You can't rank on ChatGPT the way you rank on Google — but you can measure your share of voice in AI answers over time. Traditional search gives you ranking reports; AI search gives you nothing by default. This post fixes that with a repeatable, mostly-free quarterly routine.
Why Share of Voice matters now
AI assistants increasingly own the "which tool should we use" question. Rankings fragment across ChatGPT, Perplexity, Claude, and Gemini — and none of them exposes a ranking dashboard. Share of Voice (SoV) is the practical substitute: of the answers that mention your category, how often do they mention you?
The goal isn't perfection. It's a trend line you can act on.
Building your query set
A good query set covers the three ways buyers find tools:
- Category — "best customer discovery tools for SaaS"
- Comparison — "Needle vs GummySearch" (or your category vs a known rival)
- Alternative — "alternatives to [incumbent] for small teams"
Keep 9–12 queries total (3 per category type × your niche). Write them down and never change them — comparability beats cleverness.
The free routine: prompt runs + scoring rubric
Twice a year is baseline; quarterly is better. For each query, in each engine:
- Run the query (fresh session, no logged-in personalization where possible)
- Score the answer with this rubric:
| Score | Meaning |
|---|---|
| 0 | Category mentioned, you are not |
| 1 | You are mentioned (listed or named) |
| 2 | You are recommended (positive) |
| 3 | You are recommended first |
- Log the score and the cited sources (note which domains the answer cited — this reveals your real third-party estate)
30–45 minutes per quarterly run, three engines, twelve queries. That's the whole methodology.
Tooling for scale
When manual runs feel limiting:
- Limy — AI visibility tracking plus analytics
- Otterly.ai — citation and source tracking
- SE Ranking GEO — generative-engine tracking in a familiar SEO interface
- Free start — the GEO & LLM Site Analyzer checks your own site's machine-readability
Use tools for breadth (more queries, more engines), keep the manual rubric for quality (how were you framed, not just whether).
From measurement to action
SoV numbers are only useful with an action loop:
- Score dropped → check the cited sources: did a comparison go stale? Did a review volume shift? Did your community presence fade?
- Score flat but low → your third-party estate needs work: reviews, comparisons, community presence
- Score rising → double down on whatever moved it (usually a combination of content, listings, and conversation)
The quarterly review should feed directly into the AI visibility audit checklist — docs, comparisons, llms.txt hygiene, entity consistency.
The quarterly calendar
| Quarter | Focus |
|---|---|
| Q1 | Baseline run + fix entity consistency + update comparisons |
| Q2 | Re-run + launch one community presence push |
| Q3 | Re-run + refresh docs and listing descriptions |
| Q4 | Re-run + annual summary + next-year plan |
Each run takes 45 minutes. The trend line after four runs is worth more than any single screenshot.
GEO & LLM Site Analyzer · Browse free Trending Problems · Pricing