Home / Insights / What is ad pre-testing?
What is ad pre-testing?
Ad pre-testing is research that shows an ad to a sample of its target audience before it runs, to find out whether it will work and what to change. It is done while the ad can still be fixed: at storyboard or animatic stage, on the finished edits, or on the cut-downs. A pre-test measures how people respond to the ad: whether it holds attention, what they feel and when, whether they link it to the brand, whether they understand the message, and whether it changes what they think of the product. The main methods are surveys, facial coding, eye tracking, implicit reaction-time tests, qualitative groups and, more recently, synthetic audiences and predictive AI scores that estimate a result without any real viewers. Pre-testing is different from post-testing, which measures what an ad did after it ran, and from in-market A/B testing, which spends real media money to compare versions. It earns its cost when one edit will carry a large TV or online video budget.
Updated 19 October 2026 by Jonathan Prescott, Founder, Cavefish.
Ad pre-testing methods compared
Most pre-tests combine two or more of these. The question is what you need to know, and whether you need it measured or estimated.
| Method | What it measures | Real viewers? | Strength | Limit |
|---|---|---|---|---|
| Survey | Stated liking, recall, brand linkage, message take-out, purchase intent | Yes | Cheap, familiar, easy to compare | People report what they think they felt, after the ad has ended |
| Facial coding | Facial muscle movements (FACS Action Units), turned into emotional response second by second | Yes, on webcam | Shows when the response builds or drops, without asking | Reads expression, not thought; needs consent and a clear view of the face |
| Eye tracking | Where people look, and for how long | Yes | Shows whether the brand and key frames are seen | Seeing something is not the same as feeling anything about it |
| Implicit tests | Speed of association between the brand and words or images | Yes | Gets past considered answers | Measures association after viewing, not the viewing itself |
| Qualitative groups | Discussion of the idea, the story and the characters | Yes, small numbers | Explains why, in people's own words | Group dynamics shape what people say |
| Synthetic audience | Modelled people from your audience watch the ad in their own channels and answer recall and brand questions | No, modelled | Minutes; shows which segments and channels an ad suits | A model of an audience; only as good as its calibration against real people |
| Predictive AI score | An estimate of how the ad would perform, from a model trained on past ads | No | Minutes, not days; useful for screening many ideas | A prediction, not a measurement of this ad with this audience |
For named tools that use each method, see best ad testing tools in 2026. For what facial coding can and cannot tell you, see is facial coding accurate?
What a pre-test measures
- Attention: whether people keep watching, and where they drift.
- Emotional response: what people feel, and in which seconds.
- Brand linkage: whether people know which brand the ad was for.
- Message: whether the main point lands as intended.
- Persuasion: whether the ad changes what people think of the product or how likely they are to buy it.
- Stated vs felt: whether what people say about the ad matches how they responded while watching it.
Pre-testing, post-testing and A/B testing
| When | Answers | |
|---|---|---|
| Pre-testing | Before the ad runs | Will it work, and what should we change? |
| Post-testing (tracking) | After the ad has run | Did people remember it, and did the brand move? |
| In-market A/B test | While the ad runs, with real media spend | Which version gets more clicks or sales? |
When to pre-test an ad
Pre-test at two points if you can. Test the animatic or rough cut to check the idea and the story before production money is spent. Then test two or three finished edits, each with its own sample, to choose between them and to fix the weak seconds before the 20, 10 and 6-second cut-downs are made. Skip a formal pre-test when the media budget is small enough that running two variants in market costs less than the test. The full week-by-week plan is in how to pre-test a video ad, and what it costs is in how much does ad testing cost?
How EchoDepth pre-tests an ad
Real, consenting viewers from your audience watch the ad at home on their own webcam. Their facial movements are coded against 44 FACS Action Units and plotted second by second as the Emotional Response Curve, and what they say afterwards is scored as the Cavefish Emotional Index (CEI). We recommend 15 or more viewers per audience segment for each ad or edit, and results arrive in less than 24 hours from the start of fieldwork. The response is measured on real viewers, not predicted by a model. See ad testing and the methodology.
Ad pre-testing questions
Testing a specific format? See TV ad pre-testing, YouTube ad testing and TikTok, Reels and Shorts.
What is the difference between ad pre-testing and post-testing?
Pre-testing happens before an ad runs, while you can still change it. Post-testing, often called ad tracking, measures what the ad did after it ran: recall, brand awareness and sales. Pre-testing tells you what to fix; post-testing tells you what happened.
Is A/B testing the same as ad pre-testing?
No. An in-market A/B test runs two versions with real media money and compares clicks or conversions. A pre-test shows the ad to a sample of the target audience before any media is bought. A/B tests suit cheap social ads; pre-tests suit an ad that will carry a large TV or online video budget.
When should you pre-test an ad?
Twice if you can: once at animatic or rough-cut stage, to check the idea before production money is spent, and once on the finished edits, to choose between them and fix weak moments before cut-downs are made.
How many people do you need to pre-test an ad?
EchoDepth recommends 15 or more viewers per audience segment for each ad or edit. Every group you want to report on separately is its own segment and needs its own 15.
Can AI pre-test an ad without viewers?
Predictive tools score an ad with a model trained on past ads, in minutes and without viewers. That is a prediction, not a measurement. It is useful for screening many rough ideas; for the edit that will carry the budget, a test with real viewers shows what actually happened in each second.