How do I test AI Max on my Google Search campaign before turning it on for good?

TofuAnswer by the Tofu team · checked

A fingertip touching the surface of turquoise water, sending out small ripples

Run an AI Max experiment from Campaigns, then Experiments, in Google Ads. It splits one Search campaign's traffic and budget between an arm with AI Max off and an arm with it on, without copying the campaign. By default the AI Max arm gets search term matching, text customization and final URL expansion, and you can switch the last two off at setup. If you don't apply the result, the campaign goes back to its old settings. For B2B, don't let it auto-apply, and judge it on qualified leads and pipeline, not Google's conversion count.

Ask Tofu next

Pick a question and Tofu Ads Agent answers it in chat.

As of November 2026.

How does an AI Max experiment work?

It tests AI Max inside the campaign you already run. Google says it "diverts traffic and budget within an existing campaign instead of creating a copy of the campaign to experiment on." One share of the campaign runs with AI Max off (the control) and the rest runs with it on (the trial). Google says keeping the traffic in one campaign likely gives faster results and a shorter learning period than a custom experiment, which copies the campaign.

How do I set one up?

Use the Experiments page when you want control over the settings:

  1. In Google Ads, go to Experiments in the Campaigns menu.
  2. Select the plus button, then Campaign features and settings.
  3. Select AI Max for Search campaigns and click Continue.
  4. Click Select campaigns, tick the campaign and select Done.
  5. On the "Create an AI Max experiment" page, set the arms, dates, name and auto-apply setting. If you picked one campaign, you also get the Asset optimization and brand settings.
  6. Select Create. Google says this test starts on its own.

There is also a shortcut. Google's AI Max FAQ says eligible accounts may see a 1-click experiment recommendation on the Recommendations page, and Create 1-click experiment automatically sets up a 50/50 split between your campaign and an AI Max arm. Search Engine Roundtable reported it on October 26, 2026.

Read any recommendation card before you click. Google's experiments article also lists a card called "Turn on AI Max for Search campaigns" with an Apply to all button, which acts on every campaign in the list.

What does the AI Max arm turn on, and what can I change?

All three AI Max features are on by default. What you can change, and when:

SettingIn the AI Max arm by defaultCan you change it?
Search term matchingOnPer ad group, in the AI Max arm only (see the note below)
Text customizationOnYes, at setup, under Asset optimization
Final URL expansionOnYes, at setup. It needs text customization on
Brand inclusions and exclusionsOnly if you add themYes, at setup. They apply to both arms
URL inclusionsNone unless you add themPer ad group, after setup, in the AI Max arm only

Google's two help pages disagree on timing. Its experiments article says ad group changes such as turning off search term matching "aren't supported during the initial experiment setup" and you edit them afterwards. Its AI Max FAQ says you can set them per ad group "during experiment construction." Plan to check your ad groups right after you create the test.

If you only want to learn whether wider matching finds good searches, untick text customization and final URL expansion. Then the test measures search term matching alone, and your ads and landing pages stay as written.

Why can't I select my campaign?

Google lists six blockers. You can't create an AI Max experiment if the campaign:

  • has text customization (formerly automatically created assets) on
  • targets the Display Network
  • is part of a portfolio bid strategy
  • uses a shared budget
  • uses bidding exploration
  • has any active experiment

Google also says to turn off the "Optimize your campaign with AI" setting if the campaign won't show up. Some older campaigns carry legacy settings, such as a brand list exclusion on a non-brand campaign, that block the test. Google's fix is to build a new campaign without them and test that. If Google's 2026 upgrade already turned AI Max on in your campaign, see what the AI Max auto-upgrade changed.

Should I use an AI Max experiment or a custom experiment?

Use the AI Max experiment unless your campaign also targets the Display Network. Google compares them like this:

AI Max experimentCustom experiment
How it runsInside your campaignA copied trial campaign next to the original
StartStarts on its ownYou start it by hand
Campaigns that target the Display NetworkNot allowedAllowed
Where you see reportsExperiments, Keywords and Ads pagesThose plus the Campaigns page and the report editor
Ramp-upShorterThe trial campaign may need longer

Google says the other five blockers above apply to custom AI Max experiments too.

How long should an AI Max experiment run?

Google gives no fixed length for this test. Its experiments guide says that when there is no clear winner, it recommends "allowing your experiment to run for 2 to 3 weeks," and adding budget or time if it is still undecided. Its custom experiments article says to allow 7 to 14 days for the trial arm to settle. In B2B, also leave room for the time it takes a form fill to become a qualified lead in your CRM, so the leads you judge on have had time to be scored.

How should a B2B account judge the result?

On qualified leads per dollar, not on the scorecard's default view. Google's scorecard shows clicks, CTR, cost, impressions and All conversions by default. "All conversions" includes actions you left out of your main Conversions column, so switch the metrics to Conversions and Cost / conv.

The scorecard shows the difference between the arms with a confidence interval (80% by default, and you can change it) and a blue asterisk when a result is statistically significant. Google's custom experiments article says reliable results need base campaigns that meet minimums "such as more than 100 daily conversions." Few B2B Search campaigns get close, so expect wide ranges and no asterisk on a lead-gen test.

What helps:

  1. Count the lead that matters. Import qualified-lead or opportunity stages from your CRM as conversions, so the experiment can count them per arm. See how to upload offline conversions to Google Ads.
  2. Read the new searches. In the AI Max arm, check the search terms AI Max added and block the job, student and free-tool searches. The auto-upgrade page shows where to find them.
  3. Decide on cost per qualified lead. More conversions at a similar cost can still mean more junk leads if your conversion is a raw form fill. Our match type guide covers judging wider matching on pipeline.

What happens when the experiment ends?

You choose whether to keep AI Max. You can apply it from the Experiments table, or turn AI Max on in the campaign settings. At setup there is also an option to "Apply your experiment changes if results are favorable," which applies it on its own when the test ends. Google's article doesn't say what counts as "favorable," so for B2B leave that off and decide on qualified leads yourself.

If you don't apply it, Google says all AI Max settings go back to how they were before the test. Brand controls you added for the test are removed, and any older brand controls come back.

Running Microsoft Advertising too? Its version of AI Max is covered in what happened to predictive matching in Microsoft Ads.

Sources (7)

Related resources