Research
A creative agency should test whether AI makes its ideas too similar
Judge the variety of a campaign portfolio as well as the polish of individual concepts.
Zocri · 2 min read · Sources checked
Individual quality and collective variety can diverge
Doshi and Hauser's experiment found greater similarity across AI-assisted stories alongside improved individual ratings. The setting was short-story creation, not a commercial advertising test. It nevertheless raises a useful question for a small creative agency: do polished individual outputs leave the overall portfolio feeling repetitive?
An agency can miss this by reviewing each concept separately. Every option may seem competent while all rely on the same tone, audience assumption or visual metaphor. The client then receives several versions of one idea instead of meaningful alternatives for a decision.
Establish different strategic starting points
Before generating concepts, define distinct hypotheses. One might emphasize convenience, another specialist knowledge and another a particular customer situation. Ground each in information from the client rather than requesting a large number of generic variations with no strategic distinction.
Ask a team member to develop an initial direction independently. Keeping that contribution separate until review gives the group another reference point and makes it easier to see whether the assistant is narrowing the range of possibilities.
Review the set together
Lay out the concepts without their polished presentations. Compare the underlying promise, audience and reason to believe. If those elements repeat, changing the headline or image may not create a genuinely different option.
Invite a reviewer to describe the choice each concept offers the client. An inability to explain the difference is a signal to revisit the strategy, not merely to request more variations of the same material.
Test with the intended audience
Select a few meaningfully different concepts for a modest audience test appropriate to the project. Define the response that matters and avoid treating a preference for polished wording as proof of commercial effectiveness. A small exploratory test should inform a decision without pretending to establish a universal winner.
Keep distinctive evidence in the work
Use actual product details, customer language and constraints the client can verify. These provide material competitors cannot reproduce simply by asking a general assistant for ideas in the same category. Review permission and factual accuracy before using customer material publicly.
This is our creative-process recommendation, not a transfer of the experiment's effect size into advertising. The study makes portfolio similarity worth examining. The agency still needs to demonstrate that its concepts differ in a way that matters to the client's audience and business objective.