Blog · Competitive AI visibility analysis
Competitive AI Visibility Map: Who Wins by Model and Vertical
Dharini Shah · September 5, 2026
A competitive AI visibility map is a grid that shows how often each brand in your market appears, gets cited and gets recommended across AI models, broken down by the types of questions buyers ask. It answers the question most AI visibility reports skip: not "are we visible?" but "where are competitors winning, why, and where is there room for us?"
The answer is rarely uniform. The same brand can dominate comparison prompts in ChatGPT, be absent from Perplexity because it lacks community discussion, and trail a competitor in Google AI Mode for pricing questions. A single blended score hides all of that. This guide explains how to build a visibility map that reveals those differences, how to read it, and how to turn it into a plan.
Why a single AI visibility score is not enough
AI assistants draw on different sources. Profound's citation analysis found Wikipedia was ChatGPT's most cited domain, while Reddit led for Perplexity and Google AI Overviews. Ahrefs' freshness study found ChatGPT favored newer content much more than Google AI Overviews did. If the models weigh sources differently, competitors can win in some models and lose in others for reasons that have nothing to do with the quality of their product.
Verticals differ too. Categories with active communities, heavy review-site coverage or strong analyst presence give models different material to work with. A software category discussed constantly on forums will look different in Perplexity than a regulated financial product where government and institutional sources dominate. The Ahrefs 2026 benchmark also found AI Overviews appeared on 46.4% of queries with seven or more words versus 9.5% of single-word queries, which means the kinds of questions buyers ask in a vertical also change where AI answers show up.
A map preserves those differences so you can act on them.
The dimensions of a competitive visibility map
A useful map has four dimensions.
| Dimension | What it captures | Example values |
|---|---|---|
| Brand | You and the competitors buyers compare you with | 4 to 8 brands, including indirect alternatives |
| Model | The AI assistants your buyers use | ChatGPT, Gemini, Claude, Perplexity, Copilot, Google AI Overviews and AI Mode |
| Prompt cluster | The type of question | Category discovery, comparison, alternatives, use case, pricing, integration, trust and compliance |
| Metric | What you measure in each cell | Visibility rate, recommendation share, citation share, sentiment |
Verticals become a fifth dimension when you operate in several markets, or a filter when you compare your map with benchmarks from other categories.
The metrics that belong in each cell
Visibility rate: the percentage of sampled answers that mention the brand at all.
Recommendation share: among answers in a cluster, the share in which the brand is actively recommended for the buyer's need, rather than merely listed.
Citation share: the percentage of cited sources that belong to the brand's own domain or directly discuss the brand.
Sentiment and accuracy: whether the brand is described positively, neutrally or with caveats, and whether the facts are correct.
Avoid "average position." Research by SparkToro and Gumshoe found fewer than 1 in 1,000 runs produced the same ordering of recommendations. The researchers argued that visibility percentage across many runs is a reasonable metric, while "ranking position in AI" is not.
How to build your map, step by step
Step 1: Choose the competitive set
Start with the competitors your sales team hears about. Then ask AI assistants for alternatives to your brand and to your top competitor. Models often surface brands you do not consider direct competitors, and those brands are competing for the same answers.
Step 2: Build prompt clusters
Write 5 to 15 prompts per cluster that reflect how buyers actually ask. Use sales call notes, support tickets, community threads and search queries as inputs. Include constraints buyers state, such as team size, budget, industry or required integrations, because constraints change which brands are recommended.
Example clusters for a hypothetical email marketing platform:
- Category discovery: "What are the best email marketing tools for ecommerce brands?"
- Comparison: "Brand A vs Brand B for a Shopify store"
- Alternatives: "Cheaper alternatives to Brand A"
- Use case: "Which email tool is best for abandoned cart automation?"
- Pricing: "How much does email marketing software cost for 20,000 subscribers?"
- Trust: "Is Brand B GDPR compliant?"
Step 3: Sample enough runs
AI answers vary, so each prompt needs repeated runs. The number of runs determines how precise your percentages are. As a rough guide based on standard sampling math, a visibility rate of 40% measured over 50 answers has a 95% margin of error of about 14 percentage points. Over 100 answers, it narrows to about 10 points. Aggregate prompts within a cluster to reach useful sample sizes, and be cautious about reading small week-to-week changes as real movement.
Step 4: Record answers and sources
For each run, record which brands appear, which are recommended, the reasons given and every cited URL. The sources are as valuable as the mentions, because they show why a competitor is winning.
Step 5: Build the grid
Create one grid per metric, with brands as rows and models as columns, then repeat by prompt cluster. A simple heatmap is usually enough to spot patterns.
Step 6: Classify each brand-model-cluster cell
Use a four-quadrant reading to translate numbers into decisions.
| Quadrant | Your position | Competitor position | What it means | Typical action |
|---|---|---|---|---|
| Stronghold | High | Low | You own this space | Defend: keep facts current, monitor for erosion |
| Battleground | High | High | Contested ground | Differentiate: sharper proof and positioning |
| Blind spot | Low | High | Competitor owns it | Investigate sources, close evidence gaps |
| Open lane | Low | Low | Nobody owns it yet | Move early with answer-first content and coverage |
This quadrant model is a practical framework, not a published standard. Set thresholds that fit your market. In a crowded category, 30% visibility may be strong. In a narrow niche, anything below 60% may be a gap.
