Nichola Quail, Founder & CEO of Insights Exchange, recently presented at QRCA Qual Tech Days July 2026. This is an edited summary of her presentation.
Where are you standing when you look at a customer persona?
Not theoretically. Literally. Where in the world are you, and how much does that shape what you consider normal, obvious, or correct?
It mattered before AI. It’s an imperative now. We’re racing toward agentic systems that reproduce shopper journeys, agencies generating global campaign output at scale, and synthetic personas built in minutes. Without a human check, that road leads somewhere specific: the “vanillification” of culture. The more people you ask, the closer you get to beige.
Cultural sovereignty is commercial, not philosophical
Cultural sovereignty is the right of a person to be understood on their own terms – not generated by a model trained on people who don’t look, think, or live like them.
Governments have worked this out. Dr Andrew Charlton, Australia’s Assistant Minister for Science, Technology and the Digital Economy, recently described a “sliding doors” moment: build our own AI capability, or become “a permanent renter of intelligence from abroad.”
In six months, India launched open-source foundation models trained from scratch on Indian language data, France put $830 million behind Mistral’s own data centre as an explicit act of autonomy, and Mexico passed reforms treating the human voice as a protected artistic asset.
That’s the economic argument. The cultural one is ours to make.
The data picture is narrower than most people assume
Roughly 80% of published research participants come from WEIRD populations – Western, Educated, Industrialised, Rich, Democratic. That group is under 15% of the world. Africa is 17% of the global population and under 3% of the training data.
Stanford’s 2026 AI Index tested six chatbots across six language regions. Every model performed worst on Hindi. Not because they couldn’t read Hindi – because when they couldn’t find a Hindi-language source, they substituted an English one. English Wikipedia was the most-cited domain for Hindi queries, outranking every Hindi news outlet.
Hindi has 600 million speakers. Ask yourself what’s happening for Filipino, Spanish, Swahili, or te reo Māori.
If you’re American, you’re standing in the middle of the training data, so this feels neutral. That’s how culture works – you can’t see the water you swim in. But open the same tools in Jakarta, São Paulo or Auckland and what comes back is a US consumer wearing a local costume.
What this looks like in a real brief
We ran a project for a budget airline with one of the most genuinely diverse customer bases you’ll encounter. Every age, every background, every body shape. Families, backpackers, people who saved for months to take their kids somewhere new.
The first round of AI-generated personas came back white, slim, young and beautifully dressed. They looked like they’d walked off a fashion mood board. It took hours of retraining to get personas that resembled the people actually on the plane.
Had that airline trusted the first output, they’d have built a campaign – a product, an entire view of their customer – around an idealised fiction.
“Your competitor can access the same AI tools. They cannot access the specific human truths you surfaced about your customers in their actual cultural context. That’s yours. That’s the asset.”
The insight AI structurally cannot reach
Marcus Collins defines culture as “a system of conventions and expectations that governs what people like us do.” His McDonald’s work shows why human insight still wins commercially.
McDonald’s had enormous transactional data. They knew what millions of people ordered, when and where. What they’d lost was brand love, and the data couldn’t tell them why. Collins’ team spent six months in actual restaurants and surfaced what they called “fan truths.” Your friend takes a fry after saying they didn’t want one. Refilling at the soda fountain feels like living on the edge. The universal pause at the counter — “Can I get… uhhh.”
None of it was in the data. All of it became celebrity meals and a secret menu, and the brand posted 10.3% US sales growth and 11.7% globally in a single quarter.
No prompt would have produced that. Predictive models work from known truths. Fan truths came from sitting with people and seeing the whites of their eyes.
Four things to do now
Put past research to work. Ten or fifteen years of qual with specific communities is proprietary training data. Transcripts and ethnographic archives can fine-tune models so the starting point isn’t the internet average.
Write the cultural brief. You don’t need to retrain a model to change its output. Attach your qual findings to the system so the AI reads them before generating anything about a segment.
Be the cultural QA layer. Before any AI persona enters a client brief, run a structured review. Not a vibe check – a documented process. What did the model flatten? Does a member of this community recognise themselves here?
Stress test with real people. Bring in people from the target culture and ask them to break the output. Not to find offence – to test accuracy. It’s the only reliable way to catch errors that seem right on first reading.
Why Bluey matters
Bluey is a cartoon made in Brisbane and now the number one streaming show in the United States. Its creator wrote every episode from his own experience as a dad. His brother recorded sound in real Brisbane parks and backyards, because those sounds don’t exist in any effects library.
Disney’s first instinct was to strip out the Australianisms and re-dub with American accents. Test audiences rejected it. Joe Brumm said that if he’d listened, “it would have been a show about Bluey going to the dentist, rather than what happens when she gets back and recreates a trip to the dentist.”
An AI trained on US children’s content gives you the dentist episode. It could never have given you Bluey. The cultural knowledge was never in the training set.
“Cultural sovereignty isn’t abstract. It’s a practical question about who gets to decide what your customer looks like, what they care about, and why they buy.”
The human intelligence layer isn’t a replacement for AI. It’s the thing that makes AI outputs trustworthy, specific, and actually grounded in the people they’re meant to describe.
Nichola Quail is Founder & CEO of Insights Exchange, co-founder of AI Marketers Guild APAC, and sits on the advisory board of QRCA and SheisAI.