.
What questions belong in a prompt library?
A good prompt library reflects real questions from the customer journey and combines branded and non-branded questions.
| Prompt category | Example |
|---|---|
| Definition | What is a service or concept? |
| Problem and solution | How do I solve a problem? |
| Comparison | What is the difference between solution A and B? |
| Recommendation | Which companies, tools or specialists suit a situation? |
| Local choice | Who helps with a service in a certain region? |
| Costs and conditions | What does the service cost and what factors determine the price? |
| Risk and control | What should I pay attention to and how do I recognize quality? |
| Brand question | What does the brand do and what does it specialize in? |
| Personal question | Who is the person and what expertise does he or she have? |
The library only contains questions for which the organization has or wants to develop relevant expertise, an appropriate source page and verifiable evidence.
How do you select representative prompts?
Prompts are selected based on commercial value, target group, customer phase, expertise and the likelihood that the system will provide a composite answer.
| Selection factor | Control |
|---|---|
| Business value | How important is the question for a decision, application or brand position? |
| Target group fit | Would a real prospect or customer ask the question this way? |
| Entityfit | Which organization, person, service, location or product does the answer correspond to? |
| Source Availability | Is there a strong page or source of evidence for the question? |
| Competitive relevance | What alternatives should be followed in the same analysis? |
| Repeatability | Can the question be tested periodically in the same way? |
What does a reliable testing protocol look like?
A test protocol records the same conditions and registration fields for every measurement moment.
- Step 1Determine scopeGoal, target group, market, language and demand clusters capture.
- Step 2Choose systemsAI systems and any web or search mode select.
- Step 3Capture Prompt SetExact Prompts, Branded Variants, and Competitors register.
- Step 4Register contextDate, account context, location, language, and model or mode information save.
- Step 5Testing equivalentExecute prompts in the same agreed upon order.
- Step 6Save answersMentions, recommendations, citations, and competitors save.
- Step 7Assess qualityFactual correctness and message with fixed criteria rate.
- Step 8Classify sourcesCustom and external source URLs by type organize.
- Step 9Summary resultsReport by query cluster and channel.
- Step 10RemeasureRe-execute the same measurement set after implementations.
Do you need to test a prompt multiple times?
Repeated testing can help reveal variation, especially on important recommendation questions and when answers vary widely.
The measurement method determines in advance whether each prompt is tested once or multiple times. Do not mix single and repeat tests in the same trend without clearly labeling it. More repetitions increase reliability, but also cost and analysis time.
Which KPIs are useful?
The best KPI set combines attendance, resource usage, response quality and commercial signals.
| KPI | Definition |
|---|---|
| Mention rate | Percentage of tested prompts in which the brand or person appears. |
| Recommendation rate | Percentage of prompts recommending the brand as an appropriate choice. |
| Citation rate | Percentage of prompts that use at least one custom URL as a visible source. |
| Prompt coverage | Number of priority questions with a relevant mention or citation. |
| Answer share | Share of mentions of the brand versus a fixed competitor set. |
| Factual correctness | Percentage of audited key facts that are presented correctly. |
| Message agreement | Extent to which the answer matches the desired and substantiated positioning. |
| Source quality | Relevance, topicality and reliability of the sources used. |
| AI referrals | Recognizable visits from AI tools to the website. |
| Commercial signals | Brand searches, assist conversions, appointments or leads where reliably measurable. |
How do you calculate mention rate and citation rate?
Mention rate and citation rate are calculated on the same fixed prompt set and may only be compared over comparable measurement moments.
| Metric | Calculation |
|---|---|
| Mention rate | Branded prompts divided by total valid prompts. |
| Recommendation rate | Prompts with explicit recommendation divided by relevant recommendation prompts. |
| Citation rate | Prompts with at least one custom source citation divided by total number of valid prompts. |
| Prompt coverage | Query clusters with at least one relevant appearance shared by priority clusters. |
Always report the absolute number of prompts as well. A percentage based on five questions has less evidential value than the same percentage on a larger, balanced set.
How do you measure factual correctness and message?
Factual correctness assesses whether core information is correct; message assesses whether the answer places the organization in the correct, demonstrable context.
| Classification | Meaning |
|---|---|
| Correct | The claim corresponds to a current and verifiable source. |
| Partially correct | The core is correct, but context, limitation or topicality is missing. |
| Incorrect | The claim deviates from current source information. |
| Not controllable | There is no clear source or the claim is too vague. |
Use pre-determined key facts, such as services, location, founder, pricing model, target group or certification. Do not change the fact sheet during the measurement without registering a new version and date.
How do you assess source citations?
