How do you build a personal research feed with AI agents?
The short answer
I built mine with OpenMausBot and my Claude subscription: the bots search Instagram, YouTube, TikTok, Reddit, X and Threads (every 2 hours in my setup; schedules vary by bot) and group what they find into emerging topics and topics gaining or losing momentum.
They make mistakes and false positives, so I keep checking them; to start your own, pick one topic and two sources, have your agent save the original URLs, and check one finding yourself.
Chapters
Receipts (captured October 18, 2026)
| What | Number | Source |
|---|---|---|
| Who does the research | OpenMausBot and my Claude subscription | in the video |
| Where it looks | Instagram, YouTube, TikTok, Reddit, X, Threads | in the video |
| How often it searches | Every 2 hours (schedules vary by bot) | in the video |
| How findings are grouped | Emerging / Gaining momentum / Losing momentum | in the video |
| How the bots are checked (analytics table header) | Rated useful / High + medium | in the video |
| What the usefulness score is not | A usefulness score is not factual accuracy | in the video |
| Thumbnail click performance | Unmeasured | in the video |
Try it on your own computer
- Pick one topic and two sources you are allowed to access, and write down the decision the research should help you make.
- Ask your agent to save the original URL, the time it observed each item, and the exact evidence behind each finding.
- Check one finding yourself: open the original, mark it useful or irrelevant, and explain why in the next run.
- Keep usefulness and accuracy apart: a usefulness rating is not a factual accuracy percentage, and a cumulative view count is not growth.
Build a research feed you can actually use (starter research brief) Free, on GitHub
Transcript
0:00 No more newspaper. It's like customized newspaper for you. I read this daily. You should not trust the bots at first go. It's going to make a lot of mistakes, false positives.
0:11 Today I want to show you how I made my own social media dashboard where it researches all the keywords and all the topics that I want to search. It will automatically search every 2 hours and bring in the posts that I want to see and group them. So I created this to make content easier.
0:37 The research part is done by OpenMausBot and my Claude subscription. There is Instagram, YouTube, TikTok, Reddit, X, Threads.
0:48 So on the left you can see the emerging content. Right now it's one-person builder, AI agent automation stack verification and production reality. This is something new.
1:02 And then in the middle there's gaining momentum. Usually I look at here, and gaining momentum is more like a long-term trend. So I look at this and try to analyze it, read it, get into it and see where the puck is going. I try to have some balance between gaining momentum and losing momentum, because sometimes the bots will make a big mistake on gaining momentum and losing momentum.
1:28 I have even set up the analytics for the bots researching. So you have to evaluate, verify, fix, fix, fix, fix over and over again.
1:39 And if you go down here, you see AI-powered user frustration limits and subscription fatigue. Yeah, especially me. I have three subscriptions and I honestly don't think I'm fully utilizing all of them. And now it's really fun to actually read it instead of going through all these different social media and clicking all these stuff.
2:00 And lastly, this is just a few hours ago, I have created this YouTube thumbnail generator, and it'll help me create that next content for me.
2:11 So yeah, I hope this works out well, but everything is experimental right now.
