AI has a cultural blind spot. And it's scaling.

Written by: Joycelyn David

The marketing campaigns you're building with AI tools are more monocultural than you think. Here's the proof.

Last year, a marketing team I know high-fived over their "multicultural" campaign. The creative checked every box: diverse cast, inclusive visuals, a perfect rainbow of faces.

When I pointed out that the copy defaulted to individualistic values, that the cultural references were surface-level at best, that the campaign would land as tone-deaf to the communities it claimed to celebrate – they looked at me like I'd grown a second head. They had let AI write the brief and the copy. Nobody had questioned it. Nobody had thought to.

That's not a technology failure. It's a mindset one.

The hidden cultural bias inside AI

Here is what the research confirms.

Harvard researchers compared the psychological profile of today's top AI models to data from over 94,000 people across 65 countries. The conclusion? AI doesn't think like humanity. It thinks like a very specific, very narrow slice of it.

The framework they used is called WEIRD – Western, Educated, Industrialized, Rich and Democratic. The further your audience sits from those traits, the further the AI drifts from understanding them. Correlation: r = -0.70. Strong. Systemic. Not a glitch.

The training data tells the same story. The original GPT-3 dataset was 93 per cent English. Today, English still represents over 42 per cent of the foundational web data most AI systems draw from. Amharic – spoken by 57 million people – sits at 0.004 per cent. Arabic at 0.64 per cent.

Your multicultural consumers are not meaningfully represented in this data. And the AI you are using to reach them was built on it.

The cultural gap no one is closing in AI marketing

I have watched this play out in boardrooms for twenty years.

A brand tells me they are committed to multicultural marketing. Then they hand me a brief written by ChatGPT, for a campaign where ChatGPT wrote the copy and an AI tool that has never been trained on the culture they are trying to reach generated the creative. Then they wonder why the campaign doesn't land.

A Sept. 2024 study in PNAS Nexus tracked five consecutive generations of GPT models and found that every single one defaults to Western cultural values – particularly American ones – unless specifically prompted otherwise. And even when you prompt it correctly, GPT-4o only improves cultural alignment for 71–81 per cent of countries. At best.

Who on your team is writing those culturally specific prompts? Consistently? For every single campaign touchpoint?

That gap is where your multicultural marketing fails before it ever launches.

This problem requires a new model

Layering AI on top of broken multicultural marketing strategy doesn't fix the problem. It scales the failure.

This is what drove me to build TULONG Technologies. It's also the foundational argument of The Multicultural Mindset: AI will only ever be as inclusive as the humans and the data behind it. Right now, both represent one very specific version of the world.

That version is not your multicultural customer.

What to do right now

Stop briefing AI for demographics. Start briefing it for culture.

Your next prompt shouldn't be "write for Hispanic millennials." It should be built around questions like these:

• Community: Are we targeting first-generation Mexican-Americans in Texas or second-generation Cuban-Americans in Florida? The cultural difference is enormous.
• Values: Does this community lead with collective, family-oriented values or individual achievement? Which one does your brand currently reflect?
• Context: What specific traditions, cultural moments or linguistic nuances need to be present for this message to feel like it was made for them – not about them?
• Exclusion: What imagery, phrases or assumptions would immediately signal to this audience that this message was never really meant for them?

Then – and this is non-negotiable – install a human from that community as a strategic lead in the approval process. Not a final-glance sensitivity reader. Someone with the standing and authority to veto creative that misses the mark. That is your cultural intelligence layer. No AI can replicate it.

The campaigns that earn genuine loyalty from multicultural consumers aren't built on the biggest budgets. They are built by teams that understood culture isn't a demographic. It's a lived experience.

And lived experience is not in the training data. Your marketing team's lived experience is.

Make sure it's the right one.

Joycelyn David is CEO of AVC Communications, founder of TULONG Technologies and author of The Multicultural Mindset, published by Advantage Books. She was named one of the Most Influential Filipinas in the World in 2022.