Brand lift study: what it measures and how to read it

What a brand lift study actually measures

It measures a change in belief.

Not a click. Not a view. A survey question is put to people who had the chance to see your ads and to people who did not. The platform compares how many in each group answer positively. That difference is the lift.

So this page is about measuring awareness. If you are still working out how to build it in the first place, start with our guide to building brand awareness and come back here when there is a campaign worth testing.

Google's Brand Lift lets you choose up to three things to ask about: ad recall, awareness, association, consideration, favourability, purchase intent, or a custom metric (Google Ads Help, accessed 15 October 2026).

Meta's version is narrower. Every test must include a standard ad recall question — for example, "do you recall seeing an ad for [Page name] online or on a mobile device in the last 2 days?" — and you can add up to two more from brand awareness, favourability, familiarity and recommendation (Meta Business Help Centre, accessed 15 October 2026).

One line on scope: brand lift is the awareness twin of conversion lift. Same randomised design, different outcome. If the outcome you care about is sales or sign-ups, that is incrementality testing, and it has its own page.

One gap, three names.

How the exposed-versus-control design works

This is the part worth understanding, because it is the reason the result means anything at all.

1. The platform holds back a random slice of your target audience. These people are eligible for your ads but are deliberately not shown them. This is the control group, or holdout.
2. Everyone else sees the campaign as normal. This is the exposed group.
3. Both groups are served the same survey question, in feed or before a video.
4. The platform compares the positive response rate of each group.
5. It then tests whether the gap is large enough to be unlikely from chance alone.

Because the split is random, the two groups should be alike in every way except one: whether they saw your ads. That single controlled difference is what lets you say the ads caused the change.

Nothing else in your stack does that. Attribution models divide credit among touchpoints that already happened. Marketing mix modelling fits a statistical model to historical spend. Both are inference. A brand lift study is an experiment.

Meta describes its own test exactly this way: it "surveys both people who had the opportunity to see your ads and people in a holdout group", then "calculates the incremental effect your ads had on brand awareness, perception or recall" (Meta Business Help Centre, accessed 15 October 2026).

What Google and Meta offer, and the gate you have to clear

Four products, all live as of October 2026. The eligibility figures below are the ones the platforms publish, and they change — check them before you plan around them.

Both platforms publish these thresholds in US dollars and convert them to rupees at their own exchange rates — Google refreshes its rates four times a year. So read the figure your own account shows you rather than doing the sum yourself.

Two timing details matter. Google's standard study completes in roughly 3 to 14 days, and a study qualifies for Enhanced Lift if its budget is 3x the minimum. Meta recommends running the test for at least four weeks and long enough to collect more than 250 survey responses.

How to read the result without overclaiming

This is where most brand lift decks go wrong. Google reports three different lift numbers and they are not interchangeable.

Working through the maths: say the control group's positive response rate comes back at 8% and the exposed group's at 10%. Those are made-up inputs, but the arithmetic is the real formula set from Google's own documentation.

All three describe the same two percentage points. The misreading that inflates a small change is reporting "25% lift" and letting the room hear it as one in four people. One in fifty changed their answer. Say which number you are quoting, every time.

Then read the uncertainty. Google's statuses page says lift results use 80% two-sided confidence intervals, and defines certainty of lift as 1 minus the p-value. The same page mentions a 90% interval in one section, so read the interval printed in your own report rather than assuming. Google shows one of three statuses: a lift percentage, "Not enough data", or "No lift detected" — the last meaning there was no statistically significant difference between the groups.

Meta's threshold is stricter and plainly stated: "for lift tests, a 90 percent or higher confidence percentage represents a statistically reliable result" (Meta Business Help Centre, accessed 15 October 2026).

A flat result is not a failed study. "No lift detected" is information. It is usually cheaper to learn it from one experiment than from a year of marketing KPIs that drift sideways.

The honest caveats

Four, and you should put them on the slide yourself before a client finds them.

• The platform grades its own exam. Google and Meta design the holdout, serve the survey, run the statistics and publish the verdict. The method is sound. The marking is still in-house, and neither platform lets you audit the raw response data.
• Survey response bias is real. Only people willing to answer a poll in their feed are counted. Meta states it tries to reach people it thinks can respond to your questions. That is sensible targeting, and it also means your respondents are not a clean random sample of humanity.
• The result is about the measured surface only. A Google study tells you what YouTube ads did. A Meta study tells you what Meta did. Neither says anything about your print, your connected TV buying outside DV360, or your whole media plan.
• Comparison benchmarks are directional. Meta asks you to pick a region and a vertical so it can show your result against an average. Useful context. Not a target.

Who should actually bother

Be blunt about this, because the thresholds decide it for you.

If you are spending less than the platform minimum, a brand lift study is not something you have chosen against. It is not available. Google's cheapest India gate is $5,000 of campaign budget inside ten days for a single question. Meta's India gate is nearly nine times that.

Most Indian advertisers reading this are below both. That is fine, and the honest alternatives are:

• Branded search volume and direct traffic trended over the campaign window — crude, free, and directionally useful.
• A small market research survey of your own, run before and after a burst. Weaker than a randomised holdout, because you have no control group, but it is yours and you can see the raw data.
• Geo holdouts, if your media buying can switch whole cities on and off cleanly.

And if you are above the gate: do not spend the one experiment you qualify for on a question you already know the answer to.

What we want agreed before a brand lift study runs

DH's own view, from running measurement conversations on client accounts: the study is the easy part. The arguing happens afterwards, when a number nobody pre-committed to lands in an inbox. So we get six things in writing first.

1. Which metric, and why that one. Ad recall for a launch. Consideration for a brand with recall but no shortlist position. Pick before you see data.
2. Which lift number goes in the report. Absolute, relative or headroom. Named now, so nobody swaps to the flattering one later.
3. What a flat result means. Write the sentence in advance. "No lift detected means we change the creative, not the channel" is a decision. "We'll see" is not.
4. The surface being measured, stated as a limit. One platform, one date range, one audience. On the slide, not in a footnote.
5. Who holds the campaign steady. No creative swaps, no budget jumps, no audience edits mid-study. A changed campaign measures nothing.
6. What happens to the holdout group's lost reach. Someone will ask why a slice of the audience got nothing. Answer it before they do.

Agree those six and the study is usable whichever way it lands. Skip them and a positive result gets celebrated, a flat one gets quietly dropped, and you have paid for an experiment you did not learn from.

Where brand lift fits in a measurement stack you can defend

Treat it as one instrument, not the dashboard.

Brand lift answers whether a specific platform shifted belief. Incrementality testing answers whether it shifted sales. Marketing mix modelling answers how the whole plan splits credit across channels and seasons. They disagree sometimes, and that is useful — three methods agreeing is far stronger evidence than one method shouting.

Run brand lift when you have the budget to clear the gate, a campaign you can leave alone for a fortnight, and a decision waiting on the answer. If any of the three is missing, save it for the quarter when all three are there. Our performance marketing guide sets out how the pieces sit together.