Interview with Jodie Cook, Founder of Coachvox.ai

Full video transcript

All right everyone, I am here joined by Jodie Cook, the founder of Coachvox, where they make AI coaches and all sorts of things where you can clone yourself utilizing their platform. Very cool — I’m actively building and testing my own Darby AI clone and wanted to reach out to Jodie to see if she could come share a little bit more about what Coachvox is all about, and then some of the more questions we were discussing here before this conversation got going. So Jodie, tell us a little bit about how you came across the idea of Coachvox in the first place, and what’s the number one problem that you realized you were going to be solving with this tool?

It started actually as a tool for voice-note coaching — coaching by voice note — and this was a platform totally different from the one we have now. We were set up to onboard coaches, and they would use our platform to provide coaching services to their clients. We had an idea that AI would be involved at some point in the future, but we didn’t exactly know how. What we thought we might be doing was incorporating AI to create a kind of helper for our coaches and mentors, to help them coach their clients better.

When it was November/December 2022 and ChatGPT came out, I remember a friend sent me the link and I was like, “Oh my God, what is this?” For the whole day I don’t think I ate anything, I don’t think I got dressed — I think I was just working it out, just talking to it, seeing what was coming back. And it just — this lightbulb moment was like, we can do what we want to do, and more, so much faster. One of the quite early things I did with ChatGPT: I was being really polite, I didn’t really know how to approach it, so I typed, “Please can you write me an article called…” and gave it the title of an article I had already written, called “How to Raise Entrepreneurial Kids.” I said, “Can you use the subheadings,” listed out a lot of subheadings from the article I’d already written, “and make it a thousand words.” I clicked submit, and what came out was pretty much my original article. I was looking at these words on the page thinking, “Hey, I wrote this — that’s not fair.” So at first I was probably impressed, but then after that I was like, this doesn’t make sense — I wrote this, but now it’s coming out, and if someone else used the same prompt (which I probably didn’t know was called a prompt at the time), they would essentially have my content.

So it started this kind of love-hate relationship with large language models, where I was like, this is kind of cool because I can get new headline ideas, I can generate ideas for my articles, I can create stuff with it — but at the same time it’s using my content and repurposing it and giving it to other people, so I wasn’t super sure what that would mean for the future. But then when we started thinking about how we could incorporate it into Coachvox, we pretty much realized we could use this as a tool that would help creators protect their work from everyone else using it, and let them ring-fence their content and take ownership of it. So the big premise behind Coachvox AI is that someone can use their content to generate leads, or add value for their audience, or charge for access to this package that is essentially an AI version of them — which they can’t do with something like ChatGPT because it’s not got their name on it. It’s their content, but not their name. That’s the background to Coachvox AI and how it works.

It’s interesting how you end up at the products that we do through all this AI innovation — it’s opening up new opportunities but also new challenges at the same time, like the protests against AI in Hollywood, AI for artists and authors in general. Something I’ve observed is that as general as all these LLMs get, the trend seems to be toward more customized, fine-tuned, smaller, really specific knowledge bases — kind of like what you’re doing with Coachvox — because then people can trust the source they’re getting it from. What are you seeing in terms of what’s drawing people to interact with these AI coaches — feedback from your coaches, from their clients?

Firstly, what’s drawing the creators to start making them is that they realize what they’re selling — and that’s because all our creators who create their Coachvox AI are knowledge workers in some way: authors, consultants, coaches, people with courses and podcasts. What they’re realizing is that more and more people are going to ChatGPT and AI tools to get their knowledge, which poses this massive question: how can these creators continue to sell their knowledge — their courses, their books, everything else — when ChatGPT knows everything? So the problem they come to us with is, “I want to keep ownership, and I want to stay relevant — I want to stay being that voice of authority that people come to, rather than going to a generic place for their information.”

