What Is Ad Testing, and How Do You Test Ads Before Launch?

Learn what ad testing is, how pre- and post-launch tests differ, and which methods and metrics help you pick the right ad before buying media.

Melissa Tovin
What Is Ad Testing, and How Do You Test Ads Before Launch?

Ad testing is how brand and insights teams find out, before launch, whether an ad gets noticed, gets linked to the right brand and says what it means to say. It matters because media budgets are usually committed before anyone outside the company has seen the creative. This guide covers the methods, metrics and sample sizes for testing ads with target consumers while there is still time to change them, and how to turn the results into a launch decision.

What is ad testing?

Ad testing is a research method in which a team shows an ad, or an early version of it, to people from its target audience and measures how they react. It happens at two points. Pre-testing runs before launch, on anything from a written concept to a finished film. Post-testing runs after the ad is in market and checks what it achieved. The ARF’s creative research glossary, as reported by Quirk’s, treats copy testing, ad testing and creative testing as names for the same practice.

The benefit is timing. Reactions measured before launch show whether people notice the ad, remember the brand and understand the message, so a misread line or a late logo can still be edited before the media buy.

Why brands test ads before launch

Untested ads put media spend behind creative nobody has checked. NCSolutions’ 2023 update of its campaign meta-study, published by Circana, found that ad creative drives nearly half (49%) of incremental sales, ahead of brand, reach, targeting and recency.

Consumers can also mistake a weakly branded ad for a competitor’s. When the Ehrenberg-Bass Institute tested a past comparative Pepsi ad with more than 1,600 respondents, about one-fifth came away believing it was an ad for Coca-Cola. Teams can catch these failures through pre-testing while they can still edit a line people misread, a logo that appears too late, or a claim nobody believes.

Pre-testing vs post-testing ads

Each stage answers a different question. Pre-testing predicts how an ad will perform and shows what to fix. Post-testing measures what happened in market, usually by comparing people who saw the ad with a control group who did not.

This comparison shows how they divide the work:

  Pre-testing Post-testing
Timing Before launch, from concept to near-final During or after the campaign
Purpose Choose or fix creative before the buy Measure impact, track wear-out, decide what to keep
Typical methods Survey pre-tests, interviews, focus groups, eye tracking, AI predictive scoring Brand lift studies, ad platform experiments, ad tracking

Ad platform A/B tests sit between pre-testing and post-testing. A team serves two or more finished ad variants to live audiences through an ad platform, one of the self-serve systems where ads are bought and served, such as Meta Ads Manager, Google Ads and TikTok. The team spends a small budget, compares the variants on clicks and cost per result, and then scales the winner.

When to test ads in the creative process

Pre-testing works best as a series of checks, because each production stage answers a different question at a different cost. The stages run in this order:

  • Early concept: Researchers use a written idea, storyboard or key visual to check whether the core idea feels relevant and the message lands, and this is the cheapest point to drop a weak idea.
  • Rough cut or animatic: Researchers use a rough version to check story flow, whether the hook holds attention, and when the brand appears.
  • Near-final creative: Researchers give finished work a last check on branding, clarity and believability, scoring it against benchmarks or competing variants.
  • In-flight: Teams use brand lift studies and ad platform experiments to decide which variant to scale and when an ad starts to wear out.

The main ad testing methods

The methods below work in pairs: survey scores with interview explanations, and attention measurements with live behavior in ad platform experiments. The table shows the stage, turnaround and best use of each:

Method Stage Typical turnaround Best for
Survey-based pre-test Concept to near-final Same day to 3 weeks Scoring and comparing variants
Focus groups Early concept 2 to 6 weeks First reactions and consumer language
In-depth interviews (agency-run) Concept to near-final 3 to 6 weeks The reasons behind scores
Ad platform A/B test Live, small budget 7 to 30 days Which finished variant performs best on clicks
Eye tracking and facial coding Rough cut to near-final Several weeks for lab studies Attention and moment-by-moment reaction
AI predictive scoring Any, mostly early Minutes Screening many variants

Survey-based ad pre-testing

Starting with the quantitative option, survey pre-tests are the default method for measuring recall, appeal, clarity and purchase intent across hundreds of people. Researchers place the ad inside a panel survey and use closed questions to compare each variant with the others and with past norms. Teams can identify the winning ad from the scores, but rarely understand why it won.

