AI answer share of voice: AI meeting notetakers
Ask ChatGPT, Perplexity, Gemini, and Claude five buyer questions about AI meeting notetakers, and report how often Granola, Otter.ai, Fireflies.ai, and Fathom are recommended, where Granola is missing, which sources the engines cite, and how much answers vary between runs.
All 22 paid calls are done — 5 prompts × 4 engines (20 answers) plus 2 noise re-runs. Here is the complete result.
AI Answer Share of Voice — Granola (AI meeting notetaker)
Prompts (buyer questions, US localization, run 2026-09-30): 1) "best AI meeting notetaker for small teams" 2) "best free AI meeting notes app for Zoom and Google Meet" 3) "AI meeting notetaker that works without a bot joining the call" 4) "most accurate AI meeting notetaker for sales calls" 5) "Granola vs Otter.ai vs Fireflies.ai vs Fathom — which should I choose?"
Most important finding: Granola owns the bot-free niche — it is the top-ranked pick on the no-bot prompt (3 of 4 engines) and on the 4-way comparison (3 of 4) — but it is completely absent from the highest-intent generic prompt ("best free … Zoom and Google Meet": 0/4 engines) and nearly absent on sales accuracy (1/4). By raw mentions, Fireflies.ai leads the category.
Share of voice (mentions out of 5 prompts, first runs; 20 answers total)
| Brand | ChatGPT | Perplexity | Gemini | Claude | Total /20 |
|---|---|---|---|---|---|
| Fireflies.ai | 4/5 | 4/5 | 4/5 | 4/5 | 16/20 |
| Fathom | 5/5 | 2/5 | 4/5 | 4/5 | 15/20 |
| Otter.ai | 5/5 | 2/5 | 4/5 | 3/5 | 14/20 |
| Granola | 3/5 | 2/5 | 4/5 | 3/5 | 12/20 |
| Gong (sales-only) | 1/5 | 1/5 | 0/5 | 1/5 | 3/20 |
Top-ranked brand per engine: ChatGPT → Fathom (leads 3 of 5 prompts); Perplexity → Granola (the only engine where Granola leads, topping prompts 3 and 5); Gemini → Fathom (leads 2, Granola leads the 4-way); Claude → Fathom (leads 3, Granola leads the 4-way).
Prompts where Granola is missing (8 of 20 answers)
• "Best free AI meeting notes app for Zoom and Google Meet" — missing on all four engines (ChatGPT named Fathom/Otter/Fireflies/tl;dv/Tactiq; Perplexity named Fathom/Fireflies/tl;dv; Gemini led with tl;dv; Claude led with Fathom).
• "Most accurate … for sales calls" — missing on ChatGPT, Perplexity, Claude (Gong/Avoma/Krisp dominate); Gemini placed it 3rd, cited via arrows.to.
• "Best … for small teams" — missing on Perplexity only.
5 most-cited sources (engines citing, of 4)
1. techrepublic.com — 3 engines
2. meetingnotes.com — 3 engines
3. simular.ai — 3 engines
4. fireflies.ai blog — 3 engines
5. granola.ai blog — 3 engines (tactiq.io also at 3; krisp.ai was Perplexity's #1 citation twice)
Noise note: Re-running the small-teams prompt kept Fathom #1 on ChatGPT and kept Granola absent on Perplexity, but membership churned (ChatGPT dropped Fellow; Perplexity swapped Krisp→tl;dv and flipped its top two) — single-run third mentions are ±1 volatile, Granola's no-bot presence is stable.
Evidence snippets:
1. ChatGPT, no-bot prompt: "Granola — probably the simplest fit. It runs on your computer… and does not join the call as a participant."
2. Perplexity, 4-way: "Granola is best if you want bot-free notes and in-person/video meeting capture."
3. Claude, small-teams (anomaly worth noting): a section headed "5. Granola – Best Bot-Free Option" whose body text actually describes competitor Jamie — a name-only, mismatched mention (counted as a mention, flagged as low-quality).
Suggestion (clearly separate from observations): the free-plan and sales-accuracy questions are where Granola's AI share of voice is weakest; the engines' citations suggest listicle/comparison pages (not the vendor blog alone) drive those answers.
Data used: 22 paid calls — cloro (ChatGPT web ×6, Perplexity web ×6, ~$0.0028/$0.0024 each) and DataForSEO (Gemini 2.5 Flash ×5, Claude Sonnet 4.5 ×5, web search on) — total research-data cost ≈ $0.49.
Tools used (28) · 0 failed
Find a data sourceDataForSEO · ChatGPT answer check2.6sFind a data sourceDataForSEO · Perplexity answer check542msFind a data sourceDataForSEO · X posts615msFind a data sourceDataForSEO · X posts812msFind a data sourceCloro · ChatGPT answer check489msFind a data sourceCloro · Perplexity answer check573msCloroChatGPT answer check26.7sCloroChatGPT answer check26.5sCloroChatGPT answer check23.0sCloroChatGPT answer check23.8sCloroChatGPT answer check26.1sCloroPerplexity answer check36.9sCloroPerplexity answer check35.6sCloroPerplexity answer check43.1sCloroPerplexity answer check35.0sCloroPerplexity answer check64.2sDataForSEOX posts8.3sDataForSEOX posts6.6sDataForSEOX posts7.0sDataForSEOX posts7.2sDataForSEOX posts9.0sDataForSEOX posts17.0sDataForSEOX posts20.8sDataForSEOX posts16.2sDataForSEOX posts18.6sDataForSEOX posts19.3sCloroChatGPT answer check23.6sCloroPerplexity answer check38.3s
This list comes straight from the calls Voki saved while it worked. Each paid data call is listed under the provider that served it. We left out memory and shell inputs because they hold internal workspace notes.
A real job, unedited. Research-data cost covers paid data calls only, not model usage.