All answers

The Prevailing
Answers for
Business model:
Business model, pick up to two

How do we set up and read a Google Ads experiment (traffic split, ramp-up, length, metric) to test a bid strategy, Performance Max or AI Max change against our current setup?

The PrevailingUpdated Nov 20, 2026

Our answer

Test the change in Google's own experiment for it, split 50/50 with nothing else changed. Leave out one to two weeks of ramp-up, run at least 4 to 6 weeks and judge it on a metric you chose, not auto-apply.

Confidence
Medium
Strongest proof
1 controlled test
12 Google rules, 3 practitioners, 9 opinions or claims
Sources
7 sources
25 statements

New here? Take the 1-minute tour

Why this is the answer

3 reasons and our judgment

Google builds a separate experiment for each kind of change. It discards the first 7 days for ramp-up and recommends at least 4 to 6 weeks. For value bidding it suggests a 50/50 split after 2 weeks of ramp-up.

Tests often need longer than the minimum. HubSpot's paid ads team ran AI Max as 50/50 cookie-based Google Ads experiments for eight to 12 weeks on average. Savvy Revenue's eight-week Performance Max test swung back and forth and never reached the 95% confidence it wanted.

Google's defaults read results loosely. Auto-apply is on by default and can pick a winner on directional results. Google's product leads call the scorecard's default 80% confidence enough for a directional tweak and save 95% for a major structural change.

Our judgment

  • Use the experiment Google built for the change and split it 50/50. That means a custom experiment for a Search bid strategy, the one-click Target ROAS experiment for Standard Shopping on manual CPC, the Shopping against Performance Max experiment for a move to Performance Max and the AI Max experiment for AI Max. Google routes each change to its own flow and suggests 50/50 for value bidding, and HubSpot and Amsive ran their AI Max tests 50/50. On Search, choose the cookie-based split, which HubSpot uses so each user sees one version and Relevant Audience advises when bidding works toward a conversion days later.
  • Leave one to two weeks of ramp-up out of the read, then run at least 4 to 6 weeks, longer when orders lag. We combined Google's FAQ, which also gives a 7 to 14 day ramp-up and 1 to 2 conversion cycles, with its value bidding page, which asks for at least 30 days after ramp-up. HubSpot's eight to 12 weeks and Savvy Revenue's eight weeks without a significant result show a test can need longer. The same setup applies at every spend level, because Google sets its rules in conversions and weeks rather than spend. Its only volume floors are 50 conversions in 30 days for value bidding and 15 for the one-click Target ROAS experiment.
  • Choose the success metric before launch and don't let auto-apply decide. Make it the result the change should move, conversion value or ROAS for value bidding and profit or total sales where you can measure them. Savvy Revenue judged its test on cumulative profit and HubSpot on whether AI Max boosts revenue. Set auto-apply to require statistical significance or turn it off, because it is on by default and can apply on directional results.
  • Hold a big change to 95% and don't take a winner on Google's read alone. Treat a Search bid strategy switch or a move from Shopping into Performance Max as the kind of structural change Google reserves 95% for, and act on 80% only for a directional tweak on the tested campaign that you keep watching. Savvy Revenue held its Performance Max test to 95%, and Google's own reasons a result isn't significant include too short a run and too small a split. Check a kept Performance Max or AI Max change that can pull brand or returning-customer sales into the campaign against total store sales, because Google's experiments count Google-reported conversions (Testing whether Google spend is incremental).
  • Change nothing else in either arm while the test runs. Run one experiment at a time, and keep a record of each test's hypothesis, dates and result.

Our judgment, built from Google, HubSpot (paid ads team), Amsive, Relevant Audience and Savvy Revenue: our reading of these sources, not a finding from any one of them.

What the sources say

25 statements from 7 sources
  • GoogleNov 19, 2026Supports

    Google's Ads Decoded summary says to use Experiments templates that split traffic 50/50, judge on qualified conversions over four to six weeks, and accept 80% confidence for small tweaks but 95% for major changes.

    “An 80% confidence interval works well for directional tweaks. Reserve 95% certainty for major structural overhauls.”