A source citation is only valuable if the source is relevant, current and substantively appropriate to the answer.
| Source type | Interpretation |
|---|---|
| Own primary source | The best custom page for the question or entity. |
| Own supporting source | Case, methodology, profile, FAQ or in-depth explanation. |
| Desired external source | Reliable publication, directory or profile that correctly supports the claim. |
| Weak or outdated source | Source with limited relevance, outdated information or little verifiability. |
| Competitor source | Source that supports an alternative and provides insight into the response landscape. |
How do you measure visibility per AI system?
Measure and report each AI system separately, because source display, search mode, personalization and timeliness differ.
| Channel | Reporting focus |
|---|---|
| ChatGPT | Merkvermeldingen, aanbevelingen en zichtbare webbronnen waar de modus die toont. |
| Perplexity | Source citations, source prominence and comparison between cited websites. |
| Google Generative Functions | Signals within Google, separate from classic web results and according to available reporting. |
| Gemini or Copilot | Citations, response quality, and sources when the testing environment is reproducible. |
What does AI referral traffic say?
AI referral traffic shows that a user has clicked through to the website from a recognizable AI tool; it doesn't show how many answers mentioned the brand without a click.
- Controleer herkenbare verwijzende domeinen en campaign tagging waar beschikbaar.
- Create a custom channel group or report view for AI referrals.
- Track landing page, engagement and conversion, not just sessions.
- Take into account traffic that is recorded as direct, unknown or via an intermediate page.
- Use referral traffic as an additional metric, not as a complete measure.
For analytics configuration and interpretation, the measurement plan refers to Google Analytics 4.
How do you connect AI measurement with SEO data?
AI visibility and SEO data are compared to see which topics and source pages have traction, but the results remain separate areas of measurement.
| Data source | Use |
|---|---|
| Search Console | Impressions, clicks, queries, positions and performing pages in Google Search. |
| Analytics | Landing pages, engagement, conversions and recognizable AI referrals. |
| AI measurement | Mentions, recommendations, citations, correctness and competitive share. |
| Content and action log | Which pages, facts, links or diagram elements have been adjusted and when. |
Search Console reports performance and available signals within Google. Mentions and citations in ChatGPT, Perplexity, and other third-party systems must be tested separately. An improvement may coincide with SEO growth, brand activity or external publications, making causal attribution only appropriate with sufficient evidence.
How often should you measure AI visibility?
For many organizations, a monthly measurement is sufficient to monitor trends without reacting to daily variation.
| Frequency | When appropriate |
|---|---|
| Monthly | Structural reporting and comparison with SEO and content work. |
| After major implementation | Additional measurement after adjustment of pages, entities or profiles. |
| Quarterly | Stable markets with limited content and competitive changes. |
| More often | Only when rapid changes make the additional data decidable. |
Don't measure so often that teams start aiming for random answer fluctuations. The rhythm matches the speed at which meaningful optimizations can be carried out.
How large should the prompt set be?
The size depends on the number of demand clusters, markets, languages, services and competitors; there is no universal ideal number.
A small starter set can be useful to test the method, but a management KPI requires sufficient coverage per important question type. Therefore, report how many prompts per cluster have been tested and which categories are missing.
Can you measure AI visibility manually?
Yes. A manual measurement set is suitable for a limited baseline measurement, refining prompts and a substantive assessment of answers and sources.
Manual measurement becomes time-consuming for many prompts, systems, markets and measurement moments. Automated tooling can support testing, screenshots and classification, but human control remains necessary for relevance, factual correctness and nuance.
What requirements do you place on monitoring tools?
A monitoring tool must make transparent what is actually being tested and retain sufficient raw data for auditing.
| Tool requirement | Why important |
|---|---|
| Prompt management | Fixed prompts, categories, languages, markets and versions. |
| Channel transparency | Clear system, model or mode and test date. |
| Rough answers | Full answer text, citations, URLs and screenshots where possible. |
| Competition | Fixed set of brands and clear classification rules. |
| Export | Data exportable for own analysis and archiving. |
| Repeatability | Similar tests over multiple periods. |
| Privacy | Clear handling of entered customer and company information. |
| Human review | Automatic labels and errors can be corrected. |
Do not present a tool as an objective market standard when test method, model choice or classification are insufficiently transparent.
What does an AI visibility report look like?
