How to Build an AI Search Visibility Dashboard by Topic

TL;DR

Share of AI Voice (SAIV) is a family of measures for a defined set of AI answers. Report brand mentions, linked citations and recommendations separately, with a clear denominator for each.
One answer is only a snapshot. Outputs can vary across runs, engines, location, settings and time.
Group related questions by intent. A cluster helps organize comparison, but no fixed number of prompts guarantees a representative or stable result.
Mentions ≠ citations ≠ recommendations. These are different observations; none alone proves a visitor, enquiry or sale.
A useful GEO dashboard has 6 views: cluster-level SAIV summary, citation source map, competitor dominance, prompt volatility, intent-weighted visibility, and outcome-proxy layer.
GEO is not separate from SEO. Query clusters map to topic clusters. Cited URLs reveal which content formats win. The GEO dashboard should drive your editorial roadmap.
Operationalize monthly. Assign cluster owners, run monthly GEO reviews, and connect SAIV insights to content prioritization.

Definitions: The Language of GEO Measurement

Before building anything, agree on terms. Ambiguous definitions produce useless dashboards.

Term	Definition
Share of AI Voice (SAIV)	A family of visibility measures for a defined set of AI answers; suppliers may define the term differently. State the numerator and denominator.
Query Cluster	A versioned group of prompts expressing related intent, audience and market.
Mention	Your brand name appears in an answer.
Citation	Your domain or a specific page is linked as a source. A citation is not proof of factual accuracy or model trust.
Recommendation	Your brand is presented as an option in the answer. It does not demonstrate a buyer's preference.
GEO	Generative Engine Optimization. Optimizing content for visibility within AI-generated answers.
AEO	Answer Engine Optimization. Often used interchangeably with GEO; focuses on direct-answer interfaces.
Prompt Volatility	The degree to which AI answers vary for the same prompt across runs. High volatility = unreliable single-prompt measurement.
Cluster-Level Visibility	Separate visibility measures across eligible answers to prompts within a cluster; retain prompt-level results and counts.
Citation Answer Rate	Eligible answers containing a link to your domain divided by all eligible answers. Distinct from the share of all cited source links pointing to your domain.
Recommendation Answer Rate	Eligible answers recommending your brand divided by all eligible answers.

Why Most GEO Measurement Is Wrong

Most teams attempting GEO measurement make the same mistakes. The result is noise dressed up as insight.

Common Failure Modes

Tracking a handful of random prompts. A small convenience sample can be a useful diagnostic, but cannot establish visibility across all buyer questions or regions.
Assuming prompt results are stable. The same query may return different citations, mentions or recommendations over time. Record settings and repeat observations rather than treating one answer as a trend.
Mixing low-intent and high-intent prompts. "What is marketing attribution?" and "best marketing attribution tool for SaaS" have completely different commercial value. Averaging them together destroys insight.
Treating all appearances equally. A casual mention ("some companies like Acme...") is different from a citation ("according to Acme's research...") or a recommendation ("top options include Acme"). Classify and report them separately.
Counting brand mentions as commercial wins. Mentions and recommendations are observations in model answers, not evidence that a prospect acted on them.
Ignoring competitor share. Your absolute SAIV number means nothing without knowing who else appears—and how often. A 40% mention share sounds good until you realize one competitor has 60%.
Unstated business priorities. Not all topics matter equally to the team. Document any optional priority weights and the assumptions behind them.

These approaches create dashboards that feel productive but support zero decisions. They are measurement theater.

Why the Query Cluster Is the Right Unit of Analysis

A single prompt is the wrong unit of GEO measurement for the same reason a single keyword is the wrong unit of SEO strategy. It's too granular, too noisy, and too disconnected from how people actually research.

1. A Single Prompt Is Unstable

The same prompt can produce varying output across runs, models, and contexts. Preserve individual observations and their settings so changes can be interpreted rather than assumed.

2. A Cluster Gives Results Context

Related prompts make patterns easier to inspect, but aggregating them does not make volatility disappear. Repeat the same versioned prompts, retain individual results, and report the number of eligible answers; a small or convenient set cannot represent all real searches.

3. Clusters Reflect Real User Intent

Users don't ask one question. They explore a topic through multiple related queries. A cluster mirrors this research behavior and tells you whether you're visible across an intent space—not just for one phrasing.

