Google's three rules for A/B testing Google Ads changes

Google's Ads team laid out how to test AI Max, budgets and Smart Bidding before you scale. Here is how a local advertiser can run a clean test.

SEOchat News Desk ·

Share

Google wants advertisers to prove a change works before they roll it out across an account. On November 19, 2026, the Google Ads team published Build strategic A/B tests to grow your ad performance, a summary of a new Ads Decoded episode in which Ads Product Liaison Ginny Marvin talks with product leads Sagar Shah and Al Nevarez about using experiments to check AI Max features, budget shifts and Smart Bidding strategies.

The post is short, but it boils the advice down to three rules. For a small business spending real money on search ads, those rules are a useful filter against changing too much on a hunch, especially in a month when Google keeps shipping new automation.

The three rules Google gave

1. Start from a template. Google says its pre-built experiment templates test one variable at a time, split traffic 50/50 and cut down on setup mistakes, which gives you cleaner data.

2. Judge the test on business results. The post says to measure qualified conversions instead of shallow stand-ins, and to let a test run four to six weeks so conversions that arrive late still get counted.

3. Match your confidence level to the size of the decision. Google suggests an 80% confidence level is fine for small, directional tweaks, and that 95% certainty should be saved for big structural changes.

The post closes by pointing advertisers to the Experiments tab in Google Ads and telling them to let the data prove a change before scaling it.

How experiments work in Google Ads

Google's help page on how to set up a custom experiment fills in the mechanics. An experiment is built from an existing campaign and shares that campaign's traffic and budget, so you are comparing two versions under the same conditions rather than comparing this month to last month.

A few details from that page matter for smaller accounts:

  • Custom experiments work for Search, Display, Video and Hotel campaigns. They are not available for App or Shopping campaigns.
  • You can schedule up to five experiments, but only one runs at a time.
  • For Search campaigns you can split by cookie or by search. Display experiments always use a cookie split, so a person sees only one version.
  • Changes you make to the original campaign during a test do not copy into the experiment, and editing either side mid-test makes the result harder to read.
  • The original campaign has to stay active. If it is paused or ends before the experiment finishes, the experiment stops serving.
  • If you use a cookie split with audience lists, Google says the list should have at least 10,000 users, or the results may be less accurate.
  • You can tell Google Ads to apply the winning version automatically, and choose whether a directional result is enough or statistical significance is required.

What this means for local businesses

This is guidance, not a new product, but it lands at the right moment. Google has been pushing AI Max and new bidding options hard, and plenty of local advertisers flip those settings on across the whole account and hope. A test protects you from that.

Pick the one change you are least sure about. For a plumber or HVAC company, that might be turning on AI Max in the campaign that drives emergency calls. For a dentist, it might be moving from a manual approach to a Smart Bidding strategy. Test that single change, not three at once, which is exactly what the templates are built to enforce.

Define a conversion that pays the bills. "Qualified conversions" is the key phrase. A roofer should count booked inspections or phone calls over a set length, not page views or every form fill. A lawyer should count consultations, not newsletter signups. If your conversion tracking counts junk, the experiment will happily pick the version that produces more junk.

Budget for the time. Four to six weeks is a long wait for a small shop, but it reflects how local leads behave. Someone who clicks an ad for a med spa or a realtor often calls back days later. Cutting a test short tends to reward whichever version front-loads cheap clicks.

Use the lower bar for small calls. Google's 80% versus 95% split is a practical way to move faster. Testing a new headline angle or a modest budget shift can run on a directional read. Restructuring campaigns or handing bidding fully to automation deserves the stricter standard.

Know your volume limits. A campaign that gets a handful of conversions a month may never reach a confident result in six weeks. In that case, test in your busiest campaign first, or accept a directional answer and keep watching after you apply it.

If you are weighing whether more ad spend is worth it at all, run the numbers through our SEO ROI calculator alongside your paid results, since organic and paid leads often come from the same searches. We also covered Google's recent invalid traffic transparency card, which helps you check whether the clicks in a test were real, and the expansion of ads in AI Mode, which is changing where search ads appear. Industry playbooks for plumbers, HVAC companies and lawyers show how those trades tend to generate leads.

The bottom line

Google's message is simple: automation is powerful, but prove it on your own account first. One variable, a real business outcome, enough time and a confidence bar that fits the stakes. A local business that follows those four habits will spend less money learning what does not work.

Sources

For your business

Related stories