What AI engines we track
ChatGPT, Perplexity, and Google AI Mode — how each one differs, what they return, and why we weight them the way we do.
We measure brand visibility across three AI surfaces that together account for the lion's share of AI-mediated discovery traffic in 2026. Each one returns a different kind of answer, with different citation behaviour, so we treat them as separate signals and combine them at the end.
The three engines
ChatGPT (OpenAI)
The largest AI-search audience by daily active users. ChatGPT answers tend to be longer and more discursive than Perplexity, and citations are optional — the model decides when to surface a source. We use the official OpenAI Responses API with web search tooling enabled when fresh context is required.
Default weight in the GEO Score: 0.45.
Perplexity (Sonar)
Built for retrieval — every answer comes with an explicit list of citations and excerpts. Perplexity tends to attribute claims more rigidly than ChatGPT, which is why it's a great signal for "is the brand actually backed by web evidence?". We use Perplexity's official Sonar API.
Default weight in the GEO Score: 0.25.
Google AI Mode
The dedicated AI tab on google.com — Google's own conversational search
surface, distinct from classic AI Overviews above the organic SERP.
Google AI Mode reaches roughly 1.5 billion users per month, which is
why it carries the second-largest weight despite being a younger
surface. We measure it through DataForSEO's
serp/google/ai_mode/live/advanced endpoint to avoid scraping
google.com directly. See the dedicated
Google AI Mode article for details.
Default weight in the GEO Score: 0.30.
How they compare side by side
| Property | ChatGPT | Perplexity | Google AI Mode |
|---|---|---|---|
| Citation density | Optional | High (numbered) | High (inline) |
| Average answer length | Long, discursive | Medium, structured | Long with tables |
| Tone | Conversational | Reference-like | Encyclopedic |
| Refresh latency for new sites | Days–weeks | Hours–days | Days |
| Best signal it gives | "How AI tells the story of your brand" | "Who's actually cited as a source" | "What Google's own LLM thinks" |
| Mention extraction | LLM-classified | Engine-native references | Engine-native references |
Why three engines, not five
We deliberately don't track Gemini chat, Claude, Grok, or Meta AI yet. Most B2B buyer journeys flow through the three above; the others are either niche audiences (Gemini) or have no public, scrape-friendly search surface (Meta AI). Adding more engines would dilute the score with noisy signals before we had a clean picture.
When we do extend, our priority order is: Gemini chat (when usage among B2B buyers crosses ~20% of our audience), then Claude search (when it ships a public retrieval mode), then Meta AI (if it gets a citation surface that isn't behind a Facebook wall).
What we don't claim
Where the math lives
The per-engine effective rate is computed exactly the same way for each engine — see the GEO Score article — and the combined score is a weight-normalised average across whichever engines returned data.
Before any engine cites you, the bot has to fetch you
Each engine runs its own crawler — OAI-SearchBot for ChatGPT,
Claude-Web for Claude.ai, PerplexityBot for Perplexity, and
Googlebot plus the Google-Extended opt-out for Google AI Mode. Your
robots.txt and HTTP headers decide whether they can fetch your site
in the first place. See
How AI bots discover your site for
the bot taxonomy, and the
AI Crawlability Monitor for
the audit that catches accidental blocks.