Good morning!
Hope your week is off to a good start.
Recently, I needed to learn more about the best practices for improving questionnaire conversion. The goal was to help assemble a document for the larger product team.
I already knew some principles off the top of my head but thought, âWhy not give ChatGPT a try?â
Here's what I did.
Step 1: Write the ChatGPT prompt
Below is the prompt I wrote. When asking ChatGPT questions, providing the right amount of context is important.
However, I also prefer not to overdo it when writing prompts and like to keep things simple. It may not be perfect on the first try, but I can always ask follow-up questions to refine it.
Prompt-writing techniques I often use:
Assign a role.
Share the task.
Narrow down the answers.
Specify the ask.
Ask for references.
Here is the result from ChatGPT:
Each bullet point ended with a link for further reading, which was helpful.
At the bottom, ChatGPT also put together a list of âbest resourcesâ that I asked for.
I checked those resources out. They looked legit with helpful visual examples.
I could have ended there, but decided to test something new.
Step 2: Ask ChatGPT to summarize with visuals
As I was reading through the recommended articles, I thought why not asking ChatGPT to summarize and visualize those principles for me? So I asked this question:
This is what I received:
The visuals looked âcoolâ, but they were too cartoon-ishânot what I looked for. I hoped ChatGPT could fetch a real-world UI for every principle.
I had to iterate the prompt.
Step 3: Iterate the prompt by clarifying my needs
This is what I received:
Okay. This was the closest I could get. ChatGPT couldnât fetch live internet content and I wouldnât be able to do so anytime soon.
It was due to proprietary concerns (web scraping, bypassing paywall, etc.) and complexity.
There used to be a "Browse with bing" feature at one point, but OpenAI disabled it. I couldnât see the option anymore under the GPT-4 tab. This happened to other people too.
Someone mentioned ChatGPT on Android still had that option, but I didnât get a chance to check that out.
Update on March 17: A friend of mine told me today that there is a way to get around it to fetch live internet content, since ChatGPT has recently enabled more ways to customize on the backend. Iâll look into that.
Step 4: Explore other options
Is there a way to get around?
I tested plugins including Web Pilot and VoxScript.
I also tried other platforms including Bing Chat and Perplexity AI.
None of them worked. Not even close.
The closest result to what I looked for was simply asking ChatGPT to generate synthetic images to visually represent these principles instead of actual examples:
1. Simplify User Interface
2. Use Social Proof
3. Optimize CTA Buttons
4. Ensure Mobile Responsiveness
5. Implement Multi-Step Forms with Progress Indicators
6. Leverage Power of Colors and Images
Well, what do you think?
Was it helpful to me?
I donât think so.
But at least I explored different options and had a better understanding of what I could do and not do with AI.
Takeaways
ChatGPT was able to summarize insights and provide me with relevant resources based on my request.
Yet the limitations still existed. I had to iterate the prompt several times and didnât find a workaround to meet my needs.
What was your experience like, dealing with similar tasks?
Thatâs it for today! Thank you for reading.
Next week, Iâll share an exciting use case to leverage the power of AI.
Stay tuned.
âXinran
P.S. Questions? Thoughts? Just reply to this newsletter. Iâd love to hear from you.