The AI engine index
Each month: how often each AI answer engine cites a source, how many sources it cites, and what kind of sites they are. Counted from anonymised measurements across Ansengine workspaces, and published only when enough workspaces stand behind every number.
No figure is published yet.
The index uses only measurements our terms cover: from signup for a workspace created on or after 24 September 2026, and from 24 October 2026 for a workspace that existed before. Every number must draw on at least five workspaces, with no single workspace making up more than half of it.
Each month is counted on the 3rd of the month after it, and a person at Ansengine reviews it before anything appears here. Until a month clears both, this page shows no numbers rather than early ones.
How the index is counted
Which answers
Answers from the AI surfaces a buyer reads: ChatGPT (consumer app), Perplexity, Google AI Overview, Google AI Mode, Gemini (consumer app), Microsoft Copilot and Grok. Only real answers count: a run that failed, or a search where the engine showed no AI answer, is left out. Only questions that do not name a brand, measured for a brand as a whole rather than for one of its locations.
Which workspaces
Only measurements our terms cover: from signup for a workspace created on or after 24 September 2026, and from 24 October 2026 for a workspace that existed before. A workspace that opts out in Settings is left out of every month counted after that, and a deleted workspace leaves nothing behind. Our own workspaces, the demo workspace and test workspaces never count.
What counts as a source
A page the answer links to. The same page twice in one answer counts once, and links an engine makes to its own search or image pages are not counted.
The share of answers that cite a source
Counted over the answers above, with a likely range. The range is a 95 percent interval that allows for answers to the same question agreeing with each other: it resamples whole questions, and it is never narrower than a plain interval on the same counts. Answers to the same question agree with each other more than independent answers would, so the true uncertainty is wider than the range shows. Read every figure as directional.
The kinds of site
The classifier our Sources page uses, by site and page address. A brand's own site is counted by its kind like any other site, because a number drawn from many workspaces has no single brand. A share of sources carries no range: the sources in one answer are not separate draws.
The thresholds
Each number is checked on its own. The answers behind an answer figure, and the sources behind a share of sources, must come from at least five workspaces, and no single workspace may make up more than half of them. A number that fails is held back, never folded into another. When one kind of site is held back, the next smallest is held back with it, so the first cannot be worked out by subtraction.
Review
Each month is counted on the 3rd of the month after it, and a person at Ansengine publishes it. A published month is never changed. If the method changes, its version number changes with it.
What the index does not claim
- It is not a ranking of engines or of brands.
- It is not advice. A share on one engine is not a reason to act on that engine.
- It does not measure any one brand, workspace, agency client or question, and it never names one.
- The questions are the ones workspaces chose to track, not a random sample of what people ask.
- It is statistics, not model training. Nothing here trains a model on anyone's data.
The rule these numbers follow is in our terms, under Anonymised statistics and benchmarks. A workspace can opt out at any time in Settings, under Workspace. How every other number in the product is measured is on the methodology page.