4. Clusters Enable Pattern Detection

At the cluster level, you can spot structural patterns: "We're always mentioned but never cited in the attribution cluster" or "Competitor X dominates the CRM selection cluster." Prompt-level data cannot reveal these patterns.

5. Clusters Align with Content Strategy

Query clusters map directly to topic clusters and content pillars. This means your GEO dashboard can directly inform which content themes need investment—making it a planning tool, not just a scorecard.

What a Query Cluster Looks Like in Practice

A cluster is a grouped set of prompts expressing similar intent. Here are examples:

Attribution & Measurement: "How to measure marketing attribution in 2026," "best attribution models for B2B," "incrementality testing vs attribution," "marketing attribution without cookies"
CRM Selection: "best CRM for SaaS startups," "HubSpot vs Salesforce for mid-market," "CRM evaluation criteria," "CRM implementation checklist"
GEO/AEO Measurement: "how to measure GEO visibility," "share of AI voice definition," "AI search visibility metrics," "GEO vs SEO measurement"
Lifecycle Orchestration: "lifecycle marketing best practices," "email frequency capping strategy," "cross-channel orchestration architecture," "lifecycle control tower"
Vendor Evaluation: "best marketing analytics tools," "CDP comparison 2026," "top conversion optimization tools," "marketing attribution platforms ranked"

With cluster-level measurement, teams can answer: Where are we visible? Where do competitors dominate? Which content themes are underperforming? Where are the commercial visibility gaps?

The 5-Layer Share of AI Voice Model

A rigorous GEO measurement framework needs five dimensions. Miss any one, and your dashboard produces incomplete or misleading signals.

Layer 1: Query Cluster

The thematic grouping. As a practical starting example, choose a handful of prompts with related intent across different phrasings, angles, and specificity levels; size depends on the questions and resources available.

What teams miss: Clusters that are too broad (mixing informational and evaluative prompts) or too narrow (splitting what should be one cluster into three).

Layer 2: Prompt Set & Phrasing Coverage

The individual prompts within each cluster. Include informational ("what is X"), comparative ("X vs Y"), evaluative ("best X for Y"), and action-oriented ("how to implement X") phrasings.

What teams miss: Over-indexing on informational queries and under-indexing on comparative/evaluative queries where commercial value is highest.

Layer 3: Appearance Type (Mention / Citation / Recommendation)

For each prompt result, classify the appearance: Was the brand mentioned casually? Was the domain cited as a source? Was the brand recommended as a preferred option?

What teams miss: Lumping all three into "we showed up." A mention is not a citation. A citation is not a recommendation.

Layer 4: Competitor Capture

For each cluster, map which competitors appear, how often, and in what capacity. Your SAIV is only meaningful relative to the competitive landscape.

What teams miss: Tracking only their own visibility without knowing who else dominates each cluster.

Layer 5: Intent Weight & Business Value

Not all clusters matter equally to a team. Optional weights can express a stated planning priority, but are assumptions rather than measured revenue value; retain the unweighted rates beside any composite.

What teams miss: Treating all clusters equally, which makes the dashboard decorative rather than prioritizing.

The Metrics That Matter in a GEO Dashboard

Not every metric is equally useful. Here are the metrics that actually support decisions, organized from awareness to business value:

Metric	What It Means	Why It Matters	Common Misuse
Mention Answer Rate	Eligible answers naming the brand ÷ eligible answers	Observed name visibility	Treating a mention as a business outcome
Citation Answer Rate	Eligible answers linking to the domain ÷ eligible answers	Observed linked-source visibility	Confusing answer rate with share of all cited links
Recommendation Answer Rate	Eligible answers recommending the brand ÷ eligible answers	Visibility in evaluated options	Assuming a recommendation caused a purchase
Competitor Share	Competitor visibility per cluster	Relative positioning	Tracked only for self, ignoring competitors
Unique Cited URLs	Which pages are being cited	Content format/quality signal	Not mapped back to content strategy
Cluster Volatility	How much results vary across runs	Measurement confidence	Ignored—leads to overreaction to noise
Priority Cluster Coverage	% of high-value clusters with visibility	Strategic gap analysis	All clusters treated equally
Weighted SAIV	Business-weighted composite score	Executive summary metric	Used without understanding the inputs

Visibility observations: Mentions, citations and recommendations show what appeared in the tested answers. They do not by themselves establish influence on business outcomes.

