Use AI to call your SMEs and create content with substance, not slop

Here is the TLDR, the video below shows me putting in an old blog post I want to update, getting a phone call from a Voice AI agent (VAPI) running on an n8n workflow, and Chad built V1 in on the side in 3 weeks. The tool lets you interview SMEs via phone calls, after running deep research against a company’s old blog posts. 

In the same way that I realized SMEs are the unlock to high quality content, but they also have the least time, so how can I use AI not to produce slop, semi-slop, or barely good enough.

I wanted to build content that won for HUMANS not algorithms only.

I HATE listicles, sure they get you AI visibility in 5 minutes, but no one shares them with people they trust.

Here is the AI agent, calling me on my phone, interviewing me, asking me questions (this is a demo), and at the end of it (not in this video, sorry) I get back a remixed post based on what my interview said, removing the old outdated or incomplete thoughts, and adding my SME flair, keeping me from writing sea of sameness slop.

Here’s the voice agent doing an SME interview:

Sign up to the newsletter below, to get the stuff we’re cooking up in the lab!

Here’s the longer story on how this came to be:

Almost 2 years ago I wrote this post challenging all of us to think… what if ChatGPT disrupted some of Google’s dominance in finding answers to our questions?

I think it outperformed on social every post we had written that year.

The channels told me that it was social not search that drove the value, and I loved that.

I dug into the sites that linked as referrals and they were great high value sites.

Can I build a Voice Agent to Interview me, yes.

But then 1 year later a TON had changed, and the post was now “good” but it could use some polish.  I had an idea, give it to some “content strategist” to help me write it, or get someone to “interview me”. See I had gotten the bug a long time ago…I was at building it out, but with each week I was getting closer to using the Voice AI, Vapi to update content by interviewing me. I never got it where I wanted it though.

Shout out to my guy Christian Ulstrup who had put me on to Vapi before I even knew what voice AI was. I now knew what was possible even if I was failing.

Now I had a post I wanted to update, that I knew humans loved, and while I could be interviewed, I learned about latency, system prompts, etc, and how models impacted those things, but I wasn’t able to get the questions I really wanted and I was sinking a lot of time into learning, but not producing something of value.

You can do this without all the automation, old post + basic deep research + some good prompting and an automation to drop them into a doc, and boom, share the doc and get your SME to go for a walk.

So I started working with someone to use Perplexity Sonar to run the following workflow

1. Ingest my posts
2. Find out why they are “old”, what has changed since it was published
3. Research high quality sites, reputable sites, and look for opinions that validated or challenged my assumptions, in a question format, like “According to Deloitte, they believe the opposite of what you said here, are they wrong? why?
4. The workflow would dump a bunch of questions into a set of Google docs, so I could answer them using my voice
5. Mash up my answers with the post, helping to remove the old stuff and add in new stuff

What I loved about this was I was able to use my time walking and doing other things, to quickly voice what was in my head to update content, and I loved the questions.

I learned a few things about being interviewed w/ Deep research and Voice AI  

1. It was easier to be asked questions and give my opinions than writing
2. I gave a LOT more context and nuance when interviewed that could be cut into social
3. Approvals were fast because it was what I actually said
4. It was great being an SME writing this content, it maximized my time and let me generate more value

That lead to a test we’ve started running with a few clients. We’ve now built an end to end solution, a series of agents and sub agents that run this whole workflow.

The killer is the phone call, just this week we got a 20 year SME at a client to test this, one of his areas of feedback was “can I get these questions in advance, they were so good I would have liked to have had time to think about them.”

He learned what I did… the questions are TOUGH at times.

When you prompt these AIs right (look for high quality conflicting viewpoints form reputable sources) and let it get good deep research taking its time, its going to ask you some challenging questions.

The next big step: Evals

The other thing I am loving these days is thinking through and building out evals, TY to Chad Gingrich for informing me that n8n has those as part of the platform.  

Why are evals important?  Lets say you want to write for a CMO audience but your SME went off the rails and went super tactical, you can have real time evaluations of answers pushed back into the voice AI to ask clarifying and follow up questions to get the SME to the ideal end state, they forgot to answer quetsions about case studies, no problem the eval if failed can make a follow up call / text (you could also put some nudges into your VAPI system prompts but we think evals are going to be the way).

That’s the next big thing we’re going to learn to make this even better.