The Anti-Slop Marketing Stack: How Brands Stay Distinct When Everyone Has AI

Every June, thousands of people travel to the south of France to talk about ideas.

This is stranger than it sounds.

Advertising already produces more ideas than anyone can consume. Campaigns arrive every day. New platforms appear. New formats appear. New acronyms appear. Decks multiply in shared drives. Reports are generated, summarised, translated, visualised, and forgotten.

Entire industries now exist to help us create more content than the previous industries created.

Yet people still gather.

They cross oceans. They fill hotel lobbies. They stand in queues for keynote sessions. They argue on beaches and rooftops and crowded terraces overlooking the Mediterranean.

Why?

If ideas are abundant, why are we still meeting to discuss them?

That question feels sharper this year.

For two years, the AI conversation in marketing has been dominated by production.

Can it write?

Can it design?

Can it edit?

Can it plan?

Can it generate?

The answers are becoming difficult to argue with.

Yes.

Mostly yes.

The machines write. The machines draw. The machines edit. The machines produce things that would have looked remarkable eighteen months ago and routine today.

That story is almost finished.

A different story is beginning.

The strange thing about abundance is that it changes what matters.

When information became abundant, attention became valuable.

When software became abundant, distribution became valuable.

When content becomes abundant, something else becomes scarce.

The industry is starting to discover what that thing is.

Contents

• The Anti-Slop Marketing Stack: How Brands Stay Distinct When Everyone Has AI
  • Contents
  • The Question Under Cannes
  • The Problem With Sameness
  • Cannes As A Human Evaluation System
  • The Social Physics Problem
  • The Psychological Value Layer
  • The Invisible Stack
  • Can Taste Be Systematised?
  • If Judgment Is The Bottleneck
  • What The Stack Looks Like In Practice
    • Brand Memory
    • Cultural And Category Context
    • Creative Skills
    • Evaluation Loops
    • Performance Feedback
    • Production Governance
  • What Marketers Should Build First
  • What Engineers Should Build
  • The Quiet Machinery
  • Reference Reading

The Question Under Cannes

You can see the discovery happening in real time.

One conversation is about agents. Another is about workflows. A third is about synthetic media. A fourth is about measurement. A fifth is about whether creative departments are about to become smaller, stranger, or more important.

The topics look different.

The anxiety underneath them feels similar.

People are asking the same question from different directions:

If machines can produce almost anything, how do we decide what deserves to exist?

That question sits under much of Cannes this year.

Not because the industry lacks content.

Because people are worried about judgment.

The advertising industry has always told itself a story about creativity. A good idea appears. The difficult part is producing it. Or distributing it. Or scaling it.

AI has started to expose a different possibility.

Perhaps ideas were never the bottleneck.

Perhaps production was not the bottleneck either.

Perhaps the bottleneck was evaluation all along.

The Problem With Sameness

There is a phrase people now use for low-quality AI content.

Slop.

The trouble is that most people imagine slop incorrectly. They imagine obvious failures: broken hands, dead sentences, plastic smiles, copy that announces itself as machine-made.

The more dangerous version is harder to notice.

It is competent.

The strategy is reasonable. The copy is clean. The image looks expensive. The deck has structure. Nothing appears broken. Nobody objects. The campaign ships.

Three months later nobody remembers it.

This is a failure of distinctiveness.

AI did not invent this process. It accelerated it.

The model has learned from the average of what already exists. The average of what already exists is rarely memorable.

This creates the strange paradox of AI creative work: the better the machine becomes at producing statistically plausible content, the more valuable statistically improbable ideas become.

Most teams still behave as if the problem is output.

Make more headlines.

Make more images.

Make more variants.

Make more edits.

The market does not suffer from a shortage of assets. It suffers from a shortage of things worth remembering.

Cannes As A Human Evaluation System

Consider what Cannes actually is.

On the surface, it is an awards festival. A conference. A week of presentations, panels, activations, dinners, product launches, and carefully curated optimism.

Look more closely.

It is a giant human evaluation system.

Thousands of people arrive carrying campaigns, claims, predictions, frameworks, products, strategies, and theories. Each one hopes to answer a simple question:

Does this matter?

Most ideas disappear.

A few survive.

Even fewer are remembered.