Business measures: Track referral visits and independently verified enquiries separately, with matching time windows and consent limitations. Investigate associations without assuming causation.

How to Build the GEO Dashboard

A practical GEO dashboard needs six views. Each one answers a different strategic question.

View 1: Cluster-Level SAIV Summary

What it shows: SAIV score for each cluster, broken into mention/citation/recommendation. Trended over time.

Decision it supports: "Which clusters need content investment?" and "Where are we gaining or losing ground?"

View 2: Citation Source Map

What it shows: Which specific URLs from your site are being cited, in which clusters, how often.

Decision it supports: "Which content is working?" and "Which pages should we update or promote?"

View 3: Competitor Dominance View

What it shows: Side-by-side visibility of top competitors per cluster. Who appears most? In what capacity?

Decision it supports: "Where are competitors winning?" and "Which clusters should we contest?"

View 4: Prompt Volatility View

What it shows: How much variation exists across runs for each cluster. High volatility = low confidence in trends.

Decision it supports: "Can I trust this cluster's data?" and "Where should I increase prompt coverage?"

View 5: Intent-Weighted Visibility

What it shows: SAIV weighted by commercial importance. High-value evaluative clusters weighted higher than informational ones.

Decision it supports: "Where should we prioritize content creation for maximum business impact?"

View 6: Outcome-Proxy Layer

What it shows: Correlation between SAIV trends and proxy outcomes—branded search volume, direct traffic, assisted conversions, inbound lead quality.

Decision it supports: "Is GEO visibility actually helping the business?"

How to Design Query Clusters

The quality of your GEO dashboard depends entirely on the quality of your clusters. Here is how to build them properly.

Choosing Clusters

Start with 5–7 clusters mapped to your most important content themes and commercial priorities. Each cluster should represent a distinct intent space your business needs to be visible in.

Prompts Per Cluster

Choose enough distinct prompts to cover the intended questions and repeat them consistently. Eight to fifteen can be a manageable starting point, not a statistical threshold; inspect variation and expand coverage where the decision warrants it.

Prompt Diversity Rules

Each cluster should include prompts across four intent types:

Informational: "What is marketing attribution?"
Comparative: "Marketing attribution vs incrementality testing"
Evaluative: "Best marketing attribution tools for B2B SaaS"
Action-oriented: "How to implement incrementality testing"

Weighting Clusters

Assign each cluster a business weight from 1 (low) to 5 (high) based on:

Revenue relevance of the topic
Search volume of related queries
Strategic importance to positioning
Competitive intensity in the space

Sample Cluster Map

Cluster	Prompts	Business Weight	Primary Intent
Attribution & Measurement	12	5 (critical)	Evaluative + Action
GEO / AEO Visibility	10	4 (high)	Informational + Evaluative
CDP vs CRM vs Warehouse	12	5 (critical)	Comparative
Lifecycle Orchestration	10	3 (medium)	Action + Evaluative
Conversion & Funnel Design	8	4 (high)	Action
Vendor Selection / Best Tools	15	5 (critical)	Evaluative

Why Mentions, Citations, and Recommendations Must Be Separated

This is the most common source of false confidence in GEO measurement. Teams celebrate "showing up" without understanding how they showed up.

Mention = Awareness Signal

The AI names your brand in passing. "Some companies in this space include Acme, BrandX, and Others." An answer alone does not reveal whether the name came from training, retrieved material or another source, and does not establish trust or preference.

Citation = Linked Source

The AI links to your domain or references your content as a source. "According to Acme's research on attribution modeling..." Record the linked URL and inspect the claim; a link alone does not prove the answer is correct or reveal the model's internal assessment.

Recommendation = Listed Option

The AI lists your brand as a recommended option. "For mid-market SaaS companies, strong options include Acme (best for data-driven teams)..." Record the wording and context; a recommendation is not proof that someone saw it, clicked or became a customer.

Optional Prioritization Example

Mention: 1 point
Citation: 3 points
Recommendation: 5 points

These illustrative weights are a team's assumptions, not evidence that a citation is worth three mentions or that a score predicts revenue. Keep the separate raw measures and sample counts visible.

How GEO Connects to SEO

GEO is not separate from SEO. It is an extension of it—and a feedback loop.

