More AI-generated content is not making campaigns better because the brief feeding the models is weak, not the models themselves. In WARC research with TikTok and LIONS Advisory, 88 per cent of 400 marketers said they generate significantly more creative since adopting AI, but only 45 per cent saw quality improve, and two in three still brief AI mainly with demographic data. Fix the audience brief, with behavioural signals and real audience tension, and output quality has something to stand on.
Key takeaways
- In WARC's survey of 400 marketers, 88 per cent create significantly more creative with AI, but only 45 per cent report any real improvement in quality.
- Two in three still brief generative AI mainly with demographics, although nearly six in ten say demographic segmentation no longer reflects how audiences behave; only 17 per cent consistently brief with behavioural signals, community context or audience tension.
- The human repair layer that used to catch weak briefs has disappeared, and an AI model will polish a poor concept as readily as a strong one.
- In Southeast Asia, where one team may cover six to eleven markets, demographic briefs erase the local nuance that is the whole job.
- The teams pulling ahead treat the brief itself as the product and apply judgement before an asset ships.
Is the bottleneck the AI or the brief?
Marketing teams across Southeast Asia are producing more creative assets than ever before. Nine out of ten say generative AI has multiplied their output. Yet fewer than half believe that flood of content is actually performing better. Something in the middle of the pipeline is quietly breaking, and it isn't the models.
New research from WARC, conducted alongside TikTok and LIONS Advisory and reported by PPC Land, surveyed 400 marketers and uncovered a striking pattern. While 88 per cent of respondents said they are generating significantly more creative since adopting AI tools, only 45 per cent reported any real improvement in quality. Output climbed. Results stayed flat. The problem sits somewhere most teams never think to look: the audience brief.
Why do marketers keep using data they know is outdated?
The most revealing detail in the WARC study isn't the output gap itself. It's what marketers admitted about their own inputs. Two out of three said they still brief generative AI models primarily using demographic data such as age, income or location. At the very same time, nearly six in ten of those marketers said demographic segmentation no longer reflects how audiences actually behave.
In other words, teams are knowingly feeding AI systems the exact inputs they have already identified as broken. Only 17 per cent said they consistently brief with something more meaningful, such as behavioural signals, community context or genuine audience tension. WARC calls this an intelligence gap rather than a technology gap, and that framing deserves to be repeated often, because it moves the entire conversation away from tools and back toward strategy.
What happened to the human quality check?
For decades, a creative brief was never treated as a finished document. It functioned more like the opening line of a conversation. A strategist would question the segment. A creative director would ask what the customer actually feels in the moment of decision. An art director would discard the first few concepts entirely. By the time a campaign reached a client, a thin or lazy brief had usually been reworked several times by people whose job included spotting weakness before it went live.
That process was deliberately slow, and the slowness served a purpose. It was also costly, which made it one of the first casualties when production timelines shrank from weeks to days. As that human repair layer disappeared, briefs stopped passing through a chain of sceptical professionals and started flowing directly into generative models instead.
That shift matters enormously, because an AI model has no instinct to question a weak input. It will produce forty polished, confident variations of a poor concept just as readily as forty strong ones. A weak brief used to result in visibly weak creative that someone would catch before launch. Now it results in polished, professional-looking content that slips through untouched, simply because it looks acceptable on the surface.
Why do teams keep defaulting to demographic targeting?
If marketers already know demographic briefing underperforms, why does it remain the default choice? The honest answer isn't conviction. It's inertia.
Age brackets, income bands and broad location data are already sitting inside the planning deck, the media plan and the campaign naming structure. Reusing them requires no additional research, no new tooling and no uncomfortable conversation with a client or colleague. Behavioural and community-based insight demands all three. Under tight deadlines, the path of least resistance almost always wins, and that path is demographic shorthand.
GWI made a similar point recently, noting that age-bracket targeting is one of the laziest segmentation methods in marketing, since the differences within a single generation are frequently larger than the differences between generations. It's a fair criticism, and one that is rarely said out loud. Scan a week of marketing technology commentary and a clear divide appears. Audience research companies tend to stop at the statistic. Workflow and social listening platforms tend to start at the publish button. Very few players address the space in between, which happens to be exactly where this intelligence gap lives.
Why do Southeast Asian marketing teams feel this first?
Regional compression makes this problem especially acute in Southeast Asia. A team based in Singapore might be responsible for six to eleven separate markets, across multiple languages, working with budgets that would barely stretch across two markets in Western Europe. Local nuance isn't a nice bonus in that environment. It is the entire job, and it happens to be exactly what demographic briefing erases.
