Helping Hand, or Fraudulent Foe? AI in Market Research

(In market research, and just about everywhere else, we can’t stop talking about AI. It’s the tool that’s taken over the world for good…and for bad.)

AI has been embroiled in controversy since its founding. From safety issues surrounding privacy and accuracy, to moral issues concerning ownership and the environment, the use of artificial intelligence is a hotbed for argument. Market research is no different- since the first AI chatbots came onto the market, market researchers have explored how to deal with AI use in the industry, both positively and negatively. 

AI Respondents

We’ve talked a little about synthetic respondents, one of the most prominent and well-known AI use cases in market research. As a refresher, synthetic respondents, otherwise known as digital respondents or synthetic sample, are AI-generated respondents, fed with data through an LLM and built to model certain demographics of consumers. However, another form of artificial intelligence replicating, and sometimes replacing, respondents are the synthetic personas. 

Personas, like synthetic respondents, are LLM-generated respondents meant to embody certain demographics. However, unlike synthetic respondents, synthetic personas do not represent a group of people taking a survey, but rather a single person who fits into a wider demographic. Personas also do not typically take surveys to do the necessary research, but rather exist as a way to summarize the research, by being able to ‘speak’ directly to a researcher or client about how they ‘think’ and ‘feel’. 

Both use cases have their proponents and their detractors. Supporters of digital respondents praise the rapid research times at a fraction of the cost of real respondents, and champions of synthetic personas celebrate their scalability and robustness over traditional personas. Detractors of both cite issues of accuracy, bias, AI hallucinations, and the morality of replacing real consumers with AI models. In the market research industry, a general rule of thumb is to never use entirely synthetic respondents and to always verify the results against actual respondents. However, that doesn’t stop some companies from pushing all-AI data as the stuff of the future, for better or for worse.

AI in Surveys 

The biggest use of AI in surveys is in fraud detection and reduction. AI has been used in market research to detect, flag, and catch cheaters, bots, and other survey fraudsters. AI-based solutions exist that can match a respondent’s IP address with their claimed location, detect VPNs and proxy servers, and even detect and delete accounts with fake email addresses and phone numbers. These platforms detect and defend against duplicate accounts, bots, fraudulent users, and click farms to get better, cleaner data for market researchers. 

Artificial intelligence has been used extensively in creating surveys, too. Several companies boast the ability to generate a full survey in minutes with the power of generative AI. These services have AI write the necessary questions, format the survey, and sometimes even code it, saving researchers time and money. However, these surveys often lack the depth of handwritten surveys, and sometimes come with poorly written code. 

Of course, AI isn’t just used by the researchers: plenty of panelists and survey participants use AI, too. Some use it entirely dishonestly, using AI to research topics they know nothing about in order to receive the payout at the end of a survey. More use it as a way to answer open-ended questions, writing their own thoughts into a chatbot and having it enhance their answers to better respond to the question. Though they may come from a good place, these practices hurt respondents and researchers more than they help. AI-written responses cannot be verified as coming from the respondents themselves, and so are usually tossed out, flagging the respondent and sometimes even removing them from the panel entirely. Many survey platforms have already taken to immediately disqualifying any open-ended responses that are copy-pasted in order to discourage the use of AI to answer survey questions.