How do you test two placement lists in Google Ads?
Part of Google Ads placements
The short answer
To test placement lists in Google Ads, build a custom experiment from the live Display campaign, change only the placement list in the trial arm, and split traffic and budget with the original. Display experiments always use cookie splits, so each user sees one arm. Run two to twelve weeks, read the confidence interval, then apply or convert the winner.
Key takeaways
- A custom experiment shares the original campaign's traffic and budget, and in Display it always uses cookie splits so a user sees only one arm.
- Google says a 50/50 split controls auction eligibility, not equal impressions or spend, and suggests allowing 7 to 14 days for the treatment arm to stabilize.
- The initial end date must fall 2 to 12 weeks after the start; Google recommends 2 to 3 weeks before judging a result and lets an experiment be extended.
- The scorecard reports an estimated difference with a confidence interval, 80 percent by default, and a blue asterisk when the result is statistically significant.
How do you test placement lists in Google Ads?
Google's tool for this is the custom experiment. Its About custom experiments page (opens in a new tab) describes proposing a change to a Search, Display, Demand Gen or Video campaign, running it beside the original, and measuring the result before applying it. The setup page (opens in a new tab) says the experiment shares the original campaign's traffic and budget. For a placement test, the original arm keeps list A and the trial arm swaps in list B, with nothing else changed.
Definition
- Custom experiment
- A Google Ads test in which a trial copy of one or more original campaigns runs with a proposed change, sharing their traffic and budget, so the two arms can be compared over the same dates.
Six steps to set up a placement list experiment
- Pick an active base campaign. Google's eligibility rules ask for an active campaign with an individual daily budget, not a shared budget, and a standard bid strategy, not portfolio bidding.
- Clean the ad groups first. Remove topics and Display keywords from the placement ad groups in both arms, since Google's placement page (opens in a new tab) says multiple content targeting methods reach any of the selected content.
- Create the experiment. Go to Experiments within the Campaigns menu, select the plus button, choose Custom, and pick the variable to test.
- Change only the list. In the trial campaign, replace list A's placements with list B's in the same ad groups, with the same bids, creative, exclusions and suitability settings.
- Set the split and dates. Choose the share of traffic and budget for the trial arm, then a start date late enough for ads to finish review, which Google says can take some time.
- Decide on auto-apply. Google's monitoring page (opens in a new tab) says auto-apply is enabled by default for some experiment types; turn it off if a person must sign off on the list change.
How should traffic be split in an experiment?
A placement test has to keep the two lists from reaching the same person, or each arm's results absorb the other's ads. Google handles that for Display: the setup page says "In Display campaigns, we always use cookie splits to ensure that users only ever view the experiment or your original campaign." Search experiments can choose between cookie and search based splits; Display cannot.
The setup page says the buyer decides what percentage of the original budget goes to the experiment, and a note on the same page says the experiment arm splits the budget evenly between the control and treatment arms. An even split gives each list the same chance to collect data, which shortens the test. A smaller trial share protects a campaign the client depends on, at the cost of a longer test. Google's monitoring page lists a traffic split that was too small among the reasons a result may not be statistically significant.
| Cause Google names | What it means for two placement lists |
|---|---|
| Each ad slot is a new auction | Two lists with different sites can win different shares of auctions at the same bids |
| Different bids or assets win more or fewer auctions | Keep placement bids identical across arms so the list is the only difference |
| One side hits its daily budget first | The arm that runs out stops showing ads for the day, which tilts delivery by hour |
| Automated bidding shifts volume while it learns | Leave the bid strategy alone during the stabilizing period |
Google's About page says a 50/50 traffic split controls auction eligibility but doesn't guarantee equal impressions or spend. Unequal delivery alone is not a broken test: a list that cannot spend its share is telling the planner something about its inventory.
How long should a placement test run?
Google sets the outer limits. Its find and edit page (opens in a new tab) says the initial end date must be between 2 and 12 weeks after the start date, for custom Search, Display and Hotel experiments, and that the end date can be extended during the experiment up to 12 weeks from the current date. The About page says to allow 7 to 14 days for the treatment arm to stabilize, and the monitoring page recommends 2 to 3 weeks of data before calling a winner, with more budget or more time if the result is still undecided.