Search eligibility supports discovery. Google says its normal Search technical requirements apply to AI features. A ranking does not establish inclusion in any model's training data or guarantee a citation.
Make useful answers accessible. Clear definitions, relevant evidence and readable page structure help people understand the material; no format guarantees that an AI system will select it.
Query clusters map to topic clusters. If you already run topic-cluster SEO, your GEO clusters can mirror those themes. One measurement system, two channels of visibility.
Cited URLs reveal content format winners. When you see which pages are cited, you learn what content formats the AI prefers. Comparison pages? Definition articles? Framework posts? This should drive your editorial roadmap.

The GEO dashboard should inform your content strategy: which topics to update, what comparison pages to create, which glossary entries to write, and which implementation guides to publish. Read Google's guidance for AI features and OpenAI's crawler documentation; crawler access, retrieval and public citation are separate outcomes. For help designing the measurement and implementation, see forward-deployed marketing.

Practical Examples

These are synthetic illustrations, not client results or evidence that an editorial change caused a visibility increase.

Example 1: Random Prompt Tracking (What Fails)

A B2B SaaS company picks 20 prompts across different topics, checks them once a month, and records "appeared" or "did not appear." Results: 8 out of 20 appearances. They report "40% SAIV" to leadership.

Why this fails: The 20 prompts span six unrelated topics. Some are informational, some commercial. They mix mentions with citations. They didn't check competitors. They didn't run prompts multiple times. Next month, different prompts hit, and "SAIV" swings to 55%. Leadership thinks GEO is working when nothing changed.

Example 2: Cluster-Based Measurement (What Works)

The same company reorganizes into 5 clusters with 10 prompts each. They run each prompt 3 times. They separate mentions, citations, and recommendations. They track competitors.

Findings:

Attribution cluster: 18 of 30 eligible answers mention the brand (60%), 6 link to it (20%), and 2 recommend it (6.7%). The measures can overlap and do not establish why the difference exists.
CRM selection cluster: 6 of 30 answers mention the brand (20%) and 12 link to its site (40%). Inspect those source URLs and their relevance.
Vendor evaluation cluster: 21 of 30 answers recommend Competitor X (70%). Review whether the prompts and competitors reflect the actual buying context.

The team can prioritize hypotheses: review attribution content, inspect the CRM sources, and assess whether a useful comparison page is warranted. The sample cannot establish the cause or predict impact.

Example 3: Dashboard-Driven Content Prioritization

A content team reviews the monthly GEO dashboard and sees that the "CDP vs CRM vs Warehouse" cluster has low citation share (10%) despite being weighted as a critical cluster.

Action: They revise a comparison article with sourced evaluation criteria and implementation tradeoffs. In this hypothetical example, 2 of 20 eligible answers initially cite the site (10%), and 7 of 20 do so later (35%); that timing alone does not show that the edit caused the change, or that it generated business.

Operational Playbook: Running GEO as a Real Operating System

A dashboard without a cadence is a decoration. Here is how to operationalize cluster-based GEO measurement.

Monthly GEO Review

30–45 minutes. Review cluster-level SAIV trends, competitor shifts, and citation source changes. Output: 2–3 content actions for the next sprint.

Cluster Ownership

Assign each cluster to a content owner or topic lead. They are responsible for prompt maintenance, trend interpretation, and content recommendations for their cluster.

Prioritization Framework

Focus on clusters with: (1) high business weight + low SAIV, (2) declining citation share, (3) competitor dominance that threatens positioning. Ignore high-SAIV, low-weight clusters.

Link to Editorial Roadmap

Every monthly review should output specific content recommendations: update this article, create this comparison page, add FAQ schema to this post, publish this new framework. GEO insights should feed directly into your content calendar.

Connect to SEO Reporting

Run GEO and SEO reviews together. Compare visibility observations with organic traffic, branded search and enquiries over comparable periods, recording consent and reporting gaps. Correlation alone cannot establish what caused a change.

Avoid Dashboard Bloat

Start with 5–7 clusters and 6 dashboard views. Add complexity only when you've outgrown the current setup. The best GEO dashboard is the one your team actually reviews monthly.

