Walr conducted a survey with 2,000 nationally representative respondents in the US, exploring attitudes towards AI and autonomous driving.

Half of the respondents were issued with a survey containing standard, neutral wording and scales, while the other half used a more conversational survey design with natural language. Both groups passed identical fraud and quality controls.

The A/B test found that the experimental conversational survey’s responses scored higher for data quality.

For example, one question, focused on trust in autonomous vehicles, saw ‘junk’ responses fall from 19.0% in the standard survey design to 8.4% in the conversational design, while mean length of interview increased for the experimental design, according to the paper.

Additionally, 53% of respondents to the conversational survey said they would take another survey, compared with 46% of respondents to the standard version.

Staci Kinney, vice-president of data consultancy at Walr, said: "We've spent years getting better at one question: is this a real, engaged human? This research asks a second one. Once we know we're talking to a real person, how hard are we working to talk to them like one?"