Vision & methodology
A live tracker for which GEO platforms are actually winning — measured by what the LLMs themselves say.
The honest test of an AI-visibility platform is whether the LLMs themselves recommend it. We measure that — with a fixed prompt panel, transparently — and publish the leaderboard. Funding, headcount, and customer logos are lagging indicators. Share-of-voice inside the answer engines is the leading one.
Ten venture-backed platforms, founded ≤3 years ago, with notable customer references and ≥$2M raised:
Profound · Brandlight · Peec AI · Bluefish AI · Evertune · Scrunch AI · The Prompting Company · AthenaHQ · Otterly.AI · Azoma
Across five LLM surfaces, all run with live web search enabled:
| Surface | Grounding mechanism |
|---|---|
| Perplexity Sonar-Pro | Native — built around live web search |
| Google AI Overviews (via DataForSEO) | Native — Google's search index |
| ChatGPT (GPT-5.5) | OpenAI web_search tool |
| Claude Sonnet 4.6 | Anthropic web_search tool |
| Gemini 3.1 Pro | Google google_search grounding tool |
What “web-grounded” means and why it matters
A web-grounded LLM searches the live internet beforeanswering, rather than replying purely from training data. This matters because a real CMO asking ChatGPT today gets ChatGPT-with-search by default — measuring the raw model is measuring a fossil. For a category like GEO that didn’t exist in volume until 2024–25, the gap between grounded and ungrounded answers is dramatic. All five surfaces run grounded, so the leaderboard reflects what marketers actually see today.
175 prompts across 10 intent categories, designed to span the question space a real CMO, agency lead, or founder would actually ask:
| Category | Count | What it captures |
|---|---|---|
| Discovery (no jargon) | 15 | "I want to track my brand in ChatGPT — what tools exist?" |
| Jargon-aware | 12 | Direct asks for "GEO," "LLMO," "AI SEO" platforms |
| Head-to-head | 25 | Every meaningful pairwise + 3-way comparison |
| Problem-framed | 25 | Per-LLM (Overviews, Rufus, Sparky, Perplexity) and per-need |
| Investor / funding | 8 | Funding-stage, moat, market-leadership angles |
| Buyer journey | 35 | 6 verticals × 4-6 personas |
| Category leader | 10 | "Who is the leader?" "Most popular GEO software?" |
| Feature-specific | 25 | Citation tracking, sentiment, coverage, integrations, alerts |
| Adversarial / negative | 10 | "What's wrong with [vendor]?" "Why pick X over Y?" |
| Implicit / no-tool-mention | 10 | "How do I make ChatGPT recommend my brand?" |
The 35 highest-weight prompts (head-to-head + category-leader) run 3× per cadence to control for response stochasticity. Total: 1,225 LLM responses per run.
→ Read every prompt in the panel — all 175, grouped by category. Open methodology is only open if the prompts are auditable.
For each response in which a company is mentioned at ordinal position p:
subscore = 1.0 + (0.5 if p == 1 else 0) + 1/p contribution = subscore × prompt_weight × surface_weight
The composite Share-of-Voice (SoV) is the weighted average contribution across every response. We also publish three legible sub-metrics: mention rate (% of responses where the company appears), first-mention rate(% where they’re named first), and average mention order (the ordinal index of their first appearance when mentioned). Direct-ranking prompts get 1.5× weight because they ask the LLM for an explicit recommendation.
Surface audience weighting
Surfaces are not weighted equally. A CMO cares more about Google AI Overviews (billions of searches/day) than Claude (~30-50M monthly users). The scoring formula reflects that:
| Surface | Approx weekly reach | Weight |
|---|---|---|
| Google AI Overviews | Billions of Google searches/day | 2.5× |
| ChatGPT (GPT-5.5) | ~400M weekly users | 2.0× |
| Gemini 3.1 Pro | Hundreds of millions | 1.5× |
| Claude Sonnet 4.6 | ~30-50M monthly users | 1.0× |
| Perplexity Sonar-Pro | ~10-15M monthly users | 0.75× |
Numbers are best-public-estimates as of mid-2026. Per-surface mention rates are still published in the per-company drilldowns, so you can reweight in your head.
- This isn’t usage data. We measure the published narrative LLMs draw from, not customer counts or ARR.
- Stochasticity is real.We address it with prompt diversity and 3× repetition on the highest-weight prompts, not by pretending the noise doesn’t exist.
- The panel is curated. A different curator would produce a different leaderboard. We publish the full prompt list and welcome challenge.
- We don’t (yet) track sentiment. A positive and a negative mention both count as equal visibility. The scorer measures whether a company was named and how prominently, not the verdict.
- No paid spots, no vendor input.Companies don’t pay to be tracked; the list is BRXND’s editorial pick.