Every Demand Gen report answers the same question: how many conversions did this campaign get credit for? None of them answers the one you are paying to find out: how many customers would you not have had without it?
Those are different numbers, and the gap between them is the whole argument about whether Demand Gen is worth the money. Google knows this. It has built tests specifically to measure that gap, and it has spent 2026 promoting them.
Here is the problem for a small business. Read the requirements closely and most of these tests are not built for an account your size. And the workaround Google offers when your numbers are too small quietly undoes the reason for running the test.
The Question Every Demand Gen Report Avoids
The number in the Conversions column is an attribution count. It says your ad was in the path, somewhere, before someone converted. It does not say the ad caused anything.
We have covered both ways that count goes wrong. In Demand Gen Will Tell You It Worked. That Is Not the Same as Working., one audit found the campaign claiming customers who were already on their way, and another found it missing revenue it genuinely drove. And since new Demand Gen campaigns now count people who never clicked by default, which we explained in Demand Gen Now Bids on People Who Never Clicked. Google Turned It On for You., the attribution count is looser than it has ever been.
The number that matters is called incremental: the sales or leads that happened because of the campaign and would not have happened without it. The only way to measure it is to compare people or places that saw your ads against comparable ones that did not. Everything else is inference.
Google Has Three Tests. Only Two Answer That Question.
Google offers three ways to test Demand Gen from inside Google Ads, and they are easy to confuse because they all live under the Experiments menu.
Demand Gen A/B experiments compare one Demand Gen campaign against another. You can test creative, audiences, bidding or a product feed, one variable at a time, and Google tells you which version performed better. This is genuinely useful for improving a campaign you have already decided to run. It cannot tell you whether running Demand Gen at all was worth it, because both sides of the test are Demand Gen.
Uplift Experiments were announced at Google Marketing Live in May 2026. Google describes them as an automated A/B test that measures the impact of adding Demand Gen to your existing mix of comparable campaigns, such as Performance Max. That is the right question. Google's announcement cites an average of 10 percent higher return on ad spend when Demand Gen is added to Search and Performance Max. Read the footnote before quoting that anywhere: it comes from a marketing mix study Google commissioned from Nielsen, covering 1.1 million campaigns in categories like food and beverage, retail, apparel, automotive and telecommunications. That is a study of large consumer brands, not local service businesses.
Conversion Lift holds a share of your audience back from seeing your ads, either a random group of users or a set of geographic regions, and measures the difference in conversions between the two groups. This is the most rigorous option Google offers, and it is the one Google's own help pages treat as the standard for incrementality.
The Entry Requirements, in Google's Own Words
This is where the tests stop being available to most of the businesses reading this.
Demand Gen A/B experiments. Google's help page says that when you use conversion-based bidding, the experiment needs a minimum of 50 conversions per arm before it shows results, and that conversion metrics need at least 100 data points before results start to appear. The default confidence level on the report is 70 percent, which Google itself labels directional. The 95 percent level, which Google calls conclusive, takes more data still. A service business producing 20 leads a month from Demand Gen, split evenly between two arms, would need about five months for each arm to reach 50, and it would be answering a question about creative, not about whether Demand Gen works.
Conversion Lift. Google's help page opens with a sentence that settles the matter for most small advertisers: Conversion Lift is not available for all Google Ads accounts, and to use it you contact your Google account representative. Many small accounts have no dedicated representative. For the accounts that can get in, the page lists an account threshold of at least 1,000 observed conversions and a minimum campaign budget of $5,000 for the study. It recommends a study of at least 14 days, reports that studies with long conversion lag that run shorter than that have shown drops of up to 17 percent in measured lift, and notes that most advertisers run about one or two studies a year, because every study withholds your ads from part of your market. The geographic version of Conversion Lift carries the same note about contacting your representative.
Uplift Experiments. Google has not published volume requirements for these. What it has published is the framing: they measure Demand Gen as a complement to campaigns like Performance Max that you are already running at scale. An account with one Search campaign and a modest Demand Gen test is not the account this was designed around.
Google's Fix Is to Count Something Easier
Google does offer a way around the volume problem. It is worth reading carefully, because it is the most revealing sentence in the documentation.
On the A/B experiments page, Google says that to reach 50 conversions per arm, it recommends Target CPA or Maximize conversions bidding optimizing toward shallow conversions like Add to Cart or Page view. On the Conversion Lift page, it says upper-funnel and mid-funnel actions such as page views or lead forms show higher lift and increase your chances of measuring clear results.