They then use it in one of three ways, which leads to the question of how their clients interact with their AI models. First is putting it on their site as a lead-generation tool — they collect email addresses, let their AI version interact with audience members, and see in the Coachvox dashboard what’s been talked about, a summary of the conversation, and how many times a person’s come back to speak to them. Many of our creators have very big audiences and have spent years building their personal brand and creating content. The second use case is people who have an existing membership group or course where their audience is in a specific place, like a private group. One of our creators, Rose, is a coach to female entrepreneurs helping them make their first million. She has a private Facebook group for people in her coaching programs, and she was spending about five hours a week answering questions there — now “Rosebot” answers those questions. She launched it to the group as a value-add: “This is how my content and knowledge can help you, even when I’m asleep or unavailable, and you can ask it whatever you want.” The third use case is creators charging for access to their AI. I wasn’t sure the world was ready to pay for AI coaching, but it is — I’ve seen $7 a month, $9 a month, $10 a month, from hundreds of thousands of people, to get coaching from them.

The clients attracted, whether free or paid, are looking for the creator’s content applied in a way that’s specific to them. If you read a book, the book’s not going to coach you through your problems — it’s up to you to apply it. But if you talk to the AI of someone who’s written a book, it’s like being coached through that book. People love it because it’s focused on them, available 24/7, they can ask “stupid” questions because it’s not a real person and won’t judge them, it won’t turn up late or cancel a session, and it’s probably low cost if not no cost. A lot of the things we weren’t sure people would value didn’t matter, because the pros outweigh the cons in such a big way.

I loved all the different use cases — leads on your site, augmenting yourself to interact with leads, membership group value-add, and then a lower-ticket monetization play, $7–$10 a month for specialized knowledge, maybe complimentary with a book purchase. It’s a unique way for an author to stand out — “talk to my AI version of me to get more from the book you’re reading.”

That’s a really fun thing we’ve seen too. In my last book, the back page mentioned a companion course — “if you liked the book, here are video walkthroughs” — that’s where I sent people. Now I’d send people to Jodie AI. Our creators who are also authors say exactly that: “talk to my book, get coached from my book.” It helps them get more value from content they’ve already written, without having to keep on the content grind — they can turn what they’ve already made into the ultimate lead magnet that never runs out, because it’s based on the person interacting with it rather than fixed content.

I love how simple it is to charge for that — a gap I’ve seen is that not many apps make it easy to package and productize this way. And with AI accelerating so fast, this is the perfect time for creators and entrepreneurs to stand out. What’s the average setup time — five hours of focused work, ten, fifteen?

It probably took me about two to three hours of heads-down focus one morning, having already collected a lot of my documents beforehand. We see all different types. A lot depends on the breadth of questions your AI is likely to be asked. One creator wrote a series of fiction books and created an AI for each character — narrow scope, so training is quick. But if you’re a marketing entrepreneur training your AI on the entirety of marketing, you’d want to upload a lot of content, potentially create new content to give your AI a “worldview,” because we want the creator’s own opinion, not a generic ChatGPT answer. One of our creators, an astrologer named Alice, has written a book every year for twelve years in the exact format needed to train an AI — almost like she knew in advance. So there are content formats you might already have that are perfect for training. It could be up and running in a couple of hours, or a few weeks if you want to train it in depth — or you could hand the process to someone on your team.

Can we backtrack into the formatting — for a book, say — what does an ideal format look like so the time spent creating content is most effective for training?

Written content is best. There are two ways in our system: before Coachvox AI was fully built, our beta creators created their own spreadsheets and uploaded them. Now, in a training room, you upload your material and we convert it into the question-and-answer format required to train a model. We recommend uploading one document first — an article or blog post — see the training data it produces, and check whether the question-and-answer format looks like what your audience would actually ask and how you’d actually answer. Then go into a separate training room and have a live conversation with your own AI, asking it the kinds of questions your audience would ask, to see what it says based on limited training so far, and fine-tune from there. Keep uploading content incrementally, and involve your audience in the process — tweet “I’m training an AI version of me, what would you ask it?” and train on those actual questions so the responses make you proud. We made a whole promo workshop on ways to get your audience excited and involved before launch.

That builds engagement while you’re testing — the pre-marketing and pre-selling that we’ve talked about for years applies here too: share what you’re building as you build it, because a lot of people are mind-blown by this even when they’re not close to the AI conversation. Is there anything coming in 2024 that excites you, and anything people should be aware of to avoid this going wrong?