Common Mistakes to Avoid

Ignoring the cluster context. Inspect individual prompt results as well as related groups; neither view is automatically representative.
Counting appearances as wins. Mentions and recommendations do not show whether a buyer acted.
Unstated priorities. Document why a cluster matters to the team and any assumptions used to prioritize it.
No competitor view. Your SAIV means nothing without relative positioning.
No source page mapping. If you don't know which pages are cited, you can't improve them.
No content action loop. A dashboard that doesn't produce editorial actions is a vanity project.
Overfitting to unstable results. Small changes in either a prompt or a cluster can reflect variation rather than a durable trend.
Building a vanity dashboard. If no one makes a different decision because of your dashboard, it's not a dashboard—it's a screensaver.

Minimum Viable GEO Dashboard Checklist

☐ 5–7 query clusters defined and documented
☐ Versioned prompts with relevant intent diversity and documented market and engine settings
☐ Business weights assigned to each cluster (1–5)
☐ Mentions, citations, and recommendations tracked separately
☐ Competitor visibility captured per cluster
☐ Repeated observations and eligible-answer counts recorded per measurement period
☐ Cluster-level SAIV summary view built
☐ Citation source map showing which URLs are cited
☐ Competitor dominance view per cluster
☐ Monthly review cadence scheduled
☐ Cluster owners assigned
☐ Content action loop connected to editorial roadmap
☐ Outcome proxies (branded search, direct traffic) tracked alongside SAIV

Frequently Asked Questions

What is Share of AI Voice?

Share of AI Voice (SAIV) is a family of measures for a defined set of AI answers. Report mentions, citations and recommendations separately, with each numerator and denominator, and note that platforms may define their metrics differently.

How is Share of AI Voice different from SEO visibility?

SEO visibility measures ranking positions and click-through rates in traditional search results. SAIV measures inclusion in AI-generated answers—whether your brand is mentioned, cited, or recommended. You can have zero SEO visibility and high SAIV, or vice versa.

Why should I measure by query cluster instead of individual prompts?

Clusters group related questions for useful comparison; individual results remain important. Outputs can vary across runs, so document settings, report counts, and avoid assuming a cluster size guarantees stability.

What is the difference between a mention and a citation?

A mention means the answer names your brand. A citation means it links to or identifies a source. Their value depends on the context; neither an assumed multiplier nor a citation proves trust or business impact.

What is recommendation share?

Recommendation answer rate is the number of eligible responses recommending your brand divided by all eligible responses in the defined set. It does not establish buyer preference or purchase influence.

How many prompts should be in a query cluster?

There is no universal number. Start with a manageable, documented set that covers relevant informational, comparative, evaluative and action-oriented phrasings; repeat consistently and report the counts and limitations.

How often should I measure GEO visibility?

Monthly can be a practical review cadence. Repeat the same versioned prompts as resources permit, preserve the per-run results and avoid treating a small sample as a complete market estimate.

Can small sites compete on Share of AI Voice?

They can appear, but selection varies by system and query. Useful, specific, accessible material gives people and crawlers something to find; it does not guarantee a citation or recommendation.

How do I choose the right query clusters?

Start with topics that matter to the buyers you serve. Map clusters to your content pillars and document any weights as planning assumptions, rather than measured buyer influence.

How do I know if GEO visibility is helping the business?

Track referral visits and genuine enquiries alongside visibility observations, and compare equivalent periods. Branded search or direct traffic may provide context, but changes do not prove that an AI answer caused them.

What tools do I need to build a GEO dashboard?

At minimum: a systematic way to run prompts across AI engines, a spreadsheet or database to classify results (mention/citation/recommendation), and a reporting layer. Dedicated tools like Otterly, Profound, or custom scripts can automate prompt execution.

Should I track multiple AI engines or just one?

Track at least 2–3 engines (e.g., ChatGPT, Perplexity, Google AI Overviews). Each engine uses different retrieval logic and training data. A single-engine view gives you an incomplete picture of AI visibility.

Conclusion

GEO measurement becomes more useful when you combine individual observations with groups of related questions and document the limits of the sample.

A useful GEO dashboard is a prioritization system. It shows what appeared in a defined sample, separates mentions, citations and recommendations, and keeps traffic and enquiries as independent measures. It can guide content hypotheses, but cannot promise their impact.

Most teams will continue to screenshot ChatGPT responses and call it measurement. The teams that build cluster-based dashboards will have something their competitors don't: a systematic, defensible, decision-grade understanding of their AI visibility.

Use both prompt-level evidence and cluster context to support decisions.