How are metrics calculated at LLM Pulse?
When AI assistants like ChatGPT, Perplexity, or Googleâs AI answer questions about topics in your industry, are they talking about you? Are they recommending you? And when they do mention you, is it in a good light?
LLM Pulse tracks three core metrics to measure your brandâs AI visibility: Mentions , Citations , and Sentiment. This guide explains what each metric means and how itâs calculated.
Quick Overview
| Metric | What It Measures | Why It Matters |
|---|---|---|
| Mentions | When AI says your brand name | Brand awareness in AI responses |
| Citations | When AI returns your website as a visible or background source | Source visibility across AI platforms |
| Sentiment | How positively AI talks about you | Brand reputation in AI ecosystem |
Mentions
A mention occurs when your brand name appears in an AIâs response text.
Example
If someone asks âWhat are the best project management tools?â and the AI responds:
âPopular options include Asana, Monday.com, and Trello. Asana is particularly known for its timeline viewsâŠâ
Asana, Monday.com, and Trello each received a mention.
How mentions are detected
- We search for your brand name and any alternative names youâve configured (e.g., âCoca-Colaâ and âCokeâ)
- We use word-boundary matching to avoid false positives (âAppleâ wonât match inside âPineappleâ)
- If your brand name only appears inside a URL, it counts as a citation, not a mention
Mention metrics in your dashboard
- Total mentions: Number of AI responses where your brand was mentioned
- Brand Visibility: Percentage of responses that mention you
- Mention position: The position for mentions in a specific prompt response
- Competitor comparison: How your mentions compare to competitors
Citations
A citation occurs when an AI returns your website as a visible link or background source reference.
Example
âFor SEO guides, check out Moz (moz.com/learn/seo) and Ahrefs (ahrefs.com/blog).â
Both Moz and Ahrefs received citations here.
How citations are detected
- We extract visible links and background source references
- We match domains against your website and competitors
- We filter out images, favicons, and CDN links
- We track the position of visible citations (1st, 2nd, 3rd, etc.)
Why citation position matters
Average Citation Position shows where visible citations typically rank. Background citations have no visible position and are excluded from this metric.
| Position | Impact |
|---|---|
| 1st | Highest visibility and click likelihood |
| 2nd-3rd | Still prominent, good visibility |
| 4th+ | Lower visibility, may require scrolling |
Citation metrics in your dashboard
- Total Citations: Every citation in the filtered AI answers, repeated pages included (the first card on the Citations Overview)
- Citing Responses: Number of AI responses that cite a source (your site, or a domain, host or page in the Citations tables) at least once, as a visible or background link. A response that cites it several times counts once
- Citation rate: Percentage of tracked responses that cite you, including background sources
- Average position: Your typical ranking across visible citations
- Competitor comparison: How your citations compare to competitors
Sentiment
Sentiment measures the tone of how AI talks about your brand: positive, neutral, or negative.
You could have many mentions, but if theyâre negative, thatâs a problem. Sentiment helps you understand the quality of your visibility, not just the quantity.
Sentiment levels
| Level | Score | Meaning |
|---|---|---|
| Very Positive | +1.0 | Enthusiastic recommendation, strong praise |
| Positive | +0.5 | Favorable mention, mild endorsement |
| Neutral | 0.0 | Factual, neither positive nor negative |
| Negative | -0.5 | Critical, mentions drawbacks |
| Very Negative | -1.0 | Strong criticism or warnings |
How sentiment is analyzed
We use AI to analyze AI responses. When your brand is mentioned or cited, our system sends the response to an advanced language model that evaluates:
- The overall tone toward your brand
- Specific praise or criticism
- Key topics discussed about you
This approach catches nuances that simple keyword analysis would miss. For example:
âWhile Brand X is the industry leader, many users find their pricing prohibitive.â
This is mixed: positive (industry leader) with negative elements (pricing criticism). Our analysis captures both aspects.
Sentiment metrics in your dashboard
- Sentiment distribution: Breakdown across all five levels
- Net sentiment: (Positive + Very Positive) minus (Negative + Very Negative)
- Sentiment trends: How perception changes over time
- Competitor comparison: Your sentiment vs. competitors
Brand Visibility Score
Your Brand Visibility Score combines mentions into a single percentage:
Brand Visibility Score = (Mentions Ă· Total Responses) Ă 100
One response can mention several brands, so adding their visibility percentages does not produce an âOthersâ rate.
