What we measured ourselves.
Every study here carries its date, its method and its size, because a number without those is a story. Where a newer reading contradicted an older one of ours, the newer one replaced it and we said so.
Where AI answers actually get their sources
16–18 September 2026, a 62-query panel run across answer engines: 56 of the 62 questions returned any citation, and 27 of those 56 answers (48%) cited YouTube, LinkedIn, Reddit or Quora. Those four platforms are 54 of the 301 individual citations (18%). Roughly half the answers lean somewhere the brand does not control. We first published this as a 49% share OF CITATIONS, which mislabelled the answer-level figure; it is corrected here rather than quietly dropped.
Three industries are competing for the letters AEO
2,592 probes, 1,303 distinct buyer queries: 223 of the 503 queries in the AEO cluster belonged to other markets entirely. Read it.
The Agentic Compounding Index
A buyer-side instrument for whether a marketing tool actually learns. Read it.
The state of the marketing stack
The year’s major surveys read at source, including two widely circulated numbers that do not survive contact with the original study. Read it.
The Distinctiveness Index
Where a brand sits on the ladder from existing, to retrieved, to cited, to named, to recommended — per engine. Read it.
What we do not know yet
Whether our simulated buyer panels predict real outcomes: they are advisory until checked, one low-rated idea goes to market every round so the simulator can be caught being wrong, and the first calibration will be published whichever way it lands. How stable engine answers are over time: repeat readings of the same question disagree often enough that we report trends and never a single snapshot. How much of an AI-assisted purchase can honestly be attributed: where we cannot measure it, the report says unmeasured rather than estimating.