Variability · Non-determinism of AI answers
Volatility is the instability of AI answers from one run to the next for the same prompt. It requires measuring over repeated series rather than one-off readings. A SparkToro and Gumshoe.ai study found less than a one-in-a-hundred chance of getting the same brand list twice.
Standalone definition · 45 words
What does the SparkToro study show?
About 600 volunteers ran 12 recommendation prompts on ChatGPT, Claude and Google’s AI features, for close to 3,000 answers. Brand lists varied in content, order and length. The same brands nonetheless often came back across the set of answers.
What does it mean for measurement?
Positions within an AI answer are too unstable to serve as a metric. The study’s authors recommend tracking the percentage of answers in which a brand appears over many runs. Share of voice and citation frequency are more robust than a rank.
How do you set up reliable tracking?
Repeat each prompt several times, fix the country, language and engine, keep the raw answers, calculate frequencies and intervals rather than single values, and compare whole periods. Document the method so variations can be attributed to content and not to measurement.
Frequently asked questions
Does volatility mean GEO is useless?
No. Brands in the same consideration set often come back. You steer a frequency, not a rank.
How many repetitions do I need?
The study leaves the question open. More repetitions give more stable frequencies.
Does volatility depend on the engine?
It exists on all engines tested, to varying degrees.
Related terms
Further reading
- Checklist GEO and LLM / AI 2026
- 2026 GEO Tools Comparison: The Decision Guide
- GEO tools comparison 2026: top 10 free and paid
- ← Back to the GEO glossary (114 terms)