Base·Note Data

Blog · Incrementality · Google Ads

How Google's Conversion Lift works

How Google Ads Conversion Lift works, user-based versus geo-based studies, the requirements, the November 2026 data rules, and how to read Bayesian results.

Incrementality testing by platform: Overview · Meta · Google · TikTok · Amazon · Pinterest · Snapchat

Google's Conversion Lift hides your ads from a random slice of your audience, then measures how many extra conversions the rest produced. You can split by people (user-based) or by region (geo-based).

The idea in one picture

Google shows the control group the next ad in the auction instead of yours People or regionsrandomly split Test groupwins the auction, sees your ad Control groupsees the next ad instead Liftextra conversions User-based studies split people; geo-based studies split regions. Holdback can be 1 to 50 percent.

Both groups see ads. Only the test group sees yours, so the difference is yours alone.

Two study types

User-basedGeo-based
Groups built fromAggregated user attributesAggregated geographic data
BreakdownsAge and genderNo demographic breakdown
Offline conversionsNot the focusSupported
BudgetSmallerGenerally higher
Main limitRelies mostly on opt-in traffic; no iOS app campaignsNoisier split between groups

Sources: Google Ads Help and PPC Land.

How a study runs

  1. You pick campaigns and conversions. Google shows a feasibility rating, called study power, from 50% to 95%. Aim for 90% or higher; below that it gives budget guidance.
  2. You set a holdback of 1–50% of the audience and a length. Studies combined with Brand Lift or Search Lift fix the holdback at 30%.
  3. Google randomizes the audience. Control-group users are shown the next ad in the auction instead of yours (Display & Video 360 Help).
  4. Measuring held-out users relies on hashed customer data, so accurate setup of that data matters.
  5. Results start appearing on the fourth reporting day; processing can take up to 10 days.

What you need

  • Google Ads, user-based: at least 1,000 observed conversions (modeled ones don't count), a $5,000 budget, and a compatible conversion action. Display, Search, Video, Demand Gen, App (Android only) and Performance Max are supported; iOS app campaigns and Travel Ads aren't (Google Ads Help).
  • Display & Video 360: campaigns should reach about 3,000 conversions during the study. Studies run 7 to 56 days.
  • Length: 7 days is the minimum and more than 14 is recommended. Studies under 14 days showed absolute lift drops of up to 17% for advertisers with long conversion lags.
  • Bayesian method: results need at least 150 conversions from the treatment group and 65 from the control, and spend above $5,000 (Google Ads Help).
  • A campaign can sit in only one study at a time.
  • Access: not every account has it. Eligible Search and Performance Max campaigns became self-serve on October 7, 2026; before that they needed a Google rep (Search Engine Land).

Check your data before you launch

Google rewrote its setup page between mid-October and mid-November 2026 to spell out which conversion setups work. Search Engine Roundtable spotted the change on November 13 (Search Engine Roundtable, Relevant Audience):

  • Not supported: web conversions imported from Google Analytics. Offline imports with only a click ID (GCLID) aren't enough either; use enhanced conversions for leads with hashed customer data.
  • Consent: basic consent mode is insufficient for campaigns targeting the EEA, UK and DMA regions. You need advanced consent mode.
  • Values: conversion actions with no value, a static value or no valid currency can't report incremental value or iROAS accurately.
  • Recommended: 7 to 14 days of continuous conversion data before launch, and routing 75–80% of web conversion traffic through Google tag gateway.

The core thresholds didn't change: 1,000 observed conversions, a $5,000 budget and the same eligible campaign types.

Reading the results

Google reports incremental conversions, relative lift, incremental conversion value, incremental cost per action (iCPA) and incremental ROAS (iROAS, only when conversions carry a value).

Results use Bayesian statistics, so the interval reads as a probability: an 80% credible interval means an 80% chance the true lift sits inside it. Results appear only once certainty of lift tops 50%. PPC Land summarizes the bands as 90%+ very good, 70–90% good, and 50–70% directional. A "no lift" result doesn't prove the ads failed.

Google's own examples show the range: a beauty brand's Performance Max test returned an incremental ROAS of £6 per £1, while a financial institution's YouTube test returned £1.10 (Think with Google).

Recent changes

  • May 22, 2025: Google moved to Bayesian methods and set 1,000 conversions and $5,000 as user-based minimums (PPC Land).
  • November 11, 2025: feasibility ratings arrived. "High" means a 60–90% chance of a conclusive result.
  • May 2026: Meridian GeoX began feeding geo experiments into Google's mix model.
  • August 19, 2026: Google Ads API v25.1 exposed 24 Conversion Lift metrics to allowlisted accounts. Studies still can't be created through the API.
  • October 7, 2026: Search and Performance Max became eligible for self-serve user-based studies.
  • November 13, 2026: the setup page's new data readiness checks and list of unsupported setups were reported.

Watch-outs

  • Results are a lower bound. Browser tracking limits and cookie churn can hide real conversions.
  • Other campaigns and seasonality can contaminate the control group. Don't run overlapping experiments, especially prospecting and remarketing together.
  • Apple's tracking limits hurt click-based measurement. If many iOS conversions go missing, consider excluding iOS traffic.
  • Google's guidance links conversion-based bidding and data-driven attribution with higher measured lift, advice that also steers spend toward its automated products.
  • Google designs and scores its own test, and it sees Google only.

I design and read incrementality tests for brands, on the platforms and independently of them. See how a Measurement Audit works.

Sources

More posts Services and pricing

Work with me

Want a second set of eyes on your measurement?

I help marketing and growth teams figure out what their spend is actually driving, with MMM, incrementality testing and attribution that finance trusts too.

Book a 30-min call