Keyword research is not dead, but your volume column needs re-baselining

Your keyword tool still reports a volume figure with reassuring precision. That figure has quietly stopped meaning what your forecast assumes it means.

What broke

Search volume was never a measure of traffic — it was a measure of demand, which used to convert to traffic at a rate you could estimate from position. Zero-click results have been eroding that conversion for years, and generative answers have accelerated it sharply for exactly the informational queries most content programmes target.

The volume is still real. The relationship between volume and sessions is what broke, and it broke unevenly across query types, which is what makes it genuinely hard to correct for.

Re-baselining, practically

You cannot fix this with a global multiplier. What worked for us was segmenting the keyword set by answer-likelihood and applying separate expectations to each:

• Definitional and simple factual — assume heavy answer capture. Forecast conservatively or deprioritise entirely.
• Comparative and evaluative (“X vs Y”, “best tool for Z”) — still clicking, because people want to see the reasoning and the sources. This is where the commercial value has concentrated.
• Transactional and navigational — largely unaffected.
• Long-tail specific — mixed, and the hardest to predict. Measure rather than model.

What we would change in reporting

Stop forecasting sessions from volume. Forecast on a blended basis: sessions for click-likely segments, impressions and citation frequency for answer-likely ones. This is more work and it is less satisfying to put in a deck. It is also the only version that will not be embarrassing in six months.

The short version

Keyword research is not dead. Volume-to-traffic forecasting largely is, at least in the form most teams still use. Segment by answer-likelihood, forecast each segment separately, and move editorial weight toward comparative and original-data content — which, conveniently, is also what gets cited.