Two sets of pause experiments, two different answers
In March 2012, Google's researchers published a meta-analysis of 390 experiments in which advertisers had paused their search ads. The question was how many paid clicks the advertiser's own organic listing picked up once the ad was gone. When the site held the top organic position, half of the ad clicks were incremental: they vanished instead of moving to the free result. When the organic result sat between positions 2 and 4, 82% of ad clicks were incremental. Below position 4, 96%.
Three years later, economists who ran large field experiments at eBay published results in Econometrica that pointed the other way. Their abstract is blunt: "brand-keyword ads have no measurable short-term benefits." On non-brand terms, the ads moved new and infrequent users, but frequent users, who would have bought anyway, accounted for most of the spend, "resulting in average returns that are negative."
Both findings can be right. Google measured clicks across hundreds of advertisers, averaged. The eBay team measured sales at one enormous, famous retailer whose customers type its name to get there. Neither tells you what happens when a Toronto law firm stops bidding on "severance package lawyer". That takes a test on your own searches, and the test is easy to run badly. This piece covers how to find the overlap between paid and organic, what that overlap can and cannot claim, and how to design a pause that a small account can actually read.
Finding the overlap means joining two exports that were not built to meet
The inputs are the Google Ads search terms report and the Search Console Performance export's Queries table, for the same dates. Google has its own version of this join: the paid and organic report in Google Ads, which needs a Search Console property linked to the Ads account and, per Google's documentation, counts text ads only. Doing the join yourself from exports works without the link and lets you see every step.
Three steps decide whether the join is honest. First, normalize both sides the same way: lowercase, accents folded, punctuation turned into spaces. Do not stem. "Severance package lawyers" and "severance package lawyer" stay separate, which undercounts the overlap a little but never invents it.
Second, group and sum each side before joining. Normalization makes some distinct rows identical ("harbour & finch" and "harbour finch" become the same text), and joining before grouping lets one row match several and multiplies its numbers. When rows merge, average their positions weighted by impressions, not as a plain mean. Search Console's position is already the average position of the site's topmost result, so a weighted average of averages is the only arithmetic that keeps its meaning.
Third, respect what Search Console leaves out. Google's own explainer says the interface exports at most 1,000 rows, and that anonymized queries, "those that aren't issued by more than a few dozen users over a two-to-three month period", are omitted from the tables. A search term with no match in the export was not observed. That is different from not ranking. The Search Analytics API goes up to 50,000 rows per day per site per search type, which closes most of the gap for sites that hit the interface cap.
Five kinds of search term, decided in a fixed order
Once joined, every search term goes into exactly one of five classes. The order of the checks matters as much as the definitions.
| Class | Test | What it can claim |
|---|---|---|
| No spend | Shown, never clicked, cost nothing | Nothing was bought |
| Paying for what the site ranks for | Has spend, and Search Console records an organic position | The cost and the position; a test to run |
| Renting demand the site does not rank for | Has spend and conversions, no organic position on record | Proven demand; a page worth considering |
| Off topic | Spend, no conversions, and none of its content words appears in any query the site ranks for | A candidate negative |
| Unknown | Spent, and cannot be placed from the data | Nothing yet |
Organic presence is checked before conversions, so a converting term that the site also ranks for counts as owned, not rented. That keeps the renting class to searches where the site genuinely has nothing. The off-topic test is deliberately strict: a term with any content word in common with the site's organic queries is not called off topic, and a term that is on topic, non-converting and missing from a capped export falls into unknown instead of being quietly labelled waste.
In the sample account, most of the overlap is not worth testing
Harbour & Finch is a fictional Toronto employment and injury law firm we use as a sample client. Its four-week exports (August 25 to September 21, 2026) contain 28 paid search terms and CA$13,627.47 of spend. Run through the five classes, they split like this:
- Paying for what the site ranks for: 15 terms, CA$9,615.80.
- Renting demand the site does not rank for: 2 terms, CA$1,356.30.
- Unknown: 11 terms, CA$2,655.37.
- Off topic and no spend: none.
The headline, CA$9,615.80 on searches the firm already ranks for, is where most overlap reports stop. It overstates the opportunity badly. Of that figure, CA$4,006.30 went on four terms where the firm's average organic position is worse than 10: "car accident lawyer toronto" at 21.5, "slip and fall lawyer toronto" at 17.2, "workplace harassment lawyer" at 14.8 and "constructive dismissal lawyer" at 11.3. Google's own figures put ad incrementality at 96% below position 4. Pausing those ads would mostly just lose the clicks.
Filter to terms where the firm averages position 3.5 or better, which rounds to the top three, and four remain, with CA$2,093.00 of spend. Two are brand searches, "harbour and finch" (CA$96.00, position 1.1) and "harbour finch llp reviews" (CA$21.60, 1.4). The other two carry nearly all the money: "wrongful dismissal lawyer toronto" (CA$1,012.40, 6 conversions, position 3.1) and "severance package lawyer" (CA$963.00, 4 conversions, position 2.4). Those two searches are the whole testable question in this account.
The unknown pile, meanwhile, holds every non-converting term in the account. None reaches off topic, because each shares a word such as "lawyer", "employment" or "ontario" with something the site ranks for. That is the strict test working as intended: word overlap cannot tell a job seeker from a client, so those terms need a person reading them, which is the subject of the companion piece on negative keywords.
Why the overlap is a list of tests, not a saving
Pausing an ad does not hand its clicks to the organic listing. What sits between the two is the results page and the searcher. If a rival's ad takes the slot yours vacated, if an AI Overview or a map pack pushes the organic result down, or if the ad's copy answered something the page title does not, clicks leak to someone else. One thing does not change: Google Ads Help states that "investment in paid search has no impact on your organic search ranking." The ad neither props up the ranking nor holds it back.
