Build with AI
How I helped SurveyMonkey use AI to make it easy for anyone to create a survey
How I helped SurveyMonkey use AI to make it easy for anyone to create a survey
Build with AI is a new way to create surveys, and it's powered by OpenAI. The user types short instructions. Then AI generates a survey accordingly. It's fast, accurate, and results in high-quality surveys.
It’s hard to write a good survey. You have to know what questions to ask, how to ask them, and in what order. And even if you can do all those things, writing a survey can take a lot of time. How could AI help?
Build with AI is for all survey authors, novices and experts alike. Even people on free plans can use it, although they need to upgrade if they want to include premium question types in their surveys.
We wanted to see a high number of customers deploying (sending) surveys with this feature. People who deploy tend to stick around. And we hoped to increase conversions from free to paid accounts.
As the sole content designer, I had several key responsibilities:
Identify content needs and opportunities.
Name the feature.
Write guidance copy and microcopy.
Test content for usability.
I conducted the following foundational research:
Conversation mining
Competitive analysis
Prompt engineering exploration
Here’s my key takeaway: Nearly all GPT products offered sample prompts to help users get started. The product designer and I discussed these findings. We agreed that prompt writing would be our key challenge.
A prompt engineering tutorial I completed during my research
Flows captured during competitive analysis
I organized a feature naming workshop with cross functional stakeholders. Our activities were as follows:
Align on criteria for the name.
Brainstorm names according to those criteria.
Choose three of our best options.
Then I surveyed 200 people and asked them which of those three names they liked best. Names appeared in random order to each respondent to prevent bias. I announced the winner to stakeholders and started working it into the product.
FigJam voting widgets we used to align on naming criteria
A screenshot from the survey
Ideas from our naming workshop
I knew from my research that many users wouldn’t be able to write prompts without guidance. So I built on the idea of sample prompts from other products.
Novice users could use the prompts without making changes.
More articulate users could edit the prompts to suit their needs.
Savvy users could model their own prompts after our samples.
I worked with one of our survey methodology experts to develop and test the right prompt structure.
Sample prompts I developed and tested with a survey expert
I tested a prototype with five real users to answer these questions:
Will people know what a prompt is?
Is the purpose of the sample prompts clear?
Does the term “sample prompt” make sense?
Participants attempted tasks like using one of the sample prompts and defining the word “prompt.” All participants were able to complete all tasks.
A screenshot of one of the recordings from the usability test
“SurveyMonkey’s Build with AI feature made creating surveys a breeze, even for someone with zero AI experience like myself. The AI feature exceeded my expectations, allowing me to effortlessly craft a comprehensive and professional survey for our sister non-profit groups. In just three attempts, I had a survey ready for board approval. Thank you, SurveyMonkey, for making survey creation not only simple and intuitive but also enjoyable!”
There are five ways to create a survey with SurveyMonkey. Here's where Build with AI ranks among them all:
Indicates test surveys sent to at least one person
Indicates surveys with several real responses
In design reviews, stakeholders thought these options were too busy.