In other words: if your real outcomes are too rare to measure, measure something more common.
For some businesses that is a reasonable compromise. For a small lead generation business, it hollows out the test. The reason to run an incrementality test on Demand Gen is that Demand Gen is good at producing activity that looks like results. A page view is exactly that kind of activity, and a campaign that raises page views has shown you something you already expected.
Google's own lift page does say to measure bottom-funnel actions alongside the easier ones, and to treat the upper-funnel actions as secondary. That is the right instinct. The trouble is that a small account's bottom-funnel numbers are precisely the ones too thin to reach a conclusion, which is how you end up reading the secondary number as if it were the answer.
What a Small Account Can Do Instead
You can run a version of the geographic test yourself, without a representative, a minimum budget or anyone's permission. It is less precise than Google's tools. It is also far more honest than the Conversions column, and it measures the thing you care about: total leads and customers, from every source, in places where Demand Gen ran versus places where it did not.
This is a different test from switching a campaign off and watching what happens, which we walked through in You Will Never Know Exactly Which Ad Made the Sale. That approach compares one period with another. A geographic split runs both sides at the same time, so the weather, the season and the local economy hit both areas equally, and it tests adding Demand Gen rather than removing something you already rely on.
Pick two comparable areas. Two towns, two groups of ZIP codes, or two parts of your service area that normally produce similar business. They should not sit right next to each other, because people who live in one and work in the other will see ads meant for the other side.
Measure their normal relationship first. Look back over the previous two or three months and count total leads from each area, from every source: calls, forms, walk-ins, referrals. Areas are rarely equal, so what you need is the ratio. If one area usually produces 25 percent more leads than the other, that is your baseline.
Run Demand Gen in one area only. Set the campaign's location targeting to the test area, with presence rather than interest, and change nothing else. Search, your other channels and your pricing carry on exactly as before in both areas.
Give it long enough. Google's own guidance for lift studies is to cover your conversion lag and run at least 14 days. For a considered purchase, six to eight weeks is more realistic. Decide the end date before you start.
Compare against the baseline, not against zero. At the end, work out what the test area would have produced without Demand Gen, using the baseline ratio, and compare it with what actually happened.
Here is how the arithmetic works, using illustrative numbers rather than a real client. Over the baseline period, the test area produced 100 leads to the comparison area's 80, a ratio of 1.25. During an eight-week test with Demand Gen running only in the test area, the comparison area produces 100 leads and the test area produces 150. Without Demand Gen, you would have expected the test area to produce about 125, which is the comparison area's 100 multiplied by 1.25. So roughly 25 leads were incremental. If the campaign spent $3,000, each incremental lead cost about $120.
Now compare that with what Google Ads reported for the same eight weeks. If the Conversions column claims 70, the campaign is taking credit for nearly three times the leads it can be shown to have added. That ratio is the most useful number you will get about Demand Gen, and no report in the platform will show it to you.
Three honest limits. Small numbers are noisy, so a difference of a handful of leads proves nothing, and the test is most useful when the gap is large or when you repeat it with the areas swapped. The areas are never perfectly isolated. And a single test tells you what happened in one market over one stretch, not a law of nature. Treat the result as a far better estimate than you had, not a certainty.
When Google's Own Tests Are Worth Asking For
There is a point at which the built-in tools become the better choice. If your account clears 1,000 observed conversions, can put $5,000 behind a study, and has a Google representative, a proper Conversion Lift study will be more precise than anything you run by hand, and the April 2026 addition that projects conversions arriving after a Demand Gen study ends makes it fairer to slower-converting businesses.
If you run Performance Max at meaningful scale and are deciding whether to add Demand Gen, ask about Uplift Experiments. And once you have decided Demand Gen stays, A/B experiments are the right tool for making it better.
If Demand Gen is claiming customers who already knew you, the fix sits in the campaign settings rather than the testing tools, and we covered it in Demand Gen's New Prospects Mode Stops You Paying for Customers You Already Had.
Count the Customers It Added, Not the Credit It Claimed
Google has built serious measurement tools and, to its credit, documented their limits plainly. The trouble is only that the limits sit exactly where most small businesses live. Below the thresholds, the official advice is to measure something easier, and something easier is how Demand Gen ends up looking better than it is.
The alternative is not complicated. Pick two areas, learn how they normally compare, run the campaign in one, and count real leads in both.
If you would like help designing that test, choosing the areas and reading the result against your own records, that is part of how we run Google Ads management for small businesses. Either way, do not let a campaign's credit stand in for its results.
