Paid and Organic Together

Organic sessions on the brand terms are up 40,000 this quarter, and the content team has a slide about it.

Paid clicks on the same terms are down 31,000, and the search team has a slide about efficiency. Both slides are accurate. Both are presented as wins. The two numbers in the room add up to seventy-one thousand, and the net movement on those terms is nine thousand. Even that nine thousand is not a measurement of anything: demand, bids, organic position, and competitors moved both series at once, and no arrangement of the two lines says how much of one came out of the other.

The decision it encodes

How much of this organic gain is new, and how much came out of our own paid?

The question only exists because the same customer, on the same query, can be served by two channels the company pays for differently, and each channel counts the click as its own.

The overlap is worst where the wins look best

Brand terms are the extreme case. Somebody searching your name will usually find you either way, so a paid click and an organic click on that query are closer to substitutes than anywhere else, and moving volume between them changes the cost more than it changes the outcome. Closer is not the same as identical: the published pause studies found a meaningful share of paid clicks incremental even where the advertiser ranked first organically, with the share varying widely by advertiser. Neither “mostly substitution” nor “mostly incremental” is a safe prior, which is why this component measures the rate rather than assuming it. Paid search and SEO are not independent works through the mechanism and the measurement.

Non-brand is milder and not absent. Where you rank first organically and also bid, some share of the paid clicks would have arrived free. Where you rank eighth, almost none would.

The measurement is a paid holdout

The clean way to size the overlap is to stop bidding on a set of terms in a set of markets and watch what organic does. That is a geo design run for a different purpose, and it belongs on the same test calendar.

What it returns is the share of paid clicks that organic recovers, which is the number both teams need and neither can produce alone. Modeling it from observational data is much weaker, because bidding decisions and organic rankings both respond to demand, so the correlation between them is confounded in the usual way.

One data constraint shapes the design. Search Console reports geography at the country level and withholds anonymized queries, so organic clicks by query by U.S. market is not a report anyone can pull. A geo holdout on a term set therefore reads organic from analytics, sessions and conversions by the visitor’s market, with landing pages standing in for the term set, or it separates the question into a query-level read at the national level and a market-level read of total search.

The output is a set of numbers rather than one: the gross organic change, the change in paid clicks and spend, the incremental change in total search clicks, conversions, and value, and the substitution rate that connects them, each with its interval. The substitution rate is the one that gets quoted, and the total-search line is the one the business should act on.

What to do with the answer

Report organic gains net of recovered paid clicks, or report them gross with the recovery stated alongside. Either is defensible; reporting gross without the recovery is not, once the recovery is known.

The same logic runs the other way and is more often forgotten. Cutting brand paid spend usually shows a small net loss and a large efficiency gain, and the part organic picks up is the part that was never incremental in the first place.

Maturity, honestly staged

Stage 1: brand and non-brand separated in both channels. Most of the double-count lives in brand, and splitting it makes the size of the question visible.

Stage 2: one owner for the combined number. Somebody who is neither team reports search as a whole, so the two slides become one.

Stage 3: a paid holdout sizing the recovery rate. The overlap becomes a measured quantity rather than an argument between teams.

Stage 4: organic readouts reported net. The recovery rate is applied by default, and a gross number cannot be presented without the net beside it.

What the holdout estimates Technical

Stopping paid on a term set in a subset of markets gives the change in total sessions on those terms, and the change in organic sessions. The recovery rate is the organic gain divided by the paid loss, and the incremental value of the paid spend is what remains. On brand terms, published estimates of the recovery rate run from about half to nearly all, a range wide enough that a borrowed figure is a guess, and wide enough to explain why brand paid is the most contested line in most search budgets.

Two cautions on generalizing it. The rate depends on where you rank organically for those terms, so it does not transfer from brand to non-brand or from one term set to another. And it depends on the competitive set: if a competitor bids on your brand, switching your paid off hands them the top of the page, and the recovery rate measured in that world is not the one you would get in a quieter quarter.

Running the same holdout in reverse, adding paid where there was none, is usually easier to authorize and estimates the marginal question rather than the total one, which is normally the decision on the table anyway.

Context

Best suited for

Any organization buying search ads on terms it also ranks for, which is nearly all of them, and where paid and organic are reported by different teams.

Data requirements

Query-level data for both channels at the national level, joined on the query rather than on the landing page
A brand and non-brand classification both teams accept
Market-level paid spend from the ad platform and organic sessions and conversions by visitor market from analytics, since Search Console does not report queries below the country level
Somewhere to record the recovery rate with its interval and the date it was measured

What changes by context

B2C ecommerce: Brand volume is large and the recovery rate is high, so this is where the double count is biggest in absolute terms. Shopping formats complicate it further, since the same query can return paid and organic results of different kinds.
Marketplace: Long-tail listing queries rarely overlap with paid, so the problem concentrates on category and brand terms. Measuring the rate on those and applying it site-wide will overstate the overlap considerably.
Local / omnichannel: Map and local pack results sit between the two channels and are not cleanly either, so a query-level join misses part of the substitution. Geo holdouts work well here for the same reason they work for everything else local.

When it breaks down

Programs with no paid search at all, where organic gains are unambiguous and this component is unnecessary.
Term sets too small to power a holdout, where the recovery rate has to be borrowed from a larger set and labeled as borrowed.
Organizations where the two teams have separate targets that both depend on the same clicks, in which case the number is correct and nobody will act on it.