Can you rehearse an advertisement with AI before filming?
Your team likes an advertising idea, but you do not want to discover after filming that customers misunderstood the product. Ad pretesting examines audience responses before publication. Can an AI-produced rehearsal support that decision? This study detected no significant overall differences in measured advertising responses between real and AI ads that passed a quality screen. DijitalPi commentary: Consider a human-reviewed rehearsal for investigating message understanding. The finding does not prove production methods equivalent, validate every AI output or establish sales gains and budget savings.
ACADEMIC WORK REVIEWED
Pretesting with AI-generated static and video ads
ResearchersSteven Bellman, Sid Ali Ourradi, Duane Varan.
Scientific Reports · 15 August 2026 · Peer-reviewed, accepted Article in Press; awaiting editorial corrections · Source language: English
English review: 26 September 2026 · The source date appears in the citation above.
Editorial review draft
The source has been checked; the review still carries its original editorial status. DijitalPi did not conduct a customer experiment for this paper.
LET'S READ THE RESEARCH TOGETHER
We first explain the researchers' question, method and findings. We then discuss how to interpret the results, clearly separating DijitalPi's commentary from the source.
01 / WHAT DID THE RESEARCHERS WANT TO UNDERSTAND?
Can an AI advertising rehearsal support a preproduction decision?
The researchers examined pretest responses to AI-produced counterparts of existing advertisements. AI reproduced visuals and films rather than selecting a new marketing strategy. They also examined telling viewers about AI production in advance.
Explaining the context · DijitalPi commentary
DijitalPi commentary: The preproduction question extends beyond which image people like. Do viewers understand the offer, intended customer and product promise? A rehearsal may inform those questions without validating the final production’s response or the campaign’s future sales.
02 / HOW WAS THE RESEARCH CONDUCTED?
How did they test the question?
The researchers conducted 2 US online-panel experiments with random assignment. Reported samples were 200 people for static ads and 300 for video ads. The static experiment used ads for 8 brands; the video experiment initially used films for 9 brands. AI production involved human intervention, and effectiveness analyses included ad pairs that passed a professionalism assessment.
Understanding the method · DijitalPi commentary
For static ads, other elements such as logos and copy stayed fixed, and AI generation was not disclosed beforehand. In the video experiment, 2 groups watched a mixture of real and AI films; another watched only AI films with advance disclosure. The disclosure comparison therefore changed the content mixture as well as the information given to viewers.
Professionalism was rated by viewers within the study, not certified by an independent quality body. The video experiment excluded 1 ad pair. Human editing corrected continuity across video scenes. Findings cannot be generalized to all raw, unchecked AI outputs.
03 / RESEARCH FINDINGS
No significant difference does not prove equivalence.
Main analyses of screened ads detected no significant overall group differences on 5 measures: unaided brand recall, brand recognition, hypothetical brand choice, brand attitude and ad liking. There was also no overall negative difference in the advance-disclosure video condition. The authors state that their samples were insufficient for equivalence testing. Choosing a brand in a questionnaire is not a purchase; campaign revenue and return on advertising spend were not measured.
DijitalPi commentary: Consider a rehearsal first as a way to identify unclear messaging or mistaken product expectations. Keep product facts, viewing conditions and questions consistent. Plan participant numbers around a decision-relevant difference instead of copying the paper’s sample size; failure to detect a difference should not become the success criterion.
Quality screening, informational advertising and the viewing arrangement define the scope of the finding. It does not automatically extend to other languages, emotional storytelling or disclosure formats. Rechecking the final production after choosing a rehearsal is our application proposal, not a measured sales benefit from the paper.
05 / CONCLUSION AND OPEN QUESTIONS
What did we learn, and what do we still not know?
The reviewed text is a peer-reviewed, accepted Article in Press awaiting editorial corrections. Inconsistent secondary sample and table descriptions were not used to derive new rates. MediaScience funded ad production and data collection; Ourradi and Varan work for the company, which also partly supports Bellman’s salary. The statement that Bellman analysed the data independently is not independent replication by another team.
Our open question at DijitalPi: which concrete decision should your Turkish advertising rehearsal change, and what will you check again in the final production? The examples below are original illustrative scenarios; DijitalPi conducted no advertising experiment for this review.
We created these scenarios to make the topic concrete. They are not cases from the paper or measured client results.
EXAMPLE 01
Furniture: an attractive image, a wrong size expectation
An illustrative table manufacturer’s AI rehearsal makes the room spacious and the table smaller than its actual dimensions.
Does the viewer understand where the table will fit?
Open the recommendation for example 01+
Verify dimensions, then investigate expectations.
Check the image against actual measurements. Ask the target audience which spaces suit the table and what information they need before buying. The aim is to detect a mistaken expectation; do not promise fewer returns or more sales before measuring them.
EXAMPLE 02
B2B software: an advertised feature does not exist
An illustrative quotation tool’s AI video shows an automatic approval step that the product cannot perform.
Is a favourable response about the real product or an inaccurate promise?
Open the recommendation for example 02+
Map every scene to a real capability.
Have a product owner review the scenes. Ask viewers to explain the workflow in their own words. Assess message understanding, purchasing authority and integration fit separately. Do not count a questionnaire response as an enquiry or signed contract.
EXAMPLE 03
Retail: promotion conditions disappear
An illustrative store’s rehearsal makes a discount on selected products seem available across the entire store.
Does the viewer understand the conditions at checkout?
Open the recommendation for example 03+
Check offer understanding before judging aesthetics.
Ask participants to explain which products qualify. Revise the misleading scene or wording and reuse the questions. Attention, offer understanding and completed purchases are distinct outcomes. This scenario is not a client case from the paper.
TRY IT WITH YOUR TEAM
Write down the decision your rehearsal should support.
Decide whether you are comparing messages, presentation or production methods. Keep product facts, viewing conditions and evaluation questions consistent. Plan the sample around a decision-relevant difference. This illustrative pilot has not been conducted.
The academic work belongs to the researchers named above. This page contains DijitalPi's explanatory review and original business examples; it is not a full translation of the paper.
Cookie preferences