How to Pretest Ad Messages with AI Mock-ups Before Production

Ad pretesting is the assessment of an advertisement’s message, brand connection and intended meaning before publication. An AI mock-up makes an unfilmed idea visible as a still image or short video. It helps you identify unclear messages and decide which ideas warrant development before committing to production.

Summary: Define the decision first, then prepare mock-ups that differ only in the element being tested. Check product accuracy before showing them to your target audience. Assess understanding, recall and liking separately. Give the production team findings with their uncertainties, and measure campaign performance separately after launch.

What did the research show, and what do we recommend?

The research compared existing ads with AI recreations, rather than selecting a new advertising strategy. In the main analyses of ads passing the quality screen, no significant overall group differences were detected across five measures: unaided brand recall, brand recognition, hypothetical brand choice, brand attitude and ad liking. This does not establish equivalence.

Bellman, Ourradi and Varan’s Scientific Reports paper, published on 15 August 2026, is a peer-reviewed, accepted Article in Press awaiting editorial corrections. The authors state that their samples were insufficient for equivalence testing.

Two US online-panel experiments used random assignment, with reported samples of 200 people for static ads and 300 for video ads. They examined informational advertising under low-involvement viewing conditions, without measuring purchases or advertising returns. The rationale for more finished mock-ups comes from earlier literature, rather than these experiments.

MediaScience funded ad production and data collection. Ourradi and Varan work for the company, which also partly supports Bellman’s salary. Bellman’s statement of independent analysis is not independent replication by another team. Our research review explains the findings.

DijitalPi’s interpretation: Use mock-ups to inform production decisions. Examine understanding before preferences. These findings do not predict your sales or validate any particular generation tool.

Why is pretesting different from a live A/B test?

Pretesting examines an advertising idea before production; live A/B testing compares alternatives in an actual advertising environment. The former helps clarify what to produce. The latter informs delivery decisions using paid exposure and campaign measurements. An answer about message understanding cannot directly answer a question about performance after launch.

Meta’s A/B testing tool uses non-overlapping random audience groups to compare up to five advertising variations, changing one variable while holding others constant. It runs with advertising spend. A preferred mock-up is not necessarily a stronger live ad.

See our A/B testing guide and social media advertising service for method and implementation.

Step 1: Write down the decision you want to test

Define a single decision you will change in response to the findings. It might concern the message, promise, opening or visual approach. Changing all of them together makes reactions difficult to attribute. A decision card keeps production and evaluation focused on the same question and discourages changing the rationale afterwards.

Replace “Which looks better?” with an actionable question:

• Message: Do viewers understand the need your service addresses?
• Promise: Do they interpret the benefit too broadly?
• Opening: Does the first scene establish the subject?
• Visual approach: Does the image explain product use?

Include your commercial priority. If accurate understanding matters most, liking alone should not determine selection. Decide beforehand how conflicting findings will be weighed.

Step 2: Build the AI mock-up around the actual offer

Prepare a controlled draft that accurately represents your product or service. Static images can test message placement; short videos can examine openings and scene order. Change only the element under investigation and keep other production choices as consistent as possible, so execution differences do not obscure responses to the message.

Your production brief should specify:

• Verified product features and service scope.
• Approved logos, colours and copy.
• The usage or service situation to depict.
• Features and promises that must not be added.
• Fixed duration, voiceover, ending and call to action.

Copy may change when testing messages. When testing an opening visual, retain the same copy. Have the product owner check each scene against the actual offer.

The study’s controlled production also involved human intervention for scene continuity and used MediaScience’s proprietary tool. Within DijitalPi’s human-edited, AI-assisted working model, product verification remains part of preparing a usable mock-up.

Step 3: Screen out mock-ups that should not enter testing

A mock-up should not enter testing if production errors are likely to dominate viewers’ reactions. Screening makes alternatives understandable and comparable; it does not require a finished commercial. Complete product, brand and usage-permission checks before exposure, then review corrected versions and clearly identify the approved files for the test.

