Our reading
Why this one is on our site
We sell paid media. This is the most cited piece of evidence that a large share of paid media spend does nothing, and it comes from an experiment rather than an argument. Leaving it out of a page called Research would tell you everything about how seriously to take the rest of it.
The mechanism, in one line
Search ads are shown to people who are already searching. That is the whole point of the channel and also its central measurement problem: the ad gets credit for demand it did not create.
Brand keywords are the extreme case. Somebody typing your brand name has already decided. Buy the click and you have paid for a visit that was coming free.
What this does and does not say about your account
It does not say switch off brand search. It says the value of brand search is a real question with a testable answer, and that the answer is often much smaller than the platform reports. There are decent reasons to keep buying it — a competitor bidding on your name, control over the landing page, a thin organic listing. Those reasons are defensible. "It shows a 12x ROAS in the dashboard" is not, because that is the number the experiment showed to be an artefact.
It does not transfer to unknown brands. eBay's result depends on eBay being eBay. If nobody searches your name, there is no free traffic to cannibalise, and the calculation is different.
The non-brand finding is the underrated half. Most of the spend landed on people who were already customers. Not the same problem as brand cannibalisation, but the same shape: the channel is best at reaching people you had already reached.
The test, which almost nobody runs
Turn brand search off in one geography for a fortnight. Watch total orders, not paid orders. If total holds while paid falls, the paid number was moving from one column to another.
It costs a fortnight and it settles an argument that otherwise runs for years. In practice it is rarely run, and we understand why: it is a test where the agency can only lose. Ask anyway.
Ten years on
Search in 2026 is not search in 2015. Auctions are more automated, targeting is broadly worse to observe, and a growing share of queries never reach a results page at all because an AI answer intercepts them.
None of that undoes the finding. If anything, the drift toward automated bidding — where the system optimises toward conversions it is also crediting itself with — makes the underlying confusion harder to see, not easier.
The study above is the work of its authors and is not ours. The summary and commentary on this page are written by Big Bang Story and are our interpretation, not the authors’. We do not host copies of other people’s papers — read it at the source.