    Read it on blog.google

  • GoogleNov 18, 2026Caveat

    On Google's Ads Decoded podcast, Al Nevarez says an 80% confidence level is usually good enough for directional optimization when combined with the advertiser's own judgment.

    “I think we feel that the 80% is typically good enough when, especially when combined with the advertiser's judgment.”

    Read it on goo.gle

  • GoogleNov 18, 2026Supports

    Google's Sagar Shah recommends running Google Ads experiments at least four to six weeks to gather enough conversions for a significant result.

    “our standard recommendation, we say, is at least let it run four to six weeks.”

    Read it on goo.gle

  • GoogleNov 18, 2026Supports

    On Google's Ads Decoded podcast, Sagar Shah prefers templated experiments to custom ones: they split traffic 50/50 within the same campaign, isolate one variable and may shorten the learning period.

    “this approach may also lead to a reduced learning period because you're acting on that same campaign that you started with.”

    Read it on goo.gle

  • GoogleNov 18, 2026Supports

    Google's Sagar Shah says the sync feature for custom experiments must be turned on when the experiment starts, while both arms are matched, and not switched on partway through.

    “We can only set it up at the very beginning of running an experiment because that's when both the arms are matched”

    Read it on goo.gle

  • GoogleNov 18, 2026Supports

    Google's Al Nevarez says budget experiments compare fairly by giving the control arm half the original budget and the treatment arm half the new budget.

    “during the experiment, we'll use half of the original budget in the control and half of the new treatment budget in the treatment arm.”

    Read it on goo.gle

  • GoogleNov 18, 2026Caveat

    On Google's Ads Decoded podcast, Google says its Ads experiments use jackknife resampling to calculate confidence intervals and significance, which it says reduces bias and guards against outliers.

    “statistical model that we use called Jackknife resampling to calculate confidence into intervals and statistical significance in our AB experiments.”

    Read it on goo.gle

  • Boris BecericOct 30, 2026Caveat

    A Google Ads consultant noticed that Experiments can now auto-apply a winning treatment, either on directional results that are not statistically significant or at a confidence level you set. No data.

    “You can allow Google to auto-apply the experiment, if the treatment arm wins.”

    Read it on x.com

  • Boris BecericOct 27, 2026Caveat

    Boris Beceric relays a 14-day AI Max test with 41% more form fills but 43% fewer qualified leads, and says 14 days is not a sample size; trust depends on volume, setup and time for leads to qualify.

    “14 days isn't a sample size”

    Read it on x.com

  • GoogleOct 26, 2026Supports

    Google advises running experiments one at a time for at least 4 to 6 weeks, discards the first 7 days, and calls a spend gap under 20% between arms normal. Avoid reactive changes in the first 2 weeks.

    “It’s recommended that you run the experiment for at least 4-6 weeks or longer if you have a long conversion delay.”

    Read it on support.google.com

  • GoogleOct 26, 2026Supports

    Google says a custom experiment shares your original campaign's traffic and budget, and is available for Search, Display, Video and Hotel Ads campaigns but not App or Shopping. Only one experiment can run at a time.

    “Custom experiments are only available for Search, Display, Video, and Hotel Ads campaigns.”

    Read it on support.google.com

  • GoogleOct 26, 2026Supports

    Google's experiments advice: set a clear hypothesis, test one variable at a time, pick one or two success metrics in advance, avoid changing base campaigns unless sync is on, and keep records.

    “When it’s time to end a test, let that metric tell you the winner.”

    Read it on support.google.com

  • GoogleOct 26, 2026Caveat

    Google recommends running a Google Ads experiment for 2 to 3 weeks, uses an 80% confidence interval by default, and auto-applies a winning trial to 100% of traffic unless you turn that off.

    “We recommend allowing your experiment to run for 2 to 3 weeks to gather data.”

    Read it on support.google.com

  • GoogleOct 26, 2026Supports

    Google says you can compare an existing Shopping campaign against a Performance Max campaign with similar settings using a traffic split, and should set the same Target ROAS on both. This is a rule, not a result.

    “Ensure that you set the same Target ROAS as the one in the Shopping campaign that you’re testing against.”