Return drafts for correction if they contain:

• Incorrect product features, dimensions or usage.
• Distorted logos, incorrect colours or unreadable branding.
• Unnatural hands or faces, or broken text.
• Services or outcomes outside the offer.
• Images with unclear usage permissions or licences.
• Quality differences that make one alternative look substantially more finished.

The research’s professionalism screen determined which ads qualified for particular analyses. It used audience ratings within the same study, rather than independent quality certification.

Record internal checks separately. “Verified by the product owner” and “judged professional by viewers” describe different evidence.

Step 4: Test with your target audience

Start by investigating understanding, then compare alternatives if the decision requires it. Open-ended conversations with a small group can reveal misunderstandings. Choosing between versions requires random allocation, consistent exposure and a sample plan appropriate to the decision. Do not present impressions from exploratory interviews as measured evidence of superiority.

Understanding: Ask suitable audience members to describe the ad in their own words. Interviewers should not explain the intended answer. Ask “What is this ad telling you?” before asking whether it inspires trust.

Comparison: Allocate participants randomly and keep screen, sound, viewing time and question order consistent. Giving only one group extra explanation undermines the comparison.

Useful questions include:

• Understanding: What service is offered, and for whom?
• Promise: What would you expect to receive?
• Recall: Which brand and message do you remember?
• Liking: What appealed to you or bothered you?
• Uncertainty: What else would you need to know?

Plan for a difference that matters to your decision. Do not copy the paper’s samples of 200 and 300 people. Small samples can leave even substantial differences uncertain. Complete unaided recall before showing brand-name options.

Which decisions can a mock-up support?

An AI mock-up can support decisions about message understanding, but it does not demonstrate actual sales effects. Its usefulness depends on the question, draft accuracy and test design. The boundaries below are DijitalPi’s practical interpretation. Preserve uncertainty rather than treating mock-up findings as evidence about the finished film or other markets.

Recall, liking and demand are separate outcomes. Choosing a brand in a questionnaire is not purchasing it. Keep those distinctions explicit in reporting.

Illustrative agency scenario: two openings for a home cleaning service

For a local home cleaning service, pretesting can inform which message proceeds to filming. Avoid combining different openings with different promises, such as a trustworthy team and time saved. Testing the message first, then openings for that message, makes it clearer which creative decision each audience response can inform.

The brand plans a 20-second video. One proposed opening shows cleaners arriving; another shows a customer enjoying personal time instead of housework. A direct comparison would mix differences in message and scene.

The agency first prepares static mock-ups comparing “trustworthy team” and “time saved” within the same visual layout. It then develops video openings for the message taken forward.

The illustrative checks proceed as follows:

• An incorrect uniform is replaced.
• Interviews reveal confusion between one-off cleaning and a subscription.
• Copy is revised to explain the actual booking model.
• Corrected versions are prepared for comparison under consistent conditions.

No comparison result is supplied. A non-significant difference would leave the distinction uncertain, not establish that both ideas are equally good.

How do you carry findings into production and live testing?

Turn the findings into a decision note the production team can implement. Record the tested element, observed understanding, required change and unresolved question. After filming, check that the finished ad preserves the message. Once advertising starts, evaluate campaign performance separately through live testing and conversion data rather than projecting sales from liking.

Use these fields:

• Tested: Version names, changed element and constants.
• Learned: Findings supported by participant responses.
• Change required: Copy, scene or explanation revisions.
• Still uncertain: Unresolved differences and limits of representation.
• Preserve in production: Understood messages and essential brand elements.

DijitalPi recommends briefly checking the finished ad with the same questions. This is a practical suggestion for detecting changes in meaning during filming and editing, not a measured benefit from the paper.

The study detected no overall negative difference in the video condition with advance AI disclosure. However, that group saw only AI films, so the content mix also changed. Do not generalise across disclosure formats; check applicable platform and legal requirements separately.

Sources

Decision cards, checklists and the illustrative scenario are DijitalPi’s proposed practices, not results measured in these sources.

• Bellman, Ourradi and Varan (2026), Scientific Reports: DOI
• Reviewed Article in Press full text (PDF)
• Meta: About A/B testing
• DijitalPi research review: Pretesting ads with AI