A good report links trends and answers to concrete source pages, findings and actions.
| Report component | Content |
|---|---|
| Executive summary | Key changes, risks, opportunities and recommended priority. |
| Measurement context | Period, systems, prompts, markets, languages, settings and methodology. |
| KPI dashboard | Mentions, recommendations, citations, coverage, accuracy and competitive share. |
| Demand cluster analysis | Results by definition, comparison, recommendation, local and commercial cluster. |
| Source analysis | Proprietary, external, legacy and competitor sources. |
| Error and message log | Incorrect facts, missing context and undesirable positioning. |
| Page actions | Source page, proposed change, impact, owner and status. |
| Trend comparison | Difference with previous equivalent measurement moments. |
| Next measurement set | Committed prompts and changes for the next period. |
An operational dashboard can be built via Looker Studio, but the raw answers and methodology remain available for review.
How do you interpret an improvement?
An improvement is more credible when multiple coherent indicators move in the same direction over multiple measurement moments.
- More relevant non-branded mentions.
- More own primary source citations.
- Higher prompt coverage in priority clusters.
- Better factual accuracy and less outdated information.
- Stronger answer share compared to regular competitors.
- More recognizable AI referrals to relevant landing pages.
- Supporting growth in brand searches, SEO traction, or qualified calls.
One extra citation or one striking answer is not a structural result. Also note negative changes and lost entries.
Which pitfalls disrupt the measurement?
Measurements become unreliable when prompt sets, channels, ratings, or test conditions change without version control.
- Drawing conclusions from one prompt or one session.
- Merge branded and non-branded questions without distinction.
- Hide all AI systems in one score.
- Only measure presence and ignore message or correctness.
- Counting a source citation without assessing source quality.
- Change prompt set or competitors in the meantime without version control.
- Present Search Console data as evidence for ChatGPT or Perplexity visibility.
- Mistaking AI referrals for all no-click brand mentions.
- Perform optimizations without action log and implementation date.
- Attribute an increase entirely to GEO without mentioning other influences.
What do you receive with an AI visibility measurement from Jakency?
Jakency can perform a one-off baseline measurement or periodic reporting, depending on the existing GEO basis and desired decisions.
| Deliverable | What you receive |
|---|---|
| Scope and prompt architecture | Target groups, demand clusters, systems, markets and competitors. |
| Test protocol | Fixed conditions, classification rules and storage of answers. |
| Zero measurement or remeasurement | Complete results by prompt, system and category. |
| KPI dashboard | Mentions, recommendations, citations, coverage, correctness and share. |
| Source and competitor analysis | Which websites and brands support the answers. |
| To-do list | Concrete page, entity, evidence and technical improvements. |
| Report discussion | Interpretation, limitations and recommended next priority. |
| Measurement plan | Fixed set and frequency for subsequent comparisons. |
A GEO audit is the right first step when no baseline or source analysis exists yet. A GEO strategyplaces measurement results in a broader roadmap. The GEO training teaches an internal team to apply the method themselves.
What is not automatically included?
A measurement process records and interprets visibility; the scope states which optimization and implementation are required separately.
- Full GEO audit of all website, entity and technical factors.
- GEO strategy for all services, markets and teams.
- Rewriting and publishing all featured pages.
- Implementation of structured data, technical tickets and external profiles.
- Paid monitoring tools, API consumption or external data charges.
- Conversion, revenue or causal attribution without reliable data sources.
- Guarantee on mentions, recommendations, citations, rankings or leads.
How much does AI visibility measurement cost?
The price depends on the number of prompts, systems, languages, markets, competitors, repetitions and reporting moments.
| Price factor | Why it counts |
|---|---|
| Number of prompts | More question clusters and variants increase testing and analysis time. |
| Number of systems | Each channel requires separate testing and interpretation. |
| Replays | Multiple runs per prompt make variation visible but increase the scope. |
| Languages and markets | Localization requires separate prompt sets and assessment. |
| Competition set | More brands increase classification and analysis. |
| Reporting frequency | One-off baseline measurement, monthly or quarterly. |
| Tooling | Manual measurement, automation, exports and possible licenses. |
| Substantive review | Fact checking and message analysis require professional assessment. |
A well-defined proposal follows after a short intake. We do not publish a fixed price when prompt set, systems and measurement frequency have not yet been determined.
Which case shows separate Google and AI measurement?
The Ask Gustavo case shows why classic Google Search performance, Google generative signals, and external AI mentions are reported as separate layers of evidence.
A Search Console result proves visibility in Google Search, but does not automatically indicate that ChatGPT or Perplexity mentions the brand or cites a proprietary source. Those systems require a separate prompt set and source registration.
When is measuring AI visibility not the right first step?
Measuring is of little use if there is no stable information base or implementation capacity yet.
- The organization does not yet have clear services, entities or primary source pages.
- Important pages are not crawlable or indexable.
- There is no ownership for facts, content or technical improvements.
- The goal is just one positive screenshot with no repeatable method.
- There is no capacity to respond to the findings.
- A broad GEO audit is needed first to understand source and technical issues.