    Read it on support.google.com

  • Search Engine RoundtableOct 26, 2026Caveat

    Google's AI Max FAQ says eligible accounts may see a 1-click experiment recommendation that sets up a 50/50 test automatically. No eligibility rules or results are given.

    “Eligible accounts can now use a 1-click setup to create a 50/50 experiment automatically through Recommendations.”

    Read it on seroundtable.com

  • GoogleOct 25, 2026Supports

    Google's one-click Target ROAS experiments need a standard Shopping campaign on manual CPC with 15+ conversions in 30 days. Start the target near your past average, then change it, and test for 30 to 60 days.

    “Raising your ROAS target will limit Target ROAS to entering auctions that Google believes are more likely to lead to a conversion. This means higher ROAS, but less volume.”

    Read it on support.google.com

  • GoogleOct 24, 2026Supports

    Google says to test value-based bidding with a 50/50 campaign experiment on campaigns with at least 50 conversions in 30 days, a 2-week ramp-up, then at least 30 days judged on conversion value and ROAS.

    “Let the experiment run uninterrupted for at least 30 days.”

    Read it on support.google.com

  • GoogleOct 24, 2026Supports

    Google suggests a one-click A/B experiment that builds a Performance Max campaign from an existing Standard Shopping campaign, then keeping Performance Max if you like the results.

    “Will Performance Max outperform your Standard Shopping campaigns? Run an experiment to be sure.”

    Read it on support.google.com

  • GoogleOct 24, 2026Supports

    Google's AI Max experiments split traffic and budget inside an existing campaign rather than copying it; campaigns on portfolio bidding, shared budgets or with text customization on can't use this flow.

    “If your experiment ends without being applied to your base campaign, all AI Max settings will automatically revert to the original, pre-experiment state.”

    Read it on support.google.com

  • GoogleOct 24, 2026Caveat

    Google's Performance Max uplift experiments test adding Performance Max to your Search, Video and Display campaigns, with a 50% traffic split by default, read on Google-reported conversions.

    “The traffic split is 50% by default.”

    Read it on support.google.com

  • Relevant AudienceOct 19, 2026Supports

    Relevant Audience, an agency, says plan Google Ads experiments for a month or more with the first one to two weeks as ramp-up, and warns a 10% split on a small campaign may never reach significance.

    “Too small a split on a small campaign. A 10% treatment on a low-volume campaign may never reach significance.”

    Read it on relevantaudience.com

  • HubSpot (paid ads team)Sep 1, 2026Supports

    HubSpot's paid ads team saw a 12% rise in conversions in early AI Max tests, run as 50/50 cookie-based Google Ads experiments, usually 8 to 12 weeks. It publishes no test data.

    “From our early testing [we saw] a 12% increase in conversions.”

    Read it on marketingagainstthegrain.com

  • GoogleAug 20, 2026Caveat

    Google says a single A/B test will soon compare budgets and ROI targets across multiple Search campaigns (rolling out in September), and AI Max experiments now work with brand or location controls on.

    “test different budgets and ROI targets across multiple Search campaigns in a single A/B test.”

    Read it on blog.google

  • AmsiveAug 13, 2026Supports

    Amsive, an agency, says not to test every AI Max feature at once: test one, such as search term matching, then combine features that prove themselves. More in-campaign conversions are not proof of incremental ones.

    “More conversions inside the Search campaign do not necessarily mean more incremental conversions for the business.”

    Read it on amsive.com

  • Savvy RevenueJun 23, 2026Supports

    Savvy Revenue, which sells Standard Shopping management, ran one A/B test where simplified Standard Shopping led Performance Max on profit almost throughout, though not significantly at 95%.

    “PMax should not be treated as an automatic, one-size-fits-all upgrade from Standard Shopping.”

    Read it on savvyrevenue.com

In the full answer

For readers who sign in
  • What to do this week3 steps
  • The numbers behind it3 figures from cited sources
  • Where each source standsA chart placing each source on the question
  • The strongest disagreementThe best case against the answer, from a cited source
  • How sure we areWhat its confidence rests on, and what would change the answer

Read next

More in Measurement