Panel quality is the other limit. Pew Research Center’s 2026 study of 11,114 opt-in respondents found that standard checks for speeders and straightliners miss most bogus respondents. Ask any survey provider which checks it runs beyond those.

Focus groups and qualitative interviews

Qualitative research explains the reasons behind scores, and each interview format serves a different need:

  • Focus groups: These suit the early concept stage, where people react out loud and give a team the words they use for the category. A few people can do most of the talking and quieter participants can follow a confident first opinion, making an idea look stronger or weaker than it is.
  • One-on-one interviews: Researchers hear each person’s reaction without pressure from other participants.

A newer way to test ads is just coming to market: AI-moderated interviews, which sit between surveys and human-moderated in-depth interviews as their own method. In them, an AI moderator talks voice-to-voice with real consumers, follows the discussion guide, shows the ad, and asks follow-up questions based on what the person just said. Survey open-ends can’t ask what a confusing line meant, and an AI moderator can. Strella runs these interviews with its own AI moderator.

Anyone can try an interview to hear how the follow-ups work.

A/B testing and ad platform experiments

Ad platform experiments measure behavior on live media, so they compare creative that already exists. They cannot diagnose a concept. Meta’s A/B test in Experiments compares up to 5 versions across randomized, non-overlapping audiences and names the winner by lowest cost per result. Meta’s best practices recommend running tests for at least 7 days, up to 30.

Google’s Experiment Center tests new assets against current ones. Demand Gen experiments allow up to 10 arms and let teams read results at 70%, 80% or 95% confidence, and Google recommends running experiments for at least 4 weeks. These tools fit when a team has finished variants and wants to know which one performs. A gap in click-through rate does not explain why one idea worked. Once the A/B results are in, teams still need qualitative testing, such as interviews with people from the target audience, to learn why the winning variant won.

Eye tracking, facial coding and neuromarketing

Where surveys and interviews rely on what people say, biometric methods capture signals people cannot report in a survey: where their eyes go, and how their faces react second by second. Eye tracking shows whether the logo and product get seen. Facial coding times reactions to specific scenes, and EEG (electroencephalography) measures brain activity during viewing. Lab studies use small samples and cost the most per person. Webcam eye tracking reaches more people with less precision, and a 2023 validation study found effect sizes, the strength of the measured differences, shrank by 20% to 27% compared with lab eye tracking. Facial coding software reads posed expressions far better than spontaneous ones, so treat its output as a pointer to moments worth probing.

Reactions can also be read during the interview itself rather than in a lab. Strella’s emotion analysis reads a participant’s facial expressions and tone of voice during an interview and tags the moments that stand out, such as hesitant, confused, frustrated, or delighted, so researchers can see how people reacted, not just what they said.

AI predictive ad testing tools

Predictive tools score creative in minutes without showing it to anyone. Attention-prediction models produce heatmaps of where eyes are likely to land, and creative-scoring models compare an ad with databases of past tests. Vendors publish accuracy claims of 89% to 94% agreement with lab eye tracking, but those figures are their own, and no independent study compares a named tool’s predictions with sales. The ARF’s Phase 3 attention validation release (June 2026) recommends reading attention alongside media weight, reach and frequency, not as a stand-alone proxy for outcomes. Predictive tools are good for cutting twenty variants to four. The four that survive still go in front of people.