Two Google warnings belong in the test plan. Changes made to the original campaign during the test are not copied to the experiment, and editing either arm while it runs "may make it harder to interpret your results." The find and edit page adds that shifting the bid type or rules mid test can start a learning phase. Write the freeze into the plan, with the one person allowed to break it.
What results can a placement test support?
The monitoring page explains the scorecard. For each metric it shows the experiment's own number, the estimated difference from the original, and a range for that difference at the chosen confidence level, 80 percent by default. A blue asterisk marks a statistically significant result. The Results column names the control campaign, the treatment campaign, or no clear winner.
| What the scorecard shows | What the planner can write | What it does not show |
|---|---|---|
| Estimated difference, such as a percentage change in a metric | List B's campaign differed from list A's by about this much over these dates | Which sites on list B caused the difference |
| Confidence interval | The likely range of that difference at the chosen level | That the same range holds in another market or season |
| Blue asterisk | The data is likely not due to chance, in Google's words | That the difference is large enough to matter for the client |
| No clear winner | The test could not separate the lists at this size and length | That the lists perform the same |
What does a placement list test look like on a real brief?
Fictional example
A closet design company in St. Louis and Springfield
An unnamed installer of custom closets and home organization systems in St. Louis and Springfield, Missouri, runs a Display campaign in English for free design consultations. List A, live since spring, holds home improvement and real estate sites. The planner builds list B from home organization, interior design and local lifestyle publishers and wants evidence before switching.
The original campaign keeps list A. A custom experiment with an even split gives list B to the trial arm, with identical bids, creative, exclusions and content suitability settings and no topics in either arm. The test runs six weeks from a start date set a few days out, auto-apply is turned off, and the plan names the account lead as the only person allowed to edit either arm.
The success metric is consultation requests. At the end, the planner reports the estimated difference, its interval and whether the asterisk appeared, and states that the result covers these two Missouri markets over these six weeks. No performance figures are implied.
How do you apply a winning list?
Google's apply page (opens in a new tab) offers two routes, and both keep the performance data of the original campaign and the experiment.
| Route | What happens | Fits when |
|---|---|---|
| Update your original campaign | The experiment ends and its changes are applied to the original campaign | The client wants one campaign with continuous history |
| Convert to a new campaign | The original campaign is paused; the experiment runs as a regular campaign with the same budget and dates | The team wants list A's campaign kept intact and paused, ready to compare or restore |
Before applying, update the approved list record so the list in the account and the list the client signed match, with a version number and the test's dates. If the account is due for Google's Display to Demand Gen migration, which began a phased rollout in June 2026 and cannot be reverted for migrated campaigns, finish the test first: Google's migration page advises against using the migration tool during a running lift study, and its help pages checked did not address custom experiments.
How can PlacementGraph supply the second list for a test?
Frequently asked questions
Can a placement experiment run in a Demand Gen campaign?
Google's pages differ. The About custom experiments page lists Demand Gen among campaigns that support custom experiments, while the setup page lists Search, Display, Video and Hotel Ads. Check whether the campaign appears in the experiment's campaign selection list, as both pages were read on October 22, 2026.
Do audience lists change how a cookie split works?
Google's setup page says that with a cookie based split and audience lists, the list should hold at least 10,000 users for a custom experiment, and that smaller lists may give less accurate results. The note is about audience lists, so a placement only test with no audience lists sits outside it.
What happens to the experiment if the original campaign is paused?
Google's setup page says experiments rely on the base campaign. If it is paused, inactive or ends before the experiment's scheduled end, the experiment will not run or serve impressions. Keep the original live and its end date beyond the test.
How many experiments can run on one account at once?
Google's setup page says you can schedule up to 5 experiments, each covering multiple campaigns, but only one experiment runs at a time. The apply page says a custom experiment supports up to 100 campaigns. Plan placement tests in sequence rather than in parallel.
Sources
- About custom experiments (opens in a new tab)Google Ads Help
- Set up a custom experiment (opens in a new tab)Google Ads Help
- Monitor your experiments (opens in a new tab)Google Ads Help
- Find and edit your experiments (opens in a new tab)Google Ads Help
- Apply a custom experiment (opens in a new tab)Google Ads Help
- About placement targeting (opens in a new tab)Google Ads Help
- About the Where ads showed report (opens in a new tab)Google Ads Help
- Google Display Ads campaigns have a new home in Demand Gen (opens in a new tab)Google Ads Help
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.