How Aigeo Scores Are Calculated

Every number on your visibility panel traces back to a single unit of measurement: one probe — one prompt, sent to one AI engine, in one region. Aigeo runs many probes per analysis, scores each one independently, then rolls them up.

This page is the map: the full pipeline end to end, with the exact formulas. Each stage links to the page that documents it in depth.

The pipeline

1. Classify the prompt Each prompt is assigned one of four intents. Two of them name your brand, which makes "was the brand mentioned?" a meaningless question — so they get a different rubric.
2. Query the engine The prompt is sent to one engine, in one region. The answer text and every source it cited are captured verbatim.
3. Grade the answer A separate analysis model reads the answer against the rubric for that intent and returns position, sentiment, frequency, a verdict where applicable, the competitors named, the recommendations, and a category for each cited source.
4. Validate and clamp Every returned number is range-checked on the way in. An unparseable response is flagged as a failed analysis rather than silently scored.
5. Roll up Probes are averaged into panel metrics — each with its own denominator — and blended with the technical health of your own site to produce the Final Score.

1. Prompt intent decides the rubric

Intent	Names your brand?	Example	Bucket
Category	No	”best massage gun 2026”	Discovery
Scenario	No	”quietest massage gun for a 10-person startup”	Discovery
Comparison	Yes	”Dyson vs Dreame hair dryer”	Branded
Trust	Yes	”is Kospet legit in 2026”	Branded

Discovery prompts measure unprompted visibility — the AI chose to bring you up. Branded prompts measure how you are judged once you’re already in the conversation. They are scored with different rubrics and reported separately.

→ Prompt intents in full

2. One probe becomes an AI Score

For discovery prompts, three components add up:

AI Score = Position (0–85) + Sentiment (0–15) + Frequency bonus (0–8) Clipped to [0, 100]. A brand that isn't mentioned at all scores 0 on every component.

Component	Range	What it answers
Position	0–85	How much of the answer is actually about you — rank in a list, or role in prose
Sentiment	0–15	When the AI talks about you, how does it sound
Frequency bonus	0–8	Are you referenced repeatedly, or across several sections

For branded prompts, position and frequency are meaningless — the prompt already names you — so the score comes from a discrete verdict instead:

Prompt type	Verdict	AI Score band
Comparison	win the answer favors you over the alternatives	80–100
Comparison	tie genuinely balanced, “depends on your use case”	45–60
Comparison	lose favors a competitor, or is dismissive of you	0–40
Trust	legit endorsed as legitimate and trustworthy	80–100 (65–80 with caveats)
Trust	mixed some concerns, or inconclusive	45–60
Trust	scam flagged as illegitimate, or advises against you	0–40

→ The AI Score algorithm in full

3. Who does the grading

Scoring is not keyword matching. Each answer is read by a dedicated analysis model (Gemini 2.5 Flash), separate from the engine that produced the answer, using the rubric for that prompt’s intent plus the answer’s cited sources.

Its output is validated and clamped on the way in — position to 0–100, sentiment to 0–15, frequency to 0–8, final to 0–100 — so a malformed or out-of-range response can never inflate a score.

The same analysis pass also extracts the competitors named in the answer, the improvement recommendations, and the category of each cited source (third-party blog / brand website / community discussion).

4. Rolling probes up to the panel

Each panel card has its own denominator. They are not interchangeable, and the differences are deliberate:

Card	Averaged over	Note
AI Score	Every scorable probe in scope	All intents, all engines, all regions unless filtered
Sentiment	Mentioned probes only	An answer that never names you has no tone toward you
Mention Rate	Discovery probes only	Branded prompts guarantee a mention; reported separately
SEO Score	Not per-probe	Site-level, from one Lighthouse run

Final Score = 0.8 × average AI Score + 0.2 × SEO Score With no Lighthouse data, the Final Score is simply the average AI Score — the SEO term is dropped rather than treated as zero, so a missing audit never drags the score down.

→ Averages & denominators · Final Score

5. When grading fails

Occasionally the analysis call itself fails — a timeout, a rate limit, an unparseable response. Those probes are flagged not analyzed, and they are excluded from every metric rather than averaged in as zeros.

This includes the mention-rate denominator. A failed grade means “not measured”, not “performed badly”, and leaving it in the denominator would be equivalent to recording it as “not mentioned”. There is no keyword fallback that guesses at mention status: string matching can’t tell your brand from a same-named company, from an ordinary English word, or from a hit inside an echoed search URL, and every one of those errors points the same way — overstating visibility.

The answer text and citations are still stored, so the probe detail page shows the full answer with — in place of its scores.

6. Regional and per-model breakdowns

Region and per-model cards use the same formulas on a filtered subset of probes. Two consequences:

• The SEO Score is site-level. It is identical in every region, because your site is the same site everywhere. All regional variation in the Final Score comes from the AI half, and region cards label this so identical numbers don’t look like a bug.
• Cross-region comparison uses only engines whose region signal is adjustable. Engines with no geographic signal, or with a fixed home market, are reported separately under “Global engines” rather than mixed in — otherwise the comparison would look valid while comparing different denominators.

The cross-region average weights each region equally rather than each probe, so a region that lost a probe to an engine timeout doesn’t quietly shrink in weight.

→ Engines & regions

Quick reference

Metric	Raw range	Shown as	Counts which probes
Position	0–85	0–100	Discovery only
Sentiment	0–15	0–100	Mentioned only
Frequency bonus	0–8	(folded into AI Score)	Discovery only
AI Score	0–100	0–100	All scorable
SEO Score	0–100	0–100	Site-level, not per probe
Final Score	0–100	0–100	All scorable

Hover the small i next to any of these cards in the app for the short version, or see the full metric reference.