Nobody flies to Cannes because the industry lacks content. They come because attention is scarce. Taste is scarce. Memory is scarce. The ability to recognise a meaningful signal inside overwhelming noise remains scarce.

AI has not changed that.

It may have made it more important.

The same year machines become capable of producing content at industrial scale, the industry’s most prestigious gathering becomes a week-long exercise in deciding which content deserves attention.

Production scales.

Judgment remains stubbornly human.

At least for now.

The Social Physics Problem

Marketing is often described as persuasion.

That makes it sound too individual.

People do not choose brands as isolated rational units. They copy signals. They compare. They seek belonging. They avoid looking foolish. They follow cues. They respond to status. They interpret value through the behaviour of other people.

A product can become desirable because of who uses it, where it appears, what it lets someone signal, or how easily it fits into a story people already understand.

This is where the social physics lens helps. Social physics studies how behaviour emerges from networks, communication, imitation, incentives, timing, and feedback loops. In a market shaped by AI, those loops now include machines as well as people. Agents summarise choices. Recommendation systems frame options. Search surfaces compress categories. Consumers react to that compressed view.

A campaign carries a message. It also changes the social system around that message.

A model can generate a plausible campaign line. It cannot automatically know whether that line will become a social cue, a category memory, a creator format, a retail habit, or another sentence no one repeats.

So the useful questions change:

• Will people remember this in the category?
• Does it give the brand a sharper mental shape?
• Could someone explain it to a friend without reading the brief?
• Does it contain a signal people would repeat?
• Does it match how the audience behaves in groups?
• Does it make a behaviour easier, safer, more desirable, or more public?

Pure generation misses this layer. The system needs behavioural hypotheses before it asks for more creative options.

The Psychological Value Layer

Rory Sutherland’s work is useful here because it refuses to treat value as purely objective.

A faster product is not always more valuable than one that feels easier. A cheaper option is not always more persuasive than one that feels safer. A technically superior feature may matter less than the story that makes it legible.

AI-generated marketing often becomes too literal.

The model explains the benefit. The marketer has to find the perceived value.

The model writes the claim. The marketer has to understand the anxiety, habit, status cue, friction, or category tension around the claim.

The model produces options. The team has to recognise which option creates the stronger mental shortcut.

Some of the best marketing ideas are not cleanly rational. They work because they reframe the situation. They make a choice feel easier, more premium, more obvious, more fun, safer to defend, or more socially legible.

A weak prompt asks:

Write a campaign for our new loyalty app.

A stronger brief asks:

What anxiety does this app remove? What small status signal does it create? What behaviour does it make easier to repeat? What would make someone mention it to a friend? Which category convention can we break without confusing people?

That is where AI starts to become useful. It stops pretending to be the creative director and becomes part of the thinking apparatus.

The Invisible Stack

Walk through Cannes for long enough and a pattern begins to emerge.

Not on the stages. The stages are too polished.

The interesting pattern appears in the gaps: in conversations between sessions, in questions people ask after the presentation has finished, in the quick remarks that do not make the recap posts.

At first the conversations seem unrelated.

An engineer explains agent workflows.

A strategist talks about brand codes.

A founder demonstrates a new creative platform.

A CMO worries about whether all this speed will make the brand softer.

Different industries. Different vocabulary.

The conversations keep circling the same invisible centre:

How do we stop intelligence from becoming average?

Not artificial intelligence.

Intelligence itself.

The web contains an astonishing amount of intelligence. The problem is no longer acquiring it. The problem is turning it into judgment.

Judgment turns out to require several things that are easy to overlook:

• memory
• context
• taste
• feedback
• discipline
• accountability

Strip away the technology and most successful organisations seem to possess some version of these. They remember. They understand where they are. They recognise quality. They learn. They maintain standards. They know who is responsible.

The more you look, the more these pieces start to resemble a stack.

Not a technology stack.

A judgment stack.

A structure that sits above production and decides whether production becomes valuable or forgettable.

The industry spends enormous energy discussing generators. The generators matter. But a generator is only an engine.

An engine does not tell you where to drive.

It does not tell you whether the destination is worth reaching.

Can Taste Be Systematised?

This is usually where the conversation becomes uncomfortable.

Marketing people hear systems and governance and worry creativity is about to be processed to death.

Engineers hear intuition and taste and worry the conversation has abandoned rigour.

