How we actually check
your AI visibility
No black box. Here's exactly which vendors we call, how, and what the numbers do and don't mean.
Real vendor APIs, not screen-scraping
Every AEOBot check calls the same 4 vendors, through their own official APIs. Two ground themselves in live search natively; the other two are given fresh search results as context, so every answer reflects what's true right now — not stale training data.
OpenAI
GPT-4o miniFed live Serper.dev search results as context, then asked directly
GPT-4o mini has no low-cost native web search — pairing a cheap model with fresh search context gets current answers without paying OpenAI's pricier built-in search fee.
Claude
Haiku 4.5 (Anthropic)Fed live Serper.dev search results as context, then asked directly
Same reasoning as OpenAI — Claude Haiku is fast and cheap, and current search context does the rest.
Gemini
2.5 Flash / 2.5 Flash-Lite (Google)Native Google Search grounding — Gemini searches the live web itself before answering
Gemini's own grounding is well-priced and reliable, so there's no need to substitute search context.
Perplexity
SonarNative web search — Sonar is search-grounded by design
Search is Sonar's whole product; every answer already cites live sources.
What each check actually runs
AI Audit
25 real buyer-intent questions across 5 scenarios (procurement, research, buyer review, sales & marketing, category discovery), each asked to all 4 vendors — 100 real model calls per run. A full run is rate-limited by plan (7–30 day cooldown) to keep real API cost sane. Every question's full answer, per vendor, is in the report transcript.
AI Visibility
20 structured signal queries across 5 intent categories (Awareness, Consideration, Advocacy, Authority, Competitive), each checked against all 4 vendors — 80 real model calls per scan. Organic (unprompted) mentions are weighted double over brand-direct mentions, since they measure genuine unprompted recall.
The parts worth being explicit about
We don't log into ChatGPT.com, Claude.ai, or Gemini's consumer chat apps and scrape the browser session. That approach is fragile, breaks on every UI redesign, and — for automated, unattended use — sits outside those platforms' terms of service. We use each vendor's official API instead.
We don't reuse a single cached answer across customers. Every audit and every scan is a fresh set of real API calls, run against your specific brand and domain.
We don't show you a score without the receipts — every AI Audit report includes the full question-by-question transcript, per vendor, so you can read exactly what each model actually said.
Where the current limits are
A methodology page that only lists strengths isn't a methodology page — it's marketing. Here's what we'd tell you if you asked directly.
Non-grounded steps (like the initial "what does this business do" detection during setup) rely on a model's training knowledge, which can lag behind very recent changes to your site or brand — it self-corrects as models retrain, but isn't instant.
AI Audit is rate-limited by plan (a 7, 14, or 30-day cooldown between full runs) to keep the underlying real API cost sane — it's not an artificial paywall, it's a genuine cost control, and we'd rather be upfront about that than pretend otherwise.
We track 4 vendors today: OpenAI, Claude, Gemini, and Perplexity. We don't yet cover every AI surface (e.g. Grok, Meta AI) — expanding vendor coverage is on our roadmap, not something we claim to have already.
See it run on your own brand
Free, no signup, results in under 30 seconds — using the exact methodology above.
Run a Free AI Visibility Check