To get good YouTube video ideas out of ChatGPT, you have to give it YouTube data first. I copy the most popular videos from four or five channels in the niche, have ChatGPT clean them up into title, upload date and views, build a viewer avatar from a competitor's transcript, and upload both documents into a custom GPT. Then I ask it for content formats, the top 30 keywords and 100 video ideas, and cut the list down to the best handful. It's a tool for creativity. I still decide which ideas are worth making.
- ChatGPT doesn't know YouTube on its own. It needs the title data from the most viral videos of channels in your niche.
- A viewer avatar built from a competitor's video transcript tells the GPT who the ideas are for.
- Ask for 20 to 50 content formats, then limit the channel to three to five of them. A format is a fill-in-the-blank title like "what happens to your body when you do [blank] every day."
- Prompt for 100 video ideas from your keywords and formats, with 10% new keywords and 10% new ideas in the mix.
- Remove exact duplicates of competitor videos and anything that doesn't fit the avatar, then ask for the top 20 that all fit together on one channel.
- Don't just trust the titles it gives you. When I come up with ideas myself, I write 30 and pick one.
- First, What AI Will and Won't Do for You
- Step 1: Create the Title Data
- Step 2: Create a Viewer Avatar
- Step 3: Build a Custom GPT
- Step 4: Create Content Formats
- Step 5: Pull Keywords, Then Ask for 100 Ideas
- Step 6: Clean Up the List
- Step 7: Ask for Viral Probability and Check YouTube
- What I'd Do If You Haven't Uploaded Yet
- Video chapters
- FAQ
First, What AI Will and Won't Do for You
I added this part while editing the video, as a quick disclaimer.
Don't just trust the titles it comes up with. You need to ask yourself whether this is something people actually want to watch.
I use AI for inspiration and creativity. I don't take whatever it gives me and turn it straight into content. I'll sometimes do that, but the chances are like one in a hundred, and that's after doing YouTube for six years, making thousands of videos and getting hundreds of millions if not billions of views.
Even when I come up with ideas as a human, I'll come up with 30 and pick one. Same thing with AI.
Step 1: Create the Title Data
ChatGPT really thrives off of having data. It's really smart, but it doesn't know YouTube. So you have to give it YouTube data.
- Pick a niche. For the example in the video, I used the fitness niche.
- Go to a channel in that niche, open the Videos tab and sort by Popular, so you're looking at all of their most viral videos.
- Highlight everything on the page and copy it.
- Paste it into ChatGPT with this prompt: "Give me the data cleaned up. Just title, upload date and views."
- Paste the cleaned up result into a new Google Doc.
Feed it a ton of data. I did a total of four or five channels, and I made sure there were a lot of really viral videos in there. When everything is in one Google Doc, go to File, then Download, and download it as a doc.
Step 2: Create a Viewer Avatar
Go to one of the videos you're emulating. Open the description, click Show transcript, toggle the timestamps and copy the transcript. You can do this with multiple videos if you want more accurate data.
Then ask ChatGPT, as closely as I typed it in the video:
"Can you create a viewer avatar for me based on this transcript? I want to know what shows they watch, what hobbies they have, what books they read, what channels they watch, what YouTube videos, what websites, psychographics and everything."
There's no exact science to this. For the fitness video I used, this is what it came back with:
| Part of the avatar | What ChatGPT said |
|---|---|
| Age | 30 to 45 years old |
| Gender | Primarily female, with a significant number of males |
| Work | White collar professionals in fields such as IT, marketing, healthcare and finance |
| Income and education | Middle to upper class, college educated, some with advanced degrees |
| Location | Urban and suburban |
| Psychographics | Health conscious, time poor, ambition rich, motivated by appearance and health, skeptical yet open-minded |
| Media | Shows like The Biggest Loser |
The psychographics are the really important part, and I thought the result was pretty accurate. Copy it into a second Google Doc (the viewer avatar doc) and download that one too.
Step 3: Build a Custom GPT
Now you create a custom GPT with those two files in it. When I recorded this, you had to have the upgraded version of ChatGPT for this, and honestly the free version wasn't good enough.
- In ChatGPT, go to Explore GPTs and click Create.
- Go to the Configure tab.
- Upload both files (the title data and the viewer avatar) into the Knowledge section.
- Give it a name. Mine was a fitness title creator.
- Write a short description. Mine was along the lines of "get feedback on ideas and create new titles."
- In the instructions, say what it does. The main thing I put was that it helps create titles and helps you choose.
- Hit Create. You can share it with a link or lock it to yourself.
Step 4: Create Content Formats
Inside the new GPT, ask it for content formats based on the data. (I pasted this prompt in the video instead of reading it out, so I'm describing it here.) I usually like to ask for a lot and then clean it up. I'll say 20, and you could do way more than that. For this one I did 50.
A format is a fill-in-the-blank kind of content strategy. For example:
"What happens to your body when you do [blank] every day."
That was a huge one on the fitness channel I pulled the data from.
For a YouTube channel, you want to limit your formats. New formats all the time are very resource taxing (time, money, energy). So I recommend repetitive content formats.
To narrow the list, I asked: "Which of these are the most relatable to my viewer avatar?" Then I dropped a couple I didn't like and put the rest on a new doc.
I recommend three to five formats, and sticking with them. If you give it too many formats, it gets really complicated. Once you've used those a lot, you can branch out and try new ones.