Both concerns miss the useful middle.

Taste cannot be automated. It can be supported.

Good judgment cannot be bottled. It can be amplified.

Organisations can build systems that make thoughtful decisions more likely and thoughtless decisions less likely.

Think about how creative teams actually work. The mythology says a brilliant idea appears. Reality is messier. Campaigns emerge from hundreds of tiny judgments.

A planner notices the audience tension.

A strategist kills the obvious line.

A designer refuses the expected visual code.

A creative director senses the idea is close but not there.

A media lead knows the format will punish the execution.

A legal reviewer forces the claim to become sharper.

A client remembers a past campaign everyone else forgot.

The work improves because judgment moves through the group.

AI makes this more important, not less. It can help with production. The highest-value decisions still happen at the intersections: between strategy and creativity, performance and brand, logic and intuition, data and culture.

That is where judgment lives.

And judgment rarely lives inside one person.

It emerges from interactions.

The goal is not to build systems that replace creative teams. The goal is to build systems that preserve and distribute their judgment.

If Judgment Is The Bottleneck

Every technological shift creates a temptation.

People become fascinated by the new capability. Then, slowly, they discover the thing that did not change.

AI is changing production.

Judgment remains scarce.

If that is true, some priorities begin to look different.

The conversation shifts away from prompts, models, and generation itself. A more useful question appears:

What would an organisation look like if it treated judgment as infrastructure?

It starts with memory.

Most organisations possess a fragmented version of themselves. The brand guidelines live in one place. Research lives somewhere else. Performance data lives somewhere else. Institutional knowledge lives inside people.

The organisation knows things.

The organisation struggles to remember them.

Generation without memory produces averages.

Generation with memory begins to produce identity.

One creates content.

The other creates brands.

Then comes evaluation.

Generation attracts attention. Evaluation creates outcomes.

Sophisticated teams do not ask, “Can the model create 100 ideas?”

They ask, “How will we recognise the one worth keeping?”

Once generation becomes cheap, selection becomes valuable.

The market becomes the testing environment. Consumers become the reviewers. Culture becomes the deployment target.

The organisations that learn fastest gain an advantage models alone cannot provide.

What The Stack Looks Like In Practice

The stack is simple enough to name and hard enough to build.

Brand Memory

The system needs to know what the brand has already learned.

That means more than guidelines. It needs positioning, distinctive assets, past campaigns, category codes, voice rules, proof points, claims, rejected ideas, and examples of work that felt wrong.

The negative examples matter. They teach the boundary.

Cultural And Category Context

The system needs to know where the brand is operating.

What does the category reward? What does it overuse? What do competitors keep saying? Which visual codes are tired? Which audience behaviours are changing? Which social signals matter?

A brand can break a code only if it knows the code exists.

Creative Skills

The system needs reusable ways of working.

A skill can capture how the team writes briefs, generates territories, adapts concepts, stress-tests claims, localises work, reviews brand fit, or turns performance data into new creative hypotheses.

This is procedural memory. It stops every person from prompting from scratch.

Evaluation Loops

The system needs to judge before it publishes.

A useful review asks whether the work fits the brand, whether it is distinctive, whether the behavioural logic is plausible, whether the claim is supported, and whether the execution can survive the channel.

The goal is not to automate taste. The goal is to stop volume from drowning it.

Performance Feedback

The system needs to learn from the market.

Connect creative choices to outcomes: hook type, concept territory, brand asset, message frame, audience, channel, placement, spend, review score, conversion quality, qualitative notes.

The feedback loop should produce new hypotheses before it produces more variants.

Production Governance

The system needs a memory of what it has shipped.

Who approved the asset? Which model changed it? Which source material was used? Which claims need proof? Which rights apply? Which markets can use it? Which versions are live? Which were rejected?

Chat history is not a source of truth.

What Marketers Should Build First

Start small.

Build a brand memory pack. Include positioning, voice, distinctive assets, category codes, campaign examples, claims, proof points, and compliance boundaries.

Add bad examples. Add the work that sounds almost right and still feels wrong. That is where the brand boundary often lives.

Build an anti-slop review rubric. Keep it blunt:

Make reviewers explain high and low scores. The comments become training data.

Build a behavioural insight library. Capture patterns people actually follow: risk avoidance, social proof, status signalling, convenience seeking, habit formation, regret avoidance, price anchoring, default behaviour, choice overload, trust transfer.

