Research
Survey researchers have compared chatbots with web forms for years. Here’s what they found, including the study where chat lost, and the design decisions we took from each.
Xiao, Zhou, Liao and others
ACM Transactions on Computer Human Interaction
About 600 people answered open questions either in a chat or in a standard online survey.
Across more than 5,200 written answers, the chat version drew more engagement and answers that were more informative, relevant, specific and clear.
What we took from it
Porto asks open questions in plain language and lets people answer in their own words, without dropdowns.
Kim, Lee and Gweon
CHI 2019
A chatbot survey against a web survey, in a formal and a casual tone.
People in the chatbot survey gave more differentiated answers and were less likely to rush through. A casual tone reduced rushing further, but only in the chatbot.
What we took from it
Every intake is written in the client’s own voice, usually warm and direct.
Barari and others
Survey Research Methods, 2026
1,800 people, with AI assistants that asked follow up questions on open answers.
Follow up questions produced more detailed and informative answers, at a small cost to how much people enjoyed the survey. The AI also tended to read agreement into answers that weren’t clear.
What we took from it
Porto asks at most one follow up per question, and the visitor confirms their summary before it’s sent.
Lucas, Gratch, King and Morency
Computers in Human Behavior
People talked to a virtual interviewer they believed was either automated or run by a person.
Those who believed it was automated worried less about being judged, managed their image less, and disclosed more.
What we took from it
Porto tells every visitor it’s an AI assistant in its first message.
Schuetzler and others
Decision Support Systems
Conversational agents with more humanlike abilities, asking sensitive questions.
The more humanlike agents drew more socially desirable answers to sensitive questions, which means less honest ones.
What we took from it
No invented human names or personas. The assistant speaks for your business, as an assistant.
Comparing chatbots and online surveys
Communication Methods and Measures, 2024
A two week study where chat did worse.
The web survey often produced better data, and people rated the chatbot as less enjoyable, less useful and less secure.
What we took from it
Porto keeps conversations short, shows a privacy note, and is used where open answers matter. For a newsletter sign up, use a form.
Chat tends to get longer and more specific answers to open questions, and people are more candid when they know software is asking. It can also feel slower and less trustworthy if it drags on. So Porto asks five or six questions, follows up only when an answer is vague, says it’s AI from the start, and shows a privacy note before anyone shares contact details.