How to read the map
Look for model-specific blind spots. If a competitor wins in Perplexity but not ChatGPT, compare the sources each model cites. Perplexity's reliance on community and recent content often explains the gap.
Look for cluster-specific blind spots. Many brands are visible for category discovery but invisible for pricing or integration questions, simply because they never published clear answers to them.
Study the cited sources behind competitor wins. If three comparison articles and one analyst page account for most competitor citations in a cluster, those four sources are your target list.
Check accuracy, not just presence. A high visibility rate with outdated pricing or wrong positioning can hurt more than it helps.
Compare verticals carefully. Benchmarks from other industries are useful context, but source ecosystems differ. For cross-industry context, see AI search benchmarks across industries.
From map to action plan
A map is only useful if it changes priorities. A simple scoring approach:
- Weight clusters by business value. Pricing and comparison prompts usually sit closer to purchase than broad category prompts.
- Estimate effort by gap type. A missing page is faster to fix than a missing presence in community discussions.
- Prioritize blind spots in high-value clusters first, then open lanes where you can move early, then battlegrounds.
- Assign each gap an owner: content for missing answers, PR and partnerships for third-party coverage, product marketing for positioning, and technical SEO for crawler access.
- Re-run the same prompt set on a schedule so you compare like with like over time.
Common mistakes
Sampling too little. Running each prompt once produces noise that looks like insight.
Using only generic prompts. "Best CRM" tells you little. Buyers ask with constraints, and constraints change the winners.
Tracking one model. Winning in one assistant says little about another, because source preferences differ.
Ignoring non-obvious competitors. Models often recommend adjacent tools, open-source options or services instead of software.
Changing the prompt set constantly. Consistency matters for trend lines. Add new prompts as a separate cohort.
A hypothetical example
A hypothetical B2B payroll provider maps five competitors across four models and six clusters. It finds a stronghold in ChatGPT comparison prompts, driven by a detailed comparison page. It finds a blind spot in Perplexity for "payroll for contractors in multiple countries," where a competitor is cited through several recent community threads and a review site. It also finds an open lane in pricing prompts, where no brand is consistently recommended because none publish clear pricing. The team prioritizes the pricing open lane first, since it is fast to fix and close to purchase, then works on the contractor use case.
How Bob Builds AI helps
Bob Builds AI's Visibility Monitoring compares recommendation share, citations and positioning against the companies competing for the same conversations across multiple AI models. Prompt Research helps identify the buyer questions worth mapping, and Optimization Actions turns gaps into prioritized work.
FAQ
What is a competitive AI visibility map?
A competitive AI visibility map is a grid showing how often each brand in a market is mentioned, cited and recommended across AI models such as ChatGPT, Gemini, Claude and Perplexity, broken down by prompt type. It reveals where competitors are winning, which sources support their wins and where no brand has established a presence.
How many prompts do I need to benchmark AI visibility against competitors?
Most teams start with 30 to 100 prompts grouped into clusters such as discovery, comparison, alternatives, use case and pricing. Because AI answers vary, each prompt also needs repeated runs. Aggregating runs within clusters gives more reliable percentages than relying on a few single responses.
Why is my competitor visible in Perplexity but not ChatGPT?
The two systems lean on different sources. Citation research shows Perplexity relies heavily on community platforms like Reddit and recent content, while ChatGPT cites reference sources like Wikipedia more often. A competitor with strong community discussion may win in Perplexity even without an advantage in ChatGPT.
What is share of voice in AI search?
Share of voice in AI search is the proportion of AI-generated answers in a defined prompt set that mention or recommend a brand, relative to competitors. It is usually measured as a percentage across many sampled runs per model, because individual answers vary substantially.
Should I track AI ranking position against competitors?
No. Research by SparkToro and Gumshoe found the order of AI recommendations almost never repeats across runs. Visibility rate and recommendation share across many runs are more reliable measures than position.
How often should I update a competitive visibility map?
A monthly full refresh works for most teams, with weekly tracking of high-priority clusters. Update more often around launches, pricing changes, major competitor announcements or AI model releases, since those events can shift answers quickly.
Do different industries show different AI visibility patterns?
Yes. The sources AI models rely on vary by category. Community-heavy categories, regulated industries and markets dominated by analyst or review coverage each give models different material. Build your map within your own vertical and use cross-industry benchmarks only as context.
Conclusion
A competitive visibility map turns vague questions about AI presence into specific decisions. By separating results by model, prompt cluster and metric, it shows where you are strong, where competitors hold ground you could contest and where no one has claimed the answer yet. The cited sources behind each result tell you what to do next.
Start small: five competitors, three models and five prompt clusters, sampled enough times to trust the percentages. The first map usually surfaces at least one open lane you can claim quickly. When you are ready to run the map continuously across models, Bob Builds AI can automate the monitoring and help prioritize the gaps.