Monadic vs sequential monadic ad testing

Once a test has more than one variant, the team has to decide whether each person sees one ad or several. In a monadic design, each person sees one ad, and separate matched groups see the others. Reads are clean and comparable with norms, but every ad needs its own group: three ads at 200 people each means 600 people.

In a sequential monadic design, each person sees two or three ads one after another and rates each before moving on, which needs fewer people but runs longer. It introduces order bias: the ad seen first tends to score higher, and a strong ad can drag down the one after it. A peer-reviewed analysis of sequential monadic concept tests found concepts were rated higher in first position than in later ones. Teams reduce the bias by rotating order so each ad appears in each position equally, and by capping the sequence at about three ads. Monadic fits final go/no-go calls. Sequential monadic fits early screening of several concepts on a tighter budget.

Which creative elements to isolate in an ad test

A variant test only teaches something when the team knows exactly what changed between versions. These are the elements most worth isolating:

  • Hook: The first seconds or headline decide whether anyone keeps watching.
  • Core message: The single thing the ad asks people to believe.
  • Visuals: Casting, setting, product shots and color.
  • Call to action: What the ad asks people to do next.
  • Branding cues: When the brand appears, how often, and in what form.

Change one element per variant. If two ads differ in both hook and music, nobody can say which change produced the winner. Meta’s best practices make the same point for ad platform tests: ad sets should be identical except for the variable being tested.

Ad testing metrics and survey questions

Metrics turn reactions into comparable numbers, and each one works best paired with a follow-up that asks for the reason. Scaled questions are usually reported as top-two-box, the share choosing the two most positive points on a five-point scale. Typical pairings:

Metric Sample question Follow-up probe
Recall Which ads do you remember seeing? (asked after a reel of several ads) What do you remember about it?
Brand linkage Which brand was this ad for? What in the ad told you that?
Appeal How much did you like this ad? What did you like most, and least?
Message clarity In your own words, what is this ad trying to say? Which part, if any, was unclear?
Believability How believable is the main claim? What would make it more believable?
Purchase intent How likely are you to buy [brand] next time you shop for [category]? (definitely would to definitely would not) What would hold you back?
Brand lift The same awareness or consideration question, asked of exposed and control groups Not typically probed

Brand lift compares the two groups. Google defines absolute lift as the exposed group’s positive response rate minus the baseline group’s, so a move from 20% to 40% is a 20-point absolute lift and a 100% relative lift.

Sample size and statistical significance for ad tests

The metrics above only help if the sample is large enough to separate a true winner from noise. For monadic tests, a Quirk’s guide to pricing research recommends at least 200 respondents per cell for statistical power. At that size, only fairly large gaps, roughly 5 to 10 points or more, are likely to hold up. Small gaps need far more people: Google’s Brand Lift guidance puts the total responses needed to detect an absolute lift above 4% at 1,200 to 2,800, and a 1% lift at 20,000 to 45,000. Any subgroup a team wants to read on its own needs its own adequate base.

Qualitative samples work differently, because the goal is to hear the full range of reactions. A 2022 systematic review by Hennink and Kaiser found most themes surfaced within 9 to 17 interviews when the population was similar and the objective narrow. Mixed segments need more, so plan interviews per segment.

Worked example: testing two ad concepts before launch

This example is hypothetical and illustrative. A skincare brand has two 15-second concepts for a daily sunscreen. Concept A leads with texture in a morning-routine scene. Concept B leads with a sun-protection claim in a beach setting. The team runs a monadic survey with 250 target consumers per concept, then 20 interviews per concept with the same audience.

The survey returns these top-two-box scores:

Metric Concept A Concept B
Appeal 58% 49%
Brand linkage 41% 67%
Message clarity 62% 71%
Believability 55% 44%
Purchase intent 38% 36%

The interviews explain the gaps. People liked Concept A’s scene, but several assumed it was a moisturizer ad because the pack appears only in the final two seconds. Concept B was clear and well branded, but people who rarely spend time outdoors said the sun-protection message wasn’t meant for them. The 2-point purchase intent gap is noise at this sample size. The team’s call: launch neither version as it stands, keep Concept A’s idea, move the pack into the first three seconds, and retest brand linkage before the buy.

