How we measure AI visibility
Observed answers, transparent denominators and a record of method changes.
Repeat the questions and settings
Weekly tracking reuses the project’s saved buyer questions, selected engines, measurement language and country. You can edit these inputs. A changed question is not interchangeable with the old question: change alerts compare matching successful question-ID and engine pairs. Model changes are identified as method changes rather than counted as visibility improvements or declines.
Live web search and current models
Measurement requests use live web-search-enabled engine flows. Google AI Overviews and AI Mode are captured as Google search experiences through a SERP provider; their identifiers below are API paths, not underlying Google model names. Engine enablement, provider availability, language support and returned AI coverage can affect each check. The configured identifiers below come from the same configuration used for new runs. Older answers retain their actual model identifiers.
- ChatGPT: gpt-5-mini
- Perplexity: perplexity/sonar
- Gemini: gemini-3.8-flash
- Claude: claude-sonnet-5-5
- Google AI Overviews: serp/google/organic/live/advanced
- Google AI Mode: serp/google/ai_mode/live/advanced
Metric formulas
Metrics use successfully completed, analysed answers. Failed answers and responses with no AI answer are excluded from the successful-answer denominator. A percentage without its coverage can be misleading, so the report supports answer-level inspection and engine filters.
- Visibility
- Visibility = successful answers naming your brand ÷ successful answers × 100.
- Share of voice
- Share of voice = brand answer mentions ÷ total brand, tracked competitor and saved alternative mentions × 100. Other named entities do not count.
- Average position
- Average position = sum of recorded positive positions ÷ successful mentioned answers with a recorded positive position. Missing positions are excluded; no recorded positions means unavailable, not zero.
- Citation rate
- Citation rate = successful answers linking your site as a source ÷ successful answers × 100. A mention and a citation are independent.
- Sentiment
- Sentiment dimensions = positive ÷ (positive + negative) usable answer mentions × 100, with at least three polar mentions per dimension. Brand traits carry extracted tones; competitor reason wording is interpreted evidence, not an independent rating. Neutral or conflicting evidence is not a polar mention.
<1 in 100
chance that ChatGPT or Google's AI gives the same list of brands twice for the same question.
SparkToro · 27 Jan 2026 · 2,961 runs of 12 prompts by 600 volunteers
Answers vary between checks
SparkToro’s research published 27 January 2026 found less than a one-in-100 chance that ChatGPT or Google’s AI gave the same list of brands twice for the same question, across 2,961 runs of 12 prompts by 600 volunteers. That’s why we track patterns across repeat checks and inspect the answers behind movements. A single percentage should not be read as a permanent rank.
Read SparkToro’s research ↗What we store
A saved check includes question and engine identity, status, full answer text, source domains and exact citation URLs, brand and competitor observations, extracted traits and recommendation reasons, and actual model identity where supplied. Search queries are stored when shared by the engine. Recommendations, alerts and plan actions keep evidence links. Historical answers are not replaced by a new configuration. Settings exposes measurement changes, and trend charts show Method markers alongside separate You action markers.
What we don't claim
We measure what AI engines actually answer; we don’t have access to their ranking systems or guarantee a position. We don’t measure visits or sales. A change after a shipped task shows correlation, not proof of cause.
Beta access
Your first check is free: up to 10 questions on 4 AI engines, no credit card. Weekly tracking, change alerts over time and the Slack summary are available to founding members during the beta; wider access is Soon. Nobody is charged during the beta.