Skip to main content

​
KPIs & metrics

Finseo measures how visible your brand is in AI answers from ChatGPT, Perplexity, Gemini, Claude and other models. This page defines every metric precisely — what it counts, how it is calculated, and what a good number looks like. All metrics are available in the dashboard, the REST API and the MCP server, and every metric shows a change indicator comparing the selected timeframe to the previous period of the same length.

​
The running example

The formulas below all use this example: you track 20 prompts across 2 models, so one day produces 40 AI answers. Your brand is Acme.
  • Acme appears in 18 of the 40 answers.
  • Across all 40 answers, AI models name brands 120 times in total (all brands combined).
  • Acme is named 24 times (a single answer can mention you more than once).

​
Presence metrics

​
Visibility

The share of AI answers in which your brand appears at all.
Example: Acme appears in 18 of 40 answers → Visibility = 45%. Visibility is the headline metric: it answers “when someone asks AI about my category, how often am I part of the answer?” It counts each answer once, no matter how often you are mentioned inside it.

​
Mentions

The total number of times your brand is named in the selected timeframe. Unlike Visibility, Mentions counts every occurrence — if one answer names Acme three times, that is 1 toward Visibility but 3 Mentions. Example: 24 mentions across 18 answers. Use Mentions to gauge how prominently you feature inside answers; use Visibility to gauge how often you feature at all.

​
Model Visibility

Visibility broken down per AI model (ChatGPT, Perplexity, Gemini, Claude, Grok, Mistral, DeepSeek, Copilot, Google AI Overview, Google AI Mode). Models draw on different sources and training data, so it is normal to be strong on one model and invisible on another. A gap between models usually points to which sources each model relies on — check Top Sources per model to find out where to invest.

​
Competitive metrics

​
Share of Voice (SoV)

Your slice of the total brand conversation — your answers as a percentage of all brand appearances across every brand.
SoV differs from Visibility because the denominator is the whole market, not your prompt set. Suppose in our example the 40 answers produce 90 brand-appearances across all detected brands, 18 of which are Acme → SoV = 20%, even though Visibility is 45%. You can have high Visibility and low SoV when competitors are mentioned alongside you in almost every answer. Finseo shows SoV against two denominators so the number is never ambiguous:

​
Average Position

The average order in which your brand is named among the brands in an answer. Displayed as an absolute value, e.g. #2.6. Lower is better. If an answer recommends “1. Acme, 2. Beta, 3. Gamma”, Acme has position #1 in that answer. Averaged across all answers where you appear:
Example: Acme is named first in 9 answers, second in 6, fourth in 3 → (9×1 + 6×2 + 3×4) ÷ 18 = #1.8. Position matters because AI answers behave like rankings: the first brand named typically frames the recommendation. Ties (two brands named in the same breath) share the same ordinal.

​
#1 Share

The percentage of your answers in which you are the FIRST brand named.
Example: first in 9 of 18 answers → #1 Share = 50%. This is the “pole position” metric — it isolates how often you lead the answer rather than just appearing in it.

​
Top-3 Share

The percentage of your answers in which you are among the first three brands named.
Example: 15 of Acme’s 18 answers have it in the top three → Top-3 Share = 83%. Useful for long list-style answers where being in the opening cluster matters more than being strictly first.

​
Head-to-Head (H2H)

Against one specific competitor: the percentage of answers naming BOTH brands where yours is named first. Ties are excluded.
Example: Acme and Beta appear together in 10 answers; Acme is named first in 7 → H2H vs Beta = 70%. H2H removes the noise of answers where only one of you appears — it is the cleanest “who does AI prefer?” comparison.

​
Placement metrics

​
Mention Depth

How deep in the answer text your mention appears, on average. 0% = the very top of the answer, 100% = the very end. Lower is better.
Example: your mention starts at character 300 of a 1,200-character answer → depth = 25% for that answer. A brand can hold Average Position #1 with 40% depth when answers open with a long preamble before naming any brands — that is why Finseo reports order and depth as separate metrics.

​
Source metrics

​
Citations

The number of AI answers that cited one of your domains as a source. Each answer counts once, even if it links your domain several times. Citations measure something different from Mentions: a mention is the AI talking about you, a citation is the AI sourcing from you. You can be mentioned without being cited (the model knows you from training data) and cited without being mentioned (your content informs an answer about someone else).

​
Citation Share

Your citations as a percentage of all brands’ citations in the project.
Example: AI answers cite Acme’s domains 30 times and all tracked brands’ domains 150 times → Citation Share = 20%. This is the source-side equivalent of Share of Voice: it shows who owns the underlying evidence AI answers are built on.

​
Quality metrics

​
Sentiment

How positively AI models describe your brand, scored 0–100. Finseo analyzes the language around each mention — words like “reliable”, “market leader” or “excellent support” score positive; “expensive”, “complicated” or “mixed reviews” score negative. Scores roughly read as: Sentiment is tracked per aspect (price, quality, support, …) and per competitor, so you can see not just that perception dipped but which aspect drove it.

​
Reading the metrics together

The metrics form a funnel — each answers a different question:
A common pattern: Visibility is fine but Share of Voice is falling. That means the market conversation is growing faster than your presence in it — competitors are being added to answers you used to dominate. Check the competitor gap analysis to see which prompts drive the shift.

​
Accessing metrics programmatically