Later in the video I narrowed it further with: "I just want to pick three formats. Give me your favorite three I can make 20 videos on." For a faceless channel, the three I'd run with were:
- Some sort of challenge format (a [blank] challenge)
- "What happens when you do [blank] every day" (or for a month, every morning, or every night)
- Top 5 or top 10 lists
The list formats are always good. They've crushed it for so many faceless YouTube channels. If the word format is new to you, I explain it with my own channel's numbers in how to make $30K a month with a faceless YouTube channel.
Step 5: Pull Keywords, Then Ask for 100 Ideas
Next I have it create a bunch of keywords. The prompt:
"What are the top 30 keywords pertaining to our viewer avatar that you saw get results on the channel data?"
Then the ideas prompt, which uses the keywords and the formats together:
"Based on the keywords and the formats, come up with 100 video ideas. Come up with 10% new keywords and 10% new ideas to include in the mix."
I like adding that last line to get a good mix. When I ran it, it came back with around 20 or 30 ideas per category.
How far you cut depends on your production. If you're only producing four videos a month, clean it down to the best four. If you're doing 20 or 30 videos a month, you can keep a lot more ideas.
Step 6: Clean Up the List
Here's every prompt from the cleanup, in the order I used them:
| What I'm doing | What I asked the GPT |
|---|---|
| Remove copies | Take out exact duplicates of the competitor channels (variants are fine) and anything that doesn't fit my viewer avatar |
| Condense | "I want to keep it simple at the start. Can you give me your top 20 that all fit together on one channel?" |
| Pick favorites | "What are the best ones from this list here? Give me five." |
| Push for reach | "Which ones do you think are the most viral?" |
| Get more | "Give me another 30 ideas in the fitness category." |
On the first cleanup, it only found two that were irrelevant or a duplicate.
The GPT grouped the fitness ideas into exercise based content, diet and nutrition, lifestyle and habits, and challenges and transformations. If you're getting started on YouTube, I would pick just one of those categories. The bigger channels can cover a lot more. I'd do just the exercise based content and refine from there.
The answer changed when I switched the question from "best" to "most viral." The first set looked really good for converting to sales if you're selling a product. So you can ask for sales ideas, and you can ask for viral ideas.
It's okay if you have spelling errors in your prompts. I basically just talk to it.
Step 7: Ask for Viral Probability and Check YouTube
If you want to get fancy, this is my favorite one:
"What are the probability of each video going viral?"
It tells you which ones it thinks have the highest probability to go viral. Here's how I read some of what it gave me:
| Idea from the GPT | My read |
|---|---|
| Daily plank challenge: see the core strength results after 30 days | Hard to make on a faceless channel. Better as a personal brand |
| I did 10,000 steps every day (see my transformation) | A personal story is going to be way more viral, but you have to be on camera |
| What happens when you go sugar-free for 30 days | Good, and you can do it as a faceless creator |
| 30-day squat challenge | You can do that as a faceless channel |
| The ultimate 10-minute abs workout for busy people | A good one, and a unique idea |
On that last idea, I went to YouTube and searched it. No one had really done the busy people angle. That's how you come up with unique stuff that hasn't been done. I would still want to base it off what's already gone viral.
What I'd Do If You Haven't Uploaded Yet
Two things I've figured out from playing with this. It works better when you start giving it your own keywords as well. And if you keep finding newer keywords that haven't been done, the same formats keep working.
My biggest recommendation if you haven't uploaded videos before is to start making videos. Pick 10 of the ideas, go make them, and see which one performs the best. Then do more of whatever angle works, and build around a specific audience and a specific avatar.
For the manual version of this research, read how to find viral YouTube video ideas. For more prompts, I have a separate post on ChatGPT for faceless YouTube.
Video Chapters
- 0:00What this training covers
- 0:32Disclaimer: AI is a tool
- 2:04Step 1: create the title data
- 4:45Step 2: create a viewer avatar
- 7:21Step 3: build a custom GPT
- 9:25Step 4: create content formats
- 11:59Top 30 keywords
- 13:00Generating 100 video ideas
- 14:02Cleaning up the list
- 15:39Top 20 that fit on one channel
- 18:15Probability of each idea going viral
- 19:16Checking the idea on YouTube
- 20:18Limiting to three formats
- 22:24Where to start if you haven't uploaded
FAQ
Can ChatGPT come up with YouTube video ideas?
Yes, but Devon Canup says it needs YouTube data first. He feeds a custom GPT the most popular video titles from four or five channels in the niche plus a viewer avatar, then asks for formats, keywords and 100 video ideas before cutting the list down.
What prompt does Devon Canup use to get YouTube video ideas from ChatGPT?
After loading his data into a custom GPT, Devon asks for the top 30 keywords that got results in the channel data, then prompts: "Based on the keywords and the formats, come up with 100 video ideas," and adds a request for 10% new keywords and 10% new ideas in the mix.
How do I make a custom GPT for YouTube titles?
In Devon's walkthrough, you go to Explore GPTs, click Create, open Configure, and upload two documents into Knowledge: cleaned up title data (title, upload date and views) and a viewer avatar. When he recorded the video, this required the paid version of ChatGPT.
How many content formats should a YouTube channel use?
Devon recommends three to five formats and sticking with them, because new formats take more time, money and energy.
Should I use the video ideas ChatGPT gives me without changing them?
Devon says no. He uses AI for inspiration and estimates that taking its output as-is works about one time in a hundred for him. He recommends picking 10 ideas, making them, and doing more of whatever performs best.
Want This Built With You?
Faceless Channel Academy is the coaching program I run for people building faceless YouTube channels: niche, formats, team, packaging, with an advisor and live calls. Book a free strategy call and we'll look at where you are and whether it's a fit.
Book a Free Strategy Call →Not ready to talk? Grab the free niche research template first.
