The contradiction in the numbers
Canva's 2026 State of Marketing and AI report contains the tension marketers should not smooth over: 99% of marketing leaders are increasing AI budgets and roughly 97% use AI daily, while 70% of consumers say AI-generated ads are missing something they describe as soul.
That is not a case for abandoning automation. It is evidence that adoption and audience belief are moving at different speeds. The organization sees efficiency; the audience sees a growing sameness in the work.
eMarketer's 7%/31% trust split and Bynder's 52% disengagement finding point in the same direction. The audience is not asking whether a model was involved. It is asking whether the message contains a reason to trust the brand that made it.

Why polish is not the problem
Most generated ads are not failing because the pixels are visibly broken. They fail because the message could belong to any company in the category. Smooth language, centered products, and a correct brand palette are not the same as an observation nobody else could have made.
AI makes this failure easier to scale. A team can produce hundreds of acceptable variants before anyone asks what the campaign is actually saying. Volume then becomes a way to avoid the harder decision about the point of view.
Good creative can be rough, specific, funny, quiet, or inconvenient. The test is not whether the model can imitate a finished ad. It is whether the team still has a reason for making this one.
Craft is a system
Craft is not a mood board added after generation. It is the system that chooses the observation, the tension, the voice, and the detail that keeps the work from collapsing into category language.
Give the human team explicit control over those choices. Let models help with versioning, transcription, resizing, and exploration. Do not ask them to decide what the audience should care about and then call the result strategy.
Examples from Patagonia or Everlane are useful only as illustrations of a consistent point of view. Their lesson is not that one brand format works forever. It is that the audience can recognize the reason behind the work.

Pair soul with provenance
The adjacent authenticity problem is about proof. The creative problem is about the felt result. Keep them connected: a campaign should disclose synthetic production where required and still carry a clear human source, claim, and accountability.
C2PA Content Credentials can preserve how media was made, but a credential does not turn generic work into meaningful work. It tells the audience about the path; the creative still has to earn the response.
Build a content ledger that records source, human owner, model assistance, approvals, and the claim the ad is making. The record makes revision possible when the audience says the work feels wrong.
The creative test
Before scaling a generated campaign, ask: could a competitor publish the same idea tomorrow, can the creative owner defend the central observation, and is there one detail that came from knowing this audience rather than from a category prompt?
Then test behavior, not only preference. Look at qualified responses, repeat engagement, direct questions, and the quality of the conversations the work creates. A click is not proof of connection, but a pattern of meaningful response is better evidence than a production count.
The soul deficit is not a ban on AI. It is a warning about what happens when production becomes the strategy. Multiply the work after the team decides why it deserves to exist.
A useful boundary
What the system can show
More variants
Faster localization
Lower production friction
AI expands the production surface. The distinction matters because visible activity is not automatically evidence of a business outcome.

