Build your own AI Interview Agents with n8n Forms!

Built with n8n
1

Customer or User Interviews are time consuming and expensive. I show you how to automate them using n8n’s AI Agents and Multi-page Form nodes.

Hey there :wave: I’m Jim and if you’ve enjoyed this article then please consider giving me a “like” and follow on Linkedin or X/Twitter . For more similar topics, check out my other AI posts in the forum.

OdL45wzLi627g6OOd-EDi
OdL45wzLi627g6OOd-EDi1024×768 57.4 KB

With n8n’s recently released multi-page forms, I was curious to explore ways to utilise this functionality with AI agents. After a week of experimentation, I’m happy to report AI-powered forms are now definitely a “thing” in n8n! I have a few templates lined up to share but my fave so far has been my (totally original!) AI interviewer template I’ll talk about in this post.

What is an AI Interviewer?

User research interviews are known to be quite expensive. If you think of the time and cost for preparation, resourcing, executing and reporting, it’s not hard to see why companies do only a few per year and prefer to outsource. The alternative is to run a survey but they’re low ROI - static questions make for a boring experience and don’t serve to dig deep enough to capture insights. How might we conduct quick, cost-effective interviews that still uncover meaningful insights and provide the depth we need?

The AI interviewer could be the perfect solution! An agent can conduct a dynamic interview where it intelligently adapts its questions in response to a user’s answers, is available 24/7 to suit the customer’s availability, can handle hundreds of sessions simultaneously and effortlessly streamlines post-interview data-entry and reporting. I was able to build out the template for my AI interviewer in a matter of days and best part is, by using a free model such as LLama3.2 via Groq.com, the cost of running an interview is close to nothing!

:rocket: Try out this live demo here: Interview for uncovering people’s experiences preparing for and taking the UK practical driving test

Good to know about the demo

  • The topic is UK specific so a little roleplay and imagination is required :smiley:
  • Questions are dynamically generated Using LLama3.2-90b via Groq (free tier). It may bug out at times where the response isn’t formatted correctly. If this happens, start over and try again.
  • Responses are deleted after a short period.

How does it work

  • Start by providing the AI Agent with an interview topic and the user’s name.
  • The AI Agent’s question is presented as a n8n form and the answer is capture via form submission.
  • Each question and answer pair is then recorded in a Google Sheet.
  • A loop is then created by directing the flow back to the agent for it to generate another question, repeating step 2.
  • When the user asks to stop the interview, the workflow breaks out of the loop and shows the completion screen.

Template

Conclusion

N8n multi-page forms are an exciting addition to the n8n AI builders toolkit and opens up new channels where AI can meaningfully interact with your users. I’m still experimenting with other n8n x form scenarios and will share more templates around this combination in my creator hub soon so watch this space! Working on something similar? Let me know!

if you’ve enjoyed this article then please consider giving me a “like” and follow on Linkedin or X/Twitter. For more similar topics, check out my other AI posts in the forum.

Still not signed up to n8n cloud? Support me by using my n8n affliate link.

Need more AI templates? Check out my Creator Hub for more free n8n x AI templates - you can import these directly into your instance!

21 Likes
2

My first reaction: A survey with an “unending series of questions” must be the worst customer nightmare.
Interestingly enough, this dialogue format turned out to be really fun!

So much so, that it led me to think if this could even be a paradigm shift in how surveys are presented to users. Answer as many questions as your patience allows while AI keeps you engaged and somewhat entertained.

The dialogue - which felt very natural - was in a little conflict with the lightning fast sequence between input and following question. Maybe, a little random waiting time inbetween the steps would make this feel even more like a real conversation. My two cents from a UX standpoint.

Bravo, great idea and very nice execution!

2 Likes
3

@Jim_Le
Thank you for sharing,
There might be a bug in this workflow.

I tried it out and found that if the browser window is closed midway, the Get Answer node remains in a permanent waiting state with no way to stop the waiting. I restarted n8n, but after reopening the workflow executions panel, the node is still waiting.

I am currently unable to resolve this issue and would appreciate your assistance. Thank you very much!
Demo video:

I tested the following template:

Conversational Interviews with AI Agents and n8n Forms | n8n workflow template

4

Hey @ichat006

Not really a bug - forms put the execution in a waiting state until the user submits and there is always an expectation that users will abandon the process.

Good to know that the execution waiting state does not take up resources and rather it’s more of a paused/resumable state. If the latest version of n8n (1.70), the spinning animation is removed for waiting states if that was what was worrying you.

5

I am currently using the latest test version 1.71.1 of n8n Cloud, and so far, the node is still in a waiting state.

1 Like
6

Hello,
I’m also experiencing issues with this workflow. I’m using the self-hosted Community Edition version 1.74.3. The trigger form works, and the flow seems to function properly up until I reach the “Get Answer” node. For some reason, the initial form is triggered with the correct given path, but the “Get Answer” node expects a form submission on the path /form-waiting/868, where 868 appears to be the execution ID. If I open that path in a separate browser tab, I can continue the survey flow as expected. It seems there is a disconnect between the form trigger and the “Next Form Page” node.

2 Likes
7

I am having this same issue. I have not been able to find a way around this.

8

Hey @YCharron @coreycauble
Thanks for letting me know.

I’m struggling with reproducing the issue - the forms work fine for me on self-hosted and n8n cloud! I’ve been made aware there are certain issues still outstanding for forms on some self-hosted setups - see Bug Report: n8n forms "next form page" not triggering - which might be worth exploring.

If you could share what your setups are like - either here or privately - I’ll be happ to take a closer look.

9

I finally worked. I just takes about 1-2min before the next form renders… What could cause that? The flow itselft seems really fast, it just sit there until it renders…

10

Hard to say without digging deeper into your setup and if you’ve made any changes to the template. You may want to create a new post for this for more visibility.

I would suggest perhaps changing the LLM, preferably something fast like gpt4o-mini or Gemini-flash.

1 Like
11

Hi there,

I’m also experimenting with your workflow—thanks for sharing it! I have a question about configuring the Agent so that it selects questions or topics from a predefined list or sheet and then develops the Q&A around them. Ideally, once it has explored a topic in depth, it should transition to the next one, ensuring it covers a set of predefined topics/questions.

Do you have any ideas on how to achieve this?

1 Like
12

Would also like to know. Im thinking either using the redis or other database to have the workflow store those questions, and then inject into the user prompt message those questions so it knows what to ask.

1 Like
13

Just tested this out and it is very cool. The never ending question format (with relevant active listening-style questions) was interesting and I don’t think I’ve ever experienced anything like that before in this multi-step form way. I guess it could be simulated with a chatbot.

I can imagine a very near future where this is performed with real time voice enabled interaction. And then to take a step further, a HeyGen style Avatar.

14

@EK12 @5uperfred

Check out my new template which explores this idea of using preset questions for an AI agent - Create Product Satisfaction Surveys with Telegram, Google Sheets and AI | n8n workflow template

Yes, pretty much you’re managing your own state machine to control the flow of the session and in doing do, lose a lot of that agent autonomy.

When building this new template, I found a big issue with hallucinations ie. questions leaking into other questions, agent generating it’s own questions etc The main reason (in my opinion!) was the long-lived and overloaded chat memory. My solution was then to wipe chat memory inbetween questions so the agent starts fresh.