Google documents the mechanism itself
This is not vendor speculation. Google's guide for succeeding in AI search states that its AI features are grounded in the Search index and core ranking systems, and that they can issue multiple related searches in parallel on behalf of one user query — the query fan-out technique. Official primary developers.google.com
The same documentation sets the entry conditions: to be eligible to appear in AI experiences, a page must be indexed and eligible to be shown as a search snippet. Official primary developers.google.com Put those two statements together and the shape of the new contest is clear: one buyer query becomes a family of sub-questions, and candidacy is decided per sub-question — not per keyword.
The research lineage runs years deep
Decompose-and-recombine is not a marketing novelty; it is a published line of research. Work presented at ACL 2019 showed that breaking a complex question into sub-questions and rescoring the answers globally improves multi-document reasoning. Peer-reviewed aclanthology.org What was a research technique for question answering is now, per Google's own documentation, part of how commercial AI search retrieves and grounds its answers. The trajectory has been visible for years — the measurement practice just hasn't caught up.
Your keyword table was already undercounting question demand
Here is the uncomfortable part for anyone running US demand planning off a query report. Google Search Console's own documentation explains that rare queries are anonymized and omitted from performance results to protect user privacy, and that the reporting is centered on top rows — so the sum of visible rows can differ from chart totals. Official primary support.google.com · developers.google.com
Long, specific buyer questions are exactly the kind of query that lands in the rare, anonymized tail. The decision-grade reading of the official documentation: a query showing zero rows in your table is not evidence of zero demand. The question family your buyers actually use can be invisible in the very tool most US teams treat as the census of demand — by the tool's own documented design.
Three unit errors that follow from measuring keywords alone
1. Coverage error — the engine contests a family of sub-questions; you track one
head term.
2. Visibility error — the question-shaped tail is anonymized out of your query
report, so the demand you most need to see is the demand you least observe.
3. Inference error — repeated checks of the same query add observations, not
independent evidence. The unit that supports a decision is the question family crossed
with the page that answers it, observed over repeated windows.
Meanwhile, the questions keep moving into conversations
The question-shaped demand is not hypothetical. An NBER working paper analyzing 1.5 million real ChatGPT conversations found that more than 70% of usage now happens outside of work, up from 53% two years earlier — everyday people asking everyday questions, including the ones that end in purchases. NBER working paper nber.org
What to measure instead: question families and candidate entry
The fix is a change of unit, and it pays off immediately in decision quality.
Measure question families, not keywords. Map the buying questions in your category — the comparisons, fit checks, implementation, and switching questions your buyers actually pose — and treat each family as one demand object. That is the unit the engine is reasoning in.
Measure candidate entry, not just rankings. Google's documented entry conditions mean the first question for any page is whether it enters the contest at all for each sub-question in the family. A page can rank respectably on a head term and still be absent from most of the family the engine actually runs.
Measure repeatedly, and read the unit honestly. One observation of one query is a point estimate, not a picture. Decisions hold up when they rest on the question family and the answering page, observed across repeated windows — that is the standard CiteAngle builds its US measurement on.
This is the demand your keyword tools never show you: the families you have not mapped, the sub-questions where you never enter as a candidate, and the competitors who are already the default answer there. Every one of those is fixable — once it is measured.