Brand lift is one of the few metrics built to answer the question performance dashboards ignore: did the advertising actually change what people think? Instead of counting clicks or last-click conversions, it borrows a clinical-trial design and compares survey answers between an exposed group and a control group.
How the test is constructed
Randomisation happens before bidding begins. Eligible users are split into an exposed cell that can see the ad and a control cell that is deliberately held back. Targeting, inventory and flight dates stay the same, so exposure is the only systematic difference.
After exposure, both cells answer the same survey question. On YouTube, that is typically a single question served before a video about a day after the ad ran. Google Ads reports ad recall, brand awareness and consideration, while Display & Video 360 can also cover association, favourability, purchase intent and video ad recall, with a ceiling of three metrics per study.
Three lift numbers, three stories
- Absolute lift is the raw percentage-point gap between exposed and control responses.
- Relative lift divides that gap by the control baseline; it can flatter brands with little existing awareness.
- Headroom lift divides the gap by the distance to a perfect score, a fairer read for brands already known in their category.
Budgets, sample sizes and confidence
Sample size governs whether a study can see anything. Display & Video 360 documentation suggests about 2,000 responses per metric before strong campaigns start producing readable results. Google publishes budget floors for YouTube: roughly $5,000 to $20,000 for one to three questions in the cheapest country tier over ten days, rising to $15,000 to $60,000 in the most expensive markets. Connected TV cross-exchange studies have no formal minimum, but around $15,000 per question is recommended.
Google also reports a certainty of lift measure, equivalent to one minus the p-value. It targets 90%, treats 50% to 90% as directional, and labels anything below 50% as no lift. Google’s Enhanced Brand Lift Studies, released July 20, 2026, can detect lifts as low as 1.2% and raise the chance of detecting positive lift by 60%, but at roughly three times the budget.
Who runs it and what’s changing
Advertisers or agencies usually configure brand lift inside a demand-side platform. Amazon opened a self-service workflow in May 2026 across more than 50 third-party measurement products in 18 countries, with DISQO and Kantar supplying brand lift. On The Trade Desk, LoopMe lowered the floor to one million impressions from September 2, 2026, with control groups matched on more than 100 variables across 35 markets and reports refreshed every 24 hours.
Independent vendors such as Cint, Kantar, Dynata, DISQO and Upwave exist in part because the seller of the impression often fields the survey that judges it. That structural oddity is a major reason to validate platform-run results outside the platform. Upwave’s surveyless Predictive Brand Lift beta, opened on September 9, 2026, uses Bayesian modelling to estimate what a survey would have found. Chief executive Chris Kelly said constraints meant “advertisers could only measure a fraction of their campaigns”. It trades modelled coverage for observed evidence from the campaign itself.
Where the method trips up
Response rates remain a long-standing weakness. A Digiday report from 2014 cited sub-1% response rates for pop-up surveys versus roughly 9% for panel-based work, and panel response rates have kept falling since. Statistical power is the sharper constraint: most campaigns shift brand metrics by low single-digit points, so a flat result often reflects an underpowered test rather than a failed campaign. Control groups can also be contaminated when held-out users meet the brand elsewhere.
Stated answers are not behaviour. A FreeWheel and MediaScience study published in September 2026 found 33% unaided recall for a streaming environment against 12% for the same ad in a social feed, with heavier ad loads cutting recall from 41% to 26%. That gap rarely appears in a single platform-run study.
Use it as one causal signal
Meta’s May 2025 framework put randomised brand lift at the top of its incrementality ladder, but also found the average advertiser runs 3.8 measurement solutions and 55% get contradictory results. Amazon’s analysis of more than 3,000 measurements found display plus video produced 2.2 times the brand awareness of video alone, and that upper and lower funnel together doubled brand lift—though those numbers come from the media seller.
The practical move is to pair brand lift with conversion lift, set a realistic budget floor, and choose the lift calculation that fits your brand’s starting awareness. If you can only afford a small test, be honest that a flat result may be noise.
Source: PPC Land
