Glossary
GPI™ (Generative Presence Index)
A single executive metric for brand visibility in AI answers.
Read it with LLM-Score and Share of Voice to understand both quality and volume.
Definition
GPI™ (Generative Presence Index) combines four signals into one score: mention rate, factual correctness, sentiment, and model coverage. Use it when leadership asks one question: are we becoming more visible in AI answers month over month? GPI rewards consistency across models, not one lucky mention in one app.
Executive diagnostic map
| Signal pattern | What it usually means | Immediate move |
|---|---|---|
| GPI up, LLM-Score up, SoV up | Broad healthy growth | Keep prompt pack fixed and scale winning pages |
| GPI up, LLM-Score down | Mentions increase but quality drops | Fix factual blocks before publishing more content |
| GPI flat, SoV up | You are present but not trusted | Improve citations and source quality |
| GPI down in one region only | Coverage imbalance | Expand regional model set and local sources |
Worked example
Suppose your April to June trend looks like this:
| Month | Mention rate | Correctness | Sentiment | Coverage factor | GPI |
|---|---|---|---|---|---|
| Apr | 0.44 | 0.73 | 0.61 | 0.82 | 46 |
| May | 0.51 | 0.76 | 0.64 | 0.86 | 52 |
| Jun | 0.57 | 0.79 | 0.66 | 0.89 | 57 |
Mini trend chart (GPI):
Apr 46 ▇▇▇▇▇▇▇▇▇
May 52 ▇▇▇▇▇▇▇▇▇▇▇
Jun 57 ▇▇▇▇▇▇▇▇▇▇▇▇▇
Interpretation: this is a healthy upward move because all four components rise together. If GPI rose only from coverage expansion while correctness dropped, you would treat that as fragile growth.
How it's computed
The score is built from your prompt pack runs. Each answer is scored for presence, accuracy, and tone, then aggregated across models with coverage weights. Example: if you sell in CIS but do not track YandexGPT and GigaChat, your GPI should be treated as incomplete even with strong ChatGPT results.
How to read it
Track direction first. A rising GPI after content or schema updates usually means your changes are being picked up. If Share of Voice rises while LLM-Score falls, you are getting mentioned more often but less accurately. Fix correctness before scaling volume.
30-day GPI improvement playbook
- Freeze one prompt pack for 4 weeks so movement is comparable.
- Pick the 10 highest-intent prompts and patch factual errors first.
- Add two authoritative source pages (pricing, integrations, legal facts) and mark them with schema.
- Re-run weekly and only then broaden content scope.
Rule of thumb: stability first, volume second.
Practical benchmark bands (B2B SaaS, mid-competition)
| Stage | Typical GPI band | Reading |
|---|---|---|
| Early baseline | 20-38 | Sporadic mentions, weak consistency |
| Program in motion | 39-55 | Clear progress, still model gaps |
| Scaled execution | 56-72 | Strong repeatable visibility |
| Category leader | 73+ | High trust and resilient presence |
Use bands as orientation, not as universal truth: niche volatility and model mix can shift them.
GPI™ (Generative Presence Index) vs LLM-Score
LLM-Score is the core quality score for one check stream. GPI is the executive roll-up for cross-model and cross-market reporting.
When to use
- Executive summaries and investor updates.
- Comparing two regions or two sub-brands on one axis.
- Spot-checking whether a GEO program moved the needle overall.