Feed a generative model a brief built around urban women aged twenty-five to thirty-four with middle incomes, and it will produce content that could technically run in Jakarta, Manila or Kuala Lumpur, yet fully connect in none of them. The result will read as grammatically correct, on brand and completely forgettable. Repeat that pattern across eleven markets on a weekly publishing schedule, and the outcome is an extremely efficient system for producing content nobody remembers a week later.
What does treating the brief as the product mean?
The marketing teams currently pulling ahead have made one deliberate structural change. They now treat the brief itself as the actual product, not the finished deck and not the final asset. Whoever controls the quality of that early input effectively controls the quality of everything that follows downstream.
This shift is already visible in how tools at both ends of the production pipeline are evolving. At the front end, newer platforms are translating live category and behavioural data into evidence-backed audience tensions, ensuring the brief carries something genuinely observed rather than something simply assumed. At the back end, a fresh category of execution-focused businesses is emerging as well. One example is Touchigh, a company helping Chinese cross-border sellers reach American buyers by combining AI search visibility tools with native English social content, then scoring that content for predicted performance before it ever publishes rather than only reporting results afterwards.
That detail matters more than the specific industry it comes from. A small execution-focused company serving price-sensitive customers has independently reached the same conclusion as much larger research firms. The critical judgement call needs to happen before an asset ships, or realistically, it will never happen at all.
Can speed and quality coexist?
What the current numbers suggest is genuinely encouraging. The repair layer marketing teams lost during the AI adoption rush can be rebuilt affordably enough to survive even the tightest deadline. When agency SAMY pitched against considerably larger competitors for a UK hospitality client, category planning that once took several weeks was compressed into roughly thirteen minutes per brief, with predicted click-through accuracy running about three times more reliable than traditional human estimation. SAMY won the retainer.
The real story in that example is not the speed itself. It's the fact that a level of judgement teams previously couldn't afford to apply consistently is now something they can apply to every single brief they write, not only the ones with generous timelines. A judgement call that only happens when there's spare time available was never really a standard to begin with. It was a luxury few teams could reliably afford.
What should every marketing team ask about its briefs?
The uncomfortable truth buried inside the WARC findings is that many teams already recognise their inputs are flawed and choose to ship anyway, simply because correcting the brief costs more time this week than shipping mediocre output does.
That calculation is shifting quickly. As evidence-grade audience intelligence becomes achievable in minutes rather than weeks, the justification for briefing generative AI with outdated demographic shortcuts is disappearing, and so is the excuse when that content underperforms.
So consider the briefs your team produced this quarter. How many were genuinely traced back to something real you observed about your audience, and how many were simply last quarter's template with a new date stamped on top?
Where does SOMIN fit into this shift?
This is precisely the gap SOMIN was built to close. SODA turns scattered category and consumer signal into structured, evidence-backed intelligence that a brief can actually stand on, while the SoMonitor suite, spanning Brand Tracker, Content Library, Perspective Studies and SoInspire, keeps that intelligence connected across brand health, published creative, audience perspective and emerging cultural moments rather than leaving it siloed in separate reports. Paired with deep audience data from partners like GWI, marketing teams get something closer to a living brief than a static one, grounded in what audiences are actually doing and feeling right now rather than a demographic bracket assumed months ago. For teams trying to fix the intelligence gap before it reaches the model, that combination of observed data and connected tooling is exactly where the repair layer gets rebuilt.
FAQ
Why isn't AI improving creative quality?
Because most teams feed it weak briefs. In the WARC study, 88 per cent of marketers produce significantly more creative with AI but only 45 per cent see better quality, and two in three still brief the models mainly with demographic data.
What should you brief generative AI with instead of demographics?
Behavioural signals, community context and genuine audience tension: something observed about the audience rather than assumed. Only 17 per cent of marketers in the WARC study consistently brief this way.
Why do teams still brief AI with demographic data?
Inertia. Age brackets, income bands and location data already sit in the planning deck and media plan, so reusing them needs no new research, tooling or difficult conversations, while behavioural insight needs all three.
What is the intelligence gap in AI marketing?
It is WARC's name for the problem: AI can produce far more creative, but the audience understanding fed into it has not improved. That makes it a strategy problem rather than a technology problem.
Related reading
- AI content homogenisation: why every brand sounds the same
- Why do ads in one category all look the same?
- Creative brief template: persona, tension, angle, story
- How do you turn a social listening insight into a campaign brief?
- What is a consumer tension in marketing, and how do you find one?
