How do you use AI to create content without losing your brand's voice or your agency's standards? Stop treating AI output as finished work. Put human judgement at fixed points in the process, from the brief to the final sign-off, test concepts before you pay to run them, and judge the result on business evidence rather than on how many assets you shipped.
The commercial stakes have gone up. In a Thought Leaders piece for LBBOnline, the team at Nomix Group argues that AI is no longer a production cost sitting at the edge of advertising. Its cost is already being negotiated inside the commercial terms that govern media buying, with some holding companies absorbing AI infrastructure costs in exchange for client commitments to principal inventory. The conclusion for creative teams: the brief itself now has to say how AI will be used, governed and judged.
Why does so much AI content feel interchangeable?
Because polish has become free. For decades, a well-lit, well-written, perfectly resized asset signalled that someone had invested time and skill. Now any team can produce a hundred of them before lunch. When everyone can afford polish, polish stops telling the audience anything.
The result is what the industry has started calling slop: content that is technically fine and instantly forgettable. It tends to fail in three ways. It is generic, because the prompt was generic. It is ownerless, because nobody made a real decision about it. And it is unproven, because volume was mistaken for effectiveness.
What should an AI-era creative brief include?
The LBB piece offers a useful checklist. A brief, it argues, should now cover:
- where AI can be used;
- which decisions stay human;
- how assets and data will be reviewed;
- what disclosures apply;
- how media incentives are disclosed;
- what evidence will judge the outcome.
The last point matters most for quality. The piece tells brand marketers to judge AI work on business evidence and creative standards, not speed alone. It also points to the European Commission's July guidance on AI transparency, which sets expectations for marking and labelling AI-generated or manipulated content. If you cannot say which parts of an asset a model touched, you will struggle to meet that bar.
The same piece cites Skechers, which moved media across 31 markets to a single partner combining audience strategy, data, activation and reporting. Integrated systems at that scale produce content at that scale. Without quality gates, they produce slop at that scale too.
A seven-gate quality framework for AI-assisted content
Quality slips when the only human check is a read-through at six o'clock on deadline day. It holds when judgement sits at fixed gates, each with a named owner.
| Gate | The question | Owner | It fails when |
|---|---|---|---|
| 1. Brief | Where may AI be used, and what stays human? | Strategist and client lead | The brief says "use AI" and nothing else |
| 2. Evidence | Is the input real audience and competitor insight, not a model's guess? | Strategy or insight lead | The prompt is the brief's first paragraph, pasted in |
| 3. Voice | Does every tool have the brand's actual tone, claims and banned words? | Copy lead | The output sounds like every other brand in the category |
| 4. Edit | Has a person added something specific: a detail, an opinion, a joke? | Creative lead | Edits only fix grammar |
| 5. Pre-score | Is this concept likely to earn attention before money goes behind it? | Creative and media leads together | Testing starts after launch |
| 6. Record | Which parts did AI generate or alter, with which tools and data? | Producer | Nobody can answer when the client asks |
| 7. Outcome | Did it move the business metric named in the brief? | Account lead and client | Success is reported as an asset count |
Why do imperfect details read as authentic?
Self-determination theory, from psychology, says people are driven by three basic needs: autonomy, competence and relatedness. It explains both sides of the AI content problem.
On the audience side, relatedness is the giveaway. People connect with content that feels as if a person made it for them. The oddly specific detail, the opinion someone might disagree with, the joke that only works in one city: these are the cracks that tell the viewer a human was involved. Perfectly smooth content has no cracks, so nothing gets in.
On the team side, autonomy and competence matter. A creative reduced to approving machine output has little reason to care about it, and that indifference shows in the work. A creative who uses AI to explore more routes, faster, and then applies their own taste to the best of them keeps the ownership that good work depends on. Authenticity thrives in the cracks of perfection, and it is people who put the cracks there.
How do you build the gates into your workflow?
- Rewrite your brief template. Add the six LBB questions as mandatory fields. A brief without them goes back to the client lead.
- Build a voice pack. Real examples of the brand at its best, approved claims, words it never uses. Feed the same pack to every tool your team touches.
- Start from evidence, not prompts. Ground concepts in what the audience and competitors actually publish and respond to.
- Make one person accountable per asset. Their name goes on it, which changes how carefully it gets edited.
- Pre-score before you spend. Shortlist the strongest concepts on predicted performance, then put media behind the winners.
- Keep a simple AI log. Tool, data, what changed, who approved it. A shared sheet is enough to start.
- Report outcomes, not output. Drop asset counts from the monthly report and lead with the business metric from the brief.
Can you know an AI-assisted concept will work before you pay for it?
Not with certainty, but you can know far more than you used to. Pre-scoring, gate five, is where most agencies have the biggest hole. Traditional pre-testing was slow and expensive, so it was saved for big campaigns. AI-assisted production now generates more concepts than any research panel could review.
Defy Monitor, part of the Defy platform, scores creative concepts for predicted clickability before media spend, and that scoring is validated against real ad performance. It does not replace the creative director's judgement on the idea or the voice. It does stop the weakest options reaching the media plan, which is where slop gets expensive.
