Home/Reference/AI Search score
Defined term · Presence Growth Engine metric
AI Search score
How often AI assistants recommend and cite you when buyers ask about your category without naming you.
Definition
Your AI Search score shows how often AI assistants recommend and cite you when buyers ask about your category without naming you. We combine AI Recommendation (weight 24) and AI Citation (weight 10) across ChatGPT, Google AI, Gemini, Claude and Perplexity, weighted by how widely each is used. It is the same number as the AI Search tile on your Scoreboard.
Coverage
Which answers count
- Only prompts that do not name the brand or its aliases. Perception prompts are excluded.
- Only the standard engines: ChatGPT (including ChatGPT search), Google AI Overviews and AI Mode, Gemini, Claude and Perplexity. Bing Copilot, Grok, DeepSeek and Meta AI are shown as diagnostics only.
- Every valid answer in the last 28 days counts. Fallback results and empty answers are excluded.
- Retired prompts still count. Each one is re-measured at least once a month, and a prompt not re-measured for 28 days drops out until it is re-measured.
Measurement
The two signals
AI Recommendation
Recommended counts 1, Mentioned counts 0.5, and Negative or Absent counts 0.
AI Citation
Only answers that cite at least one source are counted. An answer counts 1 if it cites the brand's own domain, else 0.
Method
How answers become a rate
Each prompt counts once per engine, using the mean of its runs, so prompts that run more often do not carry more weight.
ChatGPT
Weight 0.41.
Google AI
Weight 0.25.
Gemini
Weight 0.21.
Claude
Weight 0.08.
Perplexity
Weight 0.05.
The weights are re-normalised over the engines the site tracks. The 95% range on each rate uses the Wilson method on the effective sample, so the range is not symmetric.
Curves
From rate to score
S(x) = 100 × Φ(ln(x / x50) / σ), with σ = ln(x90 / x50) / 1.28155 and S(0) = 0. AI Recommendation uses x50 15% and x90 50%, so 30% scores 77.0. AI Citation uses x50 5% and x90 25%, so 10% scores 71.0.
These are the v1.0 defaults. The x50 and x90 values will be calibrated at a release.
Formula
The headline
Score = (24 × AI Recommendation + 10 × AI Citation) / 34, taken over live signals.
- An engine is live when it has at least 60 valid answers from at least 10 distinct prompts in 28 days.
- A signal is live when live engines carry at least 50% of the tracked weight.
- Rated means coverage of 80% or more. Provisional means 50 to 79%. Unrated means below 50%.
Below the Building threshold.
Coverage and signal are still forming.
Consistently recommended and cited.
The default answer across the category.
A band only changes after a 1.5-point move held for 7 days.
Confidence
Range and reliable change
The tile range is score ± 1.96 × SE. Minimum detectable change = 1.96 × √2 × SE. A smaller move shows as "No reliable change". regencleo.ai on 1 October 2026 scored (24 × 23 + 10 × 82) / 34 = 40.4, band Building, and has a minimum detectable change of about 17 points.
Version break
Why your old GEO Score was higher
- The old score included prompts that name your brand.
- It carried forward retired prompts at their last result.
- It weighted every engine equally, including Bing Copilot.
- It capped Share of Voice at three competitors, so any share above 25% scored 100.
Scores before the version break used the old GEO Score formula and are not comparable. The "presence is strict" rule and the competitor cap are both retired, not moved.
Diagnostics
What the GEO page also shows
The diagnostics (Found, Preferred, Cited, Perception, Reliability) explain the score. They are not separate scores and do not add up to it.
FAQ
Frequently asked questions
Your AI Search score shows how often AI assistants recommend and cite you when buyers ask about your category without naming you. It combines AI Recommendation (weight 24) and AI Citation (weight 10) across ChatGPT, Google AI, Gemini, Claude and Perplexity, weighted by how widely each is used, into Score = (24 x AI Recommendation + 10 x AI Citation) / 34, taken over live signals.
The old GEO Score counted prompts that named your brand, carried forward retired prompts at their last result, weighted every engine equally including Bing Copilot, and capped Share of Voice at three competitors, so any share above 25% scored 100. Scores before the version break used the old GEO Score formula and are not comparable. The "presence is strict" rule and the competitor cap are both retired, not moved.
Only prompts that do not name your brand or its aliases, run on the standard engines: ChatGPT, Google AI Overviews and AI Mode, Gemini, Claude and Perplexity. Every valid answer in the last 28 days counts, and a retired prompt still counts as long as it is re-measured at least once a month.
The tile range is score plus or minus 1.96 times the standard error. The minimum detectable change is 1.96 times the square root of 2 times the standard error, and a smaller move shows as No reliable change.
See your own AI Search score.
Scan your presence →