It is worth doing the arithmetic on "severance package lawyer" to see the size of the question. In the sample four weeks it drew 41 paid clicks and 58 organic ones, 99 in all. If Google's average for positions 2 to 4 held here, 82% of paid clicks incremental, a pause would lose about 34 of the 41, organic would pick up about 7, and the total would fall to roughly 65. If this term behaved like eBay's brand ads, the total would stay near 99 and the firm would keep CA$963 every four weeks. Those are illustrations built on other people's averages, not forecasts. The gap between them is the reason to test.
A note on the Google numbers before leaning on them. They measure clicks, not conversions or revenue, and the authors themselves say results for individual advertisers vary and recommend randomized experiments. They come from the company that sells the clicks, and they predate results pages with AI answers at the top. They are a sensible prior. They do not replace a test.
A pause test a small account can actually read
Most pause tests fail on design, not on analysis. These are the choices that decide whether yours can answer.
Block the search in the right place
Pausing the keyword is the wrong tool, because the search can match other keywords. Add an exact match negative for the search term instead. Put it at the campaign level, or in a shared negative list attached to every campaign that could match the search. An ad group negative alone often just moves the search to another ad group with a broad keyword. Negatives do not match close variants, so "severance package lawyers" can still show the ad. Either add the variants you see, or accept the leak and count it. Check the search terms report after a few days to confirm the term's impressions really dropped.
Measure the total, from both sources
The outcome is paid clicks plus organic clicks on the tested searches, and conversions from both channels. Ads gives the paid side; Search Console, filtered to the exact queries by day, gives the organic side. Measuring only paid spend makes every pause look like a win.
Give it a control and a rhythm
Keep comparable terms running as a control: in the sample, "severance pay ontario" (position 4.2) and "severance calculator ontario" (3.6) share the topic and the season. If their totals rise during the pause too, part of any rise on the tested terms is not the test. Better still, alternate: two weeks on, two weeks off, two weeks on, compared on the same weekdays. One before-and-after comparison cannot separate the pause from everything else that changed that month.
Be honest about what the volume can show
Run the sample numbers forward. Over two weeks, "severance package lawyer" would be expected to draw about 50 total clicks with the ad on. If Google's average held, that would fall to about 33 with it off. Random week-to-week variation on a count of 50 is around 7, so a gap of 17 is visible, roughly, in clicks. Conversions are another matter: at four a month, a two-week pause expects about two with or without the ad. No design reads that. Test both top-three terms together, run more on and off cycles, and state plainly that the conversion side is underpowered. Google counts conversions against the date of the click, so wait a couple of weeks after the last period before reading conversions at all.
Reading the result, and the brand exception
If total clicks and conversions held while paid spend fell, leave the negatives in place and check again next quarter, because the results page changes. If organic clicks rose but the total fell, the ad was catching searchers the organic listing does not, and the better lever may be a lower bid rather than no ad. If the test cannot separate the outcomes, say so in the report and run longer. "Not yet readable" is a result.
Brand terms are the case people argue about most, and the sample shows why they are often the wrong place to start. "harbour and finch" cost CA$96.00 in four weeks for 64 clicks and 9 conversions, at an organic position of 1.1. eBay's result suggests that ad may add little. But the real variable is who else is on that page. If a competitor, say the sample's rival Lakeshore Legal, bid on the firm's name, pausing hands them the top of the results. At under CA$100 every four weeks there is little to recover and something to lose. The money in this account is in the two non-brand terms.
The renting class deserves its own look. "personal injury lawyer free consultation" (CA$816.30) and "lawyer near me" (CA$540.00) each converted once with no organic presence. One conversion each is thin evidence. It is still paid-for proof that the demand exists, and the first is a reasonable brief for a page. The second is a local-results question more than a content one.
The free Paid vs organic tool does the join and the five-way split on your own exports in the browser, and shows the top-three subset separately. In AIO Copilot it runs for every connected client, and a pause test becomes a planned change that is measured when it ends.
Start with the list sorted by spend, cut it to top-three positions, and drop the brand terms. What is left, usually a handful of searches, is the test worth setting up this week.
Frequently Asked Questions
If I rank first organically, should I stop bidding on that keyword?
Not on the ranking alone. Google's 2012 meta-analysis of 390 pause studies found that when the advertiser ranked first organically, 50% of ad clicks were incremental on average, while eBay's experiments found its brand ads added nothing measurable. The spread is wide enough that only a pause test on your own terms tells you which side you are on.
Does bidding on a keyword help its organic ranking?
No. Google Ads Help states that investment in paid search has no impact on organic search ranking. The question is only whether the ad adds clicks and conversions the organic listing would not have delivered.
How long should a paid vs organic pause test run?
Long enough to see a difference in total clicks, which for a small account usually means several weeks, ideally alternating periods with the ad on and off. Conversions on a single term are often too few to read at all, so say so rather than deciding on two or three conversions.
How do I pause an ad for one search without pausing the keyword?
Add an exact match negative for that search, at the campaign level or in a shared list attached to every campaign that could match it. An ad group negative alone can let the search match a different ad group, and because negatives do not match close variants, plurals and misspellings may still show the ad.
Why do some search terms show no organic data?
The Search Console interface exports at most 1,000 rows, and queries searched by too few people are anonymized and left out of the tables. A query missing from the export was not observed, which is not the same as not ranking. The Search Console API returns up to 50,000 rows per day per site per search type.
Never miss an update
Get the latest AI and SEO strategies delivered to your inbox.