How to choose an ad testing tool

The choice of tool follows from the question the team needs answered and how soon. These criteria separate the options:

  • Speed: The turnaround has to fit the decision window, including time for a retest.
  • Depth: Some tools return scores only; others return scores plus the reasons behind them.
  • Panel quality: The provider should say who its respondents are and how it verifies them.
  • Fraud checks: Checks should run before, during and after each response, since speed and straightlining checks alone miss most bogus respondents.
  • Cost: Count sample, incentives, analysis time and retests, not only the headline fee.

Research panels and survey tools give benchmarked scores. AI predictive tools screen many variants quickly. Ad platform experiments measure live performance. Interview tools explain why.

Strella lets teams add the ad to the discussion guide as an image, video or ad copy. Strella’s AI moderator then asks each person follow-up questions about what confused, persuaded or put them off, and emotion analysis tags the moments where people hesitated or were delighted. Strella links each finding back to the participant’s own words, so the verbatims sit behind the scores.

Teams without research bandwidth can use Strella’s full service. If the decision is which finished variant to scale, an ad platform experiment answers it; if the decision is whether an idea works and how to fix it, test with people before the buy and pair the scores with interviews.

Getting started with ad testing on Strella

Running a first ad test on Strella goes faster when the team agrees on these points before the study starts:

  1. Objective: Name the decision, such as which of two concepts gets produced. Define the objective and let Strella draft a fully editable discussion guide, or bring an existing guide. Bias detection flags leading questions.
  2. Audience: Define who the ad must reach, and set custom screeners and quotas. Recruit through the team’s own consumers via a direct link, Strella’s panel partners or expert networks.
  3. Elements to isolate: Pick the one variable each variant changes.
  4. Method and stimulus: Add the ads to the discussion guide as images, videos or ad copy, alongside open questions, rating scales and multiple choice.
  5. Design: Decide whether each person sees one ad or several. When several are shown, randomize the order.
  6. Metrics: Select the four to six measures that tie to the objective. The AI moderator asks follow-up questions on each answer, and synthesis begins as interviews finish, with themes and source-linked quotes.
  7. Decision rule: Set the score or gap that counts as a win before results arrive.

See Strella in action.

FAQ

What is the difference between pre-testing and post-testing ads?

The difference is timing and purpose: pre-testing protects the media buy by letting the team choose or fix creative while changes are cheap, and post-testing shows what to keep, rework or retire once the ad has run, usually through brand lift studies that compare exposed and unexposed groups.

How many people do you need for an ad test?

Quantitative monadic tests usually need at least 200 people per ad variant, and more when the expected difference is small or subgroups need separate reads. Qualitative interviews tend to surface most themes within 9 to 17 conversations per similar segment. Mixed audiences need interviews planned for each segment.

How long does ad testing take?

AI predictive scoring returns results in minutes; survey pre-test timing varies with the audience and study design. Focus groups take 2 to 6 weeks and agency-run interviews 3 to 6 weeks. Meta recommends running ad platform tests for 7 to 30 days, and Google recommends at least 4 weeks.

Can you test ads with AI-moderated interviews?

Yes, teams can test ads with AI-moderated interviews. They embed concept boards, images, video and ad copy in the discussion guide, and the AI moderator asks each person follow-up questions about what confused, persuaded or put them off, which gives the reasons behind survey scores across a full segment.

How does ad testing improve ROI?

Teams use ad testing to improve return by identifying weak branding, unclear messages and unbelievable claims before committing media money, and by directing budget toward the strongest variant. It matters because creative accounts for nearly half of incremental sales in NCSolutions’ 2023 analysis, so a better ad improves the return on every dollar of the buy.