Synthetic personas: helpful or harmful?

This post was written October 2024

The arrival of publicly accessible Artificial Intelligence has opened both new opportunities and challenges.

One of the more recent tools that have hit the branding, marketing, and advertising sectors is Synthetic Personas, aka virtual personas. In this article, we discuss if they are any good and the relevance and quality of these AI-generating tools.

Understanding your audience is seemingly easier than ever if you are creating or managing a brand. A trail of reviews, comments and subtle recommendations guides our every online interaction. A record of many human interactions is available on social media. If you’re conducting large-scale surveys or feedback, getting the data be that Quant (numbers-based) or Qual (behaviour-based), used to be costly; today, it’s much easier with online tools.

With ever-increasing amounts of data swilling around, filtering through the volume of data and building up accurate audience personas can be a time-consuming challenge. However, scraping succinct summaries from available information is catnip to machine learning, and AI start-ups have stepped in to make the whole process easy and instantaneous. A number of online tools allow you to (after providing them with significant audience data) not just create detailed synthetic personas – but then to ‘interview’ them to get their opinions on brands, test product ideas, or get a sense of the competitive landscape.

We’ve tried a number of them, and the results are helpful. They can be surprisingly accurate in terms of defining many of the pragmatic aspects of a persona. They can quickly draw up pen portraits and reveal information that might be difficult to discover without deeper research into your data.

Once defined by AI and then checked and edited by a human, the ability to use them as a virtual sounding board gets interesting, especially if it’s difficult to speak directly to potential customers or clients. As Michelle Greenwald notes in her Forbes article, “Synthetic personas, powered by the latest LLMs from Google and Open AI mimic the sensibilities of real human personas and can help brands test and optimise creative for different target audiences, reduce backlash risk, and potentially increase the chances campaigns will go viral”.

However, there is still an underlying feeling that you are essentially interviewing a machine. And that’s the main problem with synthetic personas; there’s a sense that they lack the randomness of a real human subject — and that’s potentially problematic as randomness is inherent in how we think, interact and behave.

Anyone who’s sat in on focus groups or customer interviews will appreciate that what’s unsaid is often as important as what’s actually discussed and that people’s opinions shift greatly depending on the context, the direction of the conversation or even who else is in the room. For example, although lots of research shows that the majority of millennials declare that their purchase decisions are driven by sustainability; data reveals that the vast majority buy based on a far simpler metric: price.

It’s not just the weirdness and contrariness of humans that plays a key factor in defining personas and undertaking research. The uncertainty of culture also plays an enormous effect on how we behave or what we believe. When we take into account unforeseen shifts and the adoption of trends from far outside the mainstream conversation, it’s clear that brands exist in a world that is not only chaotic, it often bears little resemblance to what’s gone before.

Recent high-profile missteps by major brands illustrate this challenge (Greenwald cites examples such as Apple’s Crush! iPad Pro campaign and Bumble dating app’s anti-celibacy advert). These campaigns — despite presumably being vetted through traditional channels of research — resulted in significant backlash and public apologies from the companies involved.