EasyFound methodology
How we measure AI visibility
ChatGPT, Claude, and Perplexity are measured through their APIs.
Gemini, Google AI Mode, and Google AI Overviews are acquired through a data vendor.
Gemini runs with web search on, so answers may differ from gemini.google.com.
Engine measurement basis
API · every run we got an answer
ChatGPT
Model: gpt-4o-mini
We sample 1 designated response per check.
Citations are extracted from links present in the answer text.
API · every run we got an answer
Claude
Model: claude-haiku-4-5-20251001
We sample 1 designated response per check.
Citations are extracted from links present in the answer text.
API · every run we got an answer
Perplexity
Model: sonar
We sample 1 designated response per check.
Citations come from the provider's structured citation data.
Data vendor · every run we got an answer
Gemini
We sample 1 designated response per check.
Citations come from the grounding metadata attached to the answer; proxy links are resolved to their final URL and unresolved ones are set aside.
Data vendor · every run we got an answer
Google AI Mode
We sample 3 designated responses per check.
Citations come from the data vendor's reference list attached to the answer.
Data vendor · fired days
Google AI Overviews
We sample 2 designated responses per check.
Citations come from the data vendor's reference list attached to the answer.
Counting rules
Designated response: for each prompt, engine and day we keep the earliest response the vendor captured; later samples never revise it.
AI Overviews counts fired days only: a prompt-day is in the population only when an AI Overview actually appeared.
Run live is charged at the quoted ceiling: prompts × engines, AI Overviews at one credit per prompt, whether or not an AI Overview appears.