What excites me right now is the results AI coaches get for clients. I was just added to the Slack group of a creator charging for access to his AI, reading comments of people getting marketing advice from his AI and actually making money from it — I think there will be more of those stories, and it’s great for creators’ PR. There’s also a test we recommend in the promo video: run your AI with a group of people in your audience, see the results, then turn that into headlines — “an AI coach helped me lose 20 pounds,” etc. Being first in a niche is newsworthy in a way a generic PDF checklist never is.

On the flip side, we’ve had interesting requests — a teacher messaged asking if we could remove the “AI” label from the chat screen and add a random delay to responses so people wouldn’t know it was an AI. I realized she didn’t want people to know it was an AI version of her. We said no — that’s not what we’re here for. But you can imagine a doctor or lawyer wanting their AI to give medical or legal advice without disclosure. We’re big on making sure people always know it’s AI, because you don’t want to mislead people into acting differently than they would if they knew.

On the deepfake side, I think everyone should teach their parents and grandparents about this — it’s really easy to clone someone’s voice or appearance now. Tell them: you could get a call that sounds like me, saying things I would say, asking for money, but it isn’t me. Have a family safe word, or questions only you and they would know the answer to, that aren’t online anywhere.

You mention on your website that GPT-4 isn’t learning from or training on the content creators put into their bots — that content belongs to the author, coach, or consultant. Is that right, for anyone considering platform security?

That’s right. Our platform is built on OpenAI, but we engage through their API, and the information creators put into our platform is ring-fenced — we don’t use it to train any other models, and we don’t create anything that isn’t theirs. We make sure creators continue to own and benefit from the work they’ve put in, because we ourselves are creators too, and that’s central to why we built this platform.

One more question before we wrap — non-creator use cases, like an agency building these for clients who have a book or course but not the time or bandwidth to do it themselves?

One example: a running brand based in Australia wanted to create a running coach but didn’t have much of their own written content — they had their ethos and general talking points, but nothing fully formed. I taught the owner a process: come up with a list of questions someone might ask the AI, put the company’s website into ChatGPT, ask it to generate 100 questions someone might ask this type of coach and answer them based on the site’s ethos and vision, then have the creator review and edit those answers before using them to train the Coachvox AI model. The result was a running coach built with almost no original written content.

That’s using AI to help generate the training content itself.

Yeah — using AI to build an AI, like using a 3D printer to print a 3D printer. It sounds odd, but it makes sense, especially because in our system the more important thing is the fine-tuning — testing what the AI says to a question like “I’m training for this marathon, what should I do, I’ve had this injury before,” and editing the response — even if ChatGPT provided the base-level content.

So there’s really no excuse for someone not to clone some version of themselves for their business if they want to — not that everyone needs to right now, but it’s surprisingly simple to get one of these clones up. Whether it’s adding leads to your website, adding value for members of an existing paid community, or charging directly for access, there are a number of low-hanging-fruit ways for creators, entrepreneurs, coaches, consultants, and brands to integrate AI simply and stand out from another PDF checklist.

Absolutely, and for anyone who’d like to learn more about Coachvox and sign up — we’re affiliated with Coachvox and helping people in our own community create their own clone AIs, you can check out their platform. Jodie, where can people find and connect with you directly?

I’d be happy to hear from anyone with questions — probably the best place is Twitter (I don’t want to call it X yet), my username is @jodicook, or message through the Coachvox website.

And if people go to Coachvox and scroll down, they can enter their email and name and start talking with Jodie AI directly?

Yes, exactly.

Jodie, thank you so much for taking the time to share your story, Coachvox, and the future of AI and cloning yourself to grow a business. Any parting words for anyone interested in this space?

One of the things you messaged about earlier was the emotional intelligence of AI, and whether it will ever be as good as a human. I think it’s really important to remember that AI is not emotionally intelligent at all, and perhaps never will be — it can mimic having empathy and emotions, but it doesn’t actually have them. That’s important to keep in mind whenever you’re engaging with AI, and when you’re setting expectations for what your own AI will achieve with its clients.

Jodie, thank you so much again, and we’ll see you on the other side.

Thanks Darby, see you.