Synthetic interviews - how AI is changing marketing and recruiting

When AI becomes a conversation partner

Interviews with synthetic users or candidates are currently generating discussion. After all, research thrives on real encounters: people, their emotions, idiosyncrasies and surprises. Only by observing and interviewing them can you understand what drives them - and what slows them down and perhaps even stops them from making a purchase or applying.

But traditional interviews are time-consuming and expensive. So it's no wonder that generative AI tools entice with great promises: fast, cheap, seemingly limitless. A temptation for marketing and recruiting.

What if real respondents were simply replaced by AI personas? What if they were even based on artificially generated data? Sounds convenient - and it is. But the result remains flat.

Synthetic data - easy to make, difficult to evaluate

Personas must be based on data. These can be real or synthetic. The latter are not created from real sources, but from patterns that AI derives from existing information in the real world.

The advantage is obvious: synthetic data is fast, inexpensive and privacy-friendly. It can be used specifically for niches or to compensate for distortions in real data sets.

Today, generative AI and synthetic data are also used to develop artificial personas. The process is simple: First, you describe the target customers, define their goals, needs and challenges. Then you add the right product or job opening to solve these problems. In this way, countless personas are created in no time at all, mapping target groups, imitating their behavior and responding to surveys or interviews - seemingly ideal for accelerating design and marketing processes.

But the method has its limits. We go into this in detail in the blog article "Creating personas with AI - the check". The most important weaknesses of personas from synthetic data at a glance:

• They reflect clichés and prejudices that are widespread on the Internet.
• Groups that are large in the real world but barely represented online - such as rickshaw drivers, agricultural workers or housewives - are portrayed in a distorted and stereotypical way.
• For genuine innovations - i.e. products, concepts or target groups that do not yet exist - the model fails because there are no patterns on which the AI could build.

Synthetic interviews - efficient, but soulless

Real interviews require time, planning and intuition. With synthetic interviews, the AI takes over the counterpart - via text dialog or survey. This can be useful, for example

• when testing unusual ideas and hypothetical future situations
• for sensitive topics for which there are hardly any real participants,
• or if hard-to-reach target groups are to be surveyed.

Synthetic interlocutors don't get nervous, they don't contradict, they always answer. But they remain superficial - average in the best sense of the word.

As early as 1998, researchers at Carnegie Mellon University described such "synthetic interviews": virtual figures that responded to user questions.

In these early experiments, users were to receive answers to their questions in an intensive dialog - from virtual conversation partners who mimic human behavior as realistically as possible.

To do this, the researchers recorded thousands of video clips with actors: both for potential responses and for non-verbal signals such as gestures and facial expressions. The computer-generated personas then appeared in the so-called "talking head" format and showed typical characteristics in order to appear credible.

Users could ask their questions using a microphone or keyboard. Voice recognition analyzed them and provided the appropriate answers. The technology was even designed to enable interactive experiences with celebrities, scientists or religious leaders - laying the foundation for realistic virtual characters.

Create AI interviews yourself

Today's AI makes it much easier to conduct synthetic interviews. If you want to try it out despite the known disadvantages, all you need is access to one of the common LLMs - such as ChatGPT, Gemini or LLaMA - and a Word or PDF document with the profiles of the target persons.

This should include key characteristics: Age, gender, location, professional background, purchasing behavior, pain points, motivations, triggers and media usage. Equally important: a clear objective. What should the interview or roundtable achieve?

You can also specify how many synthetic personas you want to create and in which format the answers should be available - as continuous text, interview excerpts or survey results.

Then you can get started: Upload your document with the target group information, add the product description or the position to be filled and define the desired number of personas. Then ask them your questions - and at the end, have a report created that summarizes key problems, concerns, positive aspects and concise quotes.

The result is available in just a few minutes and can be quite informative. However, it often remains superficial. Many synthetic interview partners provide almost identical answers - factual, unemotional, predictable. This is not enough for in-depth insights.
But: such tests help to develop better questions for real user interviews.

The best of both worlds

data-driven personas - such as those from the Persona Institute - are also created with the help of AI, but not by prompt and not on the basis of synthetic data. AI analyzes huge data sets for us that would be almost impossible to manage manually. It recognizes patterns in these volumes that humans would overlook and derives specific recommendations for action.

Precise personas are created from valid data sets - developed with AI, but under human supervision and control. The result is realistic profiles that make target groups tangible.

And best of all, AI helps us to bring these personas to life - as living chatbots that you can talk to directly. They conduct interviews, take part in focus groups or advise you on design issues. If you prefer talking to typing, you can also talk to them.

In short: personas from real data, created with AI - for realistic, synthetic interviews.

Our personas are based on representative data from over 12 countries and 1,000 markets. They are based on more than one million scientifically collected surveys and just as many statistics.

These profiles provide far more than just demographics. They show who your target group really is - in all its facets: Living situation, income, media use, innovativeness and attitudes towards topics such as health, finance, travel, nutrition or social behavior. You will also find out how your persona assesses their economic situation, what and where they shop, what drives them and what inhibits them.

Each persona comes across as authentic - almost like a good friend that you really know. And why not make this friend your colleague?

Thanks to our AI-powered persona bot, you can chat, talk and involve your persona in your work:

• For recruiters: Have a conversation with your candidate persona about the ideal application process, no-gos and favorite job ads.
• For product developers: Ask your buyer persona what they expect from the product, how they use it and which design appeals to them. If you have several buyer personas for a brand, you can even conduct synthetic focus group interviews with them - because they are data-based on real data and are therefore not "fishing in the dark".
• For B2B sales: Consult your entire buyer center and develop your sales strategy on this basis.

These conversations are synthetic - but surprisingly real. Thanks to millions of real data points, the answers seem credible, emotional and relevant.

Combining the best of two worlds: the precision of human research with the speed and scalability of AI.