ChatGPT’s search market share depends on what you count

Choose one measurement object

A ChatGPT visit can include writing, coding, image work or a search-like request. A Google search query is not the same unit as a monthly ChatGPT visit. A referral click is different again. Dividing one series by another without a shared definition does not produce a defensible market share.

Specify whether the question concerns search-like requests, unique users, visits or referral clicks. State the countries, platforms and period covered. Then use numerator and denominator drawn from compatible observations.

One conversation can contain more than one search

OpenAI describes rewriting a request into one or more searches and potentially searching again after seeing results. A user question and a backend query count therefore cannot simply be added to traditional search-query totals as though they were identical units.

Some answers involve no web search at all. A brand can still be mentioned. Decide whether your definition of search includes that interaction before estimating an AI share.

A panel needs a sampling frame

For an observed panel, disclose recruitment, device coverage and the available demographic or country breakdown. IP addresses are not unique people. A large conversation count does not, by itself, show that the panel represents the whole market.

Keep work or API activity separate from consumer web activity unless both are deliberately included in the same design. A shift in the panel's platform coverage can look like a market-share change.

What the older numbers can and cannot establish

The previous version combined search-engine share, AI adoption and estimated search-like use into an “effective Google share”. Those inputs did not share a common denominator, so that derived share has been removed.

A historical fall in Google's measured referral-based search share does not identify how much went to ChatGPT. A platform can influence a decision without sending a measurable click. Report the original series under its own definition instead of filling that gap with an assumption.

Use a narrower measure for a brand decision

For content prioritisation, focus on relevant observed questions and the answers you can capture. For commercial performance, track human referrals and completed outcomes under consistent attribution rules. Neither requires inventing a global share from unrelated sources.

The Aiso methodology describes sampling limits, and the measurement guide separates agent requests from human traffic.