A 45% visibility score means your brand appears in nearly half of all tracked queries relevant to your business.
AI Visibility Score (position-weighted)
The AI Visibility Score goes one step further than the visibility percentage: every response where your brand appears contributes points based on its position. First mention earns 100 points, second earns 50, third earns 33.3, and so on. The points of your date range are then divided by its tracking runs (one per day, week or month, depending on how often your prompts are tracked), so the score describes an average run: a longer range does not inflate it, and a prompt you add mid-week counts in that week's run even though it answers on its own day. It is not a percentage and it is not capped at 100, so reaching the hundreds or thousands is normal for a project with many prompts.
Every page reads the score the same way. The Overview, the projects dashboard, Custom Reports, the Overall column in AI Model Insights and the API show the same number for the same dates and filters. Chart points follow the same rule for their own period, and an AI model's cell in AI Model Insights uses the runs that model answered. Competitors are divided by your brand's runs, so their scores compare directly with yours. Brand Visibility (mention rate) is the percentage metric for a selected range.
The API summary returns that same AI Visibility Score. For rate, position and sentiment metrics, it pools the whole date range, as the Overview dashboard, its PDF and Custom Reports do: visibility is all mentions divided by all responses, citation rate all citations divided by all responses, and an average position the average of every position recorded in the range. They therefore show the same headline numbers for the same dates and filters. A chart point covers only its own period, so a range's total is not the average of its points when periods have different response counts.
Mention rate across AI models
Each tracked prompt runs on the five core models plus any add-ons enabled for the account. Mention rate is calculated for each model first and then combined, so a strong score on one model can hide a weak one on another.
The AI Model Insights view breaks this down for you. It shows your mention rate side by side for each model and how your mentions split across models, so you can see which models drive most of your visibility and where you have room to grow.
Understanding metric combinations
| Scenario | What It Suggests |
|---|---|
| High mentions, low citations | AI knows your brand but doesnât link to your site |
| High citations, low mentions | Youâre used as a source but not recommended by name |
| High visibility, negative sentiment | Youâre well-known but with reputation issues |
| Low visibility, positive sentiment | Good reputation with room to grow awareness |
Filtering your data
By AI Platform
Different platforms may show different results:
- ChatGPT: Conversational AI assistant
- Perplexity: Research-focused, citation-heavy
- Google AI Overview: AI summaries in search results
- Google AI Mode: Conversational search experience
By Geography & Language
Your visibility may vary by region. Filter by country and language to see regional performance.
By Time Period
Track daily, weekly, or monthly trends to measure progress and spot changes.
Daily reporting uses the Madrid calendar date when the response record was created. For API and MCP filters, a date-only end date includes the whole Madrid day; an explicit timestamp keeps its exact time. Sentiment time series use the response's recorded execution time, so records near a date boundary can fall into different windows. Trend charts draw one point per tracking period with responses: a period without responses has no point, and a period with responses but no mention is a real zero.
Benchmarks
Performance varies by industry, but here are general guidelines:
| Metric | Good | Great | Excellent |
|---|---|---|---|
| Mention Rate | 20%+ | 40%+ | 60%+ |
| Citation Rate | 10%+ | 25%+ | 40%+ |
| Average Position | Top 5 | Top 3 | #1 |
| Net Sentiment | +20% | +40% | +60% |
Note: Compare yourself to competitors in your industry for the most meaningful benchmarks.
Improving your metrics
Low mentions?
- Create more content that AI models can learn from
- Get mentioned on authoritative industry sites
- Ensure consistent brand naming across the web
Low citations?
- Create comprehensive, authoritative resources
- Improve technical SEO for better discoverability
- Build topical authority in your niche
Poor citation position?
- Strengthen your expertise signals
- Create definitive guides that AI would reference first
- Build more high-quality backlinks
Negative sentiment?
- Review the specific criticisms in your dashboard
- Address legitimate concerns about your product or service
- Monitor for improvements after making changes