Connect each pattern to category examples.

Build a creative territory map. For each territory, define the audience tension, behavioural trigger, brand role, proof points, visual codes, tone, channels, and examples.

Then, after each campaign, write down what surprised you.

That is often the most valuable data in the room.

What Engineers Should Build

The anti-slop stack needs infrastructure.

Build retrieval over approved brand materials, campaign archives, research, claims, and performance notes. The agent should cite the source of its guidance.

Treat briefs as structured artefacts. Store the objective, audience context, insight, behavioural hypothesis, offer, claims, constraints, mandatories, and approval state. Version them. Compare them. Learn across them.

Build a creative eval harness:

1. Generate or ingest creative.
2. Retrieve brand and category context.
3. Score against rubrics.
4. Flag risks.
5. Capture human review.
6. Store decision and rationale.

Do not ask one model to judge its own work. Use deterministic checks, separate review agents, and humans for high-impact decisions.

Track asset provenance: model, prompt or skill version, source materials, editor, approval history, rights status, and live placements.

Connect creative metadata to performance. This is how the system learns which ideas travel, which assets carry the brand, and which AI patterns are already overused.

The Quiet Machinery

There is something deceptive about Cannes.

The visible parts attract most of the attention: the films, the stages, the awards, the announcements, the beach activations.

The visible work is rarely the whole story.

Behind every campaign sits a collection of decisions nobody sees. A strategist choosing one insight over another. A creative team rejecting dozens of ideas. A planner recognising a pattern. An organisation learning, forgetting, remembering, and adapting.

The visible work is the surface.

The machinery underneath determines whether the surface matters.

That machinery is becoming more important because production is becoming easier.

For most of modern marketing, production acted as a natural filter. AI is removing those constraints faster than organisations are replacing them.

This is why so many conversations feel unsettled. The technology appears extraordinary. The future appears obvious. Something still feels unresolved.

Perhaps the real question is organisational.

What happens when our ability to create begins to exceed our ability to choose?

For years, the industry worried that AI would make human judgment less valuable. The opposite may be happening.

The easier creation becomes, the more valuable judgment appears.

The easier generation becomes, the more important memory becomes.

The easier production becomes, the more important taste becomes.

Abundance has a way of revealing what was scarce all along.

Perhaps that is the real lesson hidden beneath this year’s conversations.

Machines can create. We know that now.

The more interesting discovery is that creation was never the scarce resource.

The scarce resource was the ability to recognise what matters.

What lasts.

What deserves attention.

That may explain why thousands of people still gather on a stretch of coastline in the south of France.

The industry does not lack ideas.

It is still trying to decide which ideas deserve to survive.

The stages will be dismantled. The banners will come down. Most predictions will age badly.

A few ideas will remain.

The interesting question is not whether those ideas were generated by humans or machines.

The interesting question is whether anyone recognised their value before everyone else did.

That has always been the game.

AI has made the rules easier to see.

Reference Reading

• Cannes Lions 2026 - festival context, programme, communities, and creative effectiveness agenda.
• How to Stop Shipping Low-Quality RL Environments - useful framing for why bad evaluation environments produce bad agent behaviour.
• The Age of Async Agents - relevant to marketing teams building agentic production workflows.
• Social Physics: Uncovering Human Behaviour from Communication - overview of social physics research using communication and population-level behaviour data.
• Social Physics in the Age of Artificial Intelligence - recent framing of hybrid human-AI social systems.
• Creativity Benchmark: A benchmark for marketing creativity for LLM models - evidence that marketing creativity evaluation remains hard and human judgment still matters.
• MindFuse: Towards GenAI Explainability in Marketing Strategy Co-Creation - research direction for explainable AI in strategic marketing workflows.
• Rory Sutherland on behavioural science and marketing - useful background on psychological value, framing, and irrational-seeming marketing effects.
• Social physics discussion - background for the group-behaviour lens used in this article.
• Human behaviour and AI discussion - additional context on what AI can and cannot model about human behaviour.
• Mastercard’s former CMO on AI and marketing - Cannes-adjacent industry perspective on AI, creativity, and sameness.
• Disney’s AI-generated TV ad push - example of AI production tooling moving into mainstream advertising infrastructure.