Don't Write Crappy YouTube Title and Descriptions (And Check Your Sound Quality) | For AI Fellows
An onboarding video tells Humanitarians AI Fellows to write real YouTube descriptions, amplify their audio, and remove Zoom watermarks before submitting videos.
Every fellowship program eventually needs a video like this one: a direct, no-nonsense correction aimed at people who are doing good work but cutting corners on the presentation. Professor Bear opens by holding up real, anonymized emails sent back to Humanitarians AI Fellows whose video submissions weren't acceptable, not because the content was bad, but because the basics, a real description, decent audio, weren't there.
What counts as a real YouTube description
The core complaint is simple: a one-line placeholder is not a YouTube description. Fellows are told the media team cannot be expected to write descriptions or titles for them, and that the standard is a description the team can simply copy and paste as-is, with no editing required. To make that achievable, fellows are pointed to a specific workaround: describe your video in a prompt to Claude, and it will generate a usable title and description. The instruction that follows matters just as much as the tool recommendation. Read what Claude gives you before submitting it, because you are the one who knows whether it is actually correct, not the AI and not the media team.
Links belong in the description, not just in your head
Beyond a clear description, fellows are expected to include a link to the actual code behind their project. A LinkedIn link is treated as optional but strongly encouraged, framed not as vanity but as strategy: the fellowship exists as a bridge between graduation and a real job, so a video is also a chance to market yourself to anyone who watches it and wants to learn more about who made it.
Fixing audio without fancy equipment
The second half of the video walks through actual audio review, watching submitted videos side by side and grading them out loud. One video is judged acceptable but not great; another is called out directly as too quiet to be usable. The fix demonstrated is a simple editing step: open the audio effects, find compression or amplification, and increase the level so the recording is more audible, then save it again. That's the entire fix. It does not require professional recording gear, just a few minutes in whatever editing tool is available, and a willingness to actually listen back to your own recording before submitting it.
The expectation, stated plainly
Running through this checklist, Professor Bear repeats the same point from a few different angles: fellows made the video, so fellows are responsible for making it presentable. That includes writing a description that doesn't need editing, providing links to the code and (optionally) a LinkedIn profile, and checking that the audio is actually audible before it goes to the media team. None of these are difficult individually, but skipping all three is what generates the kind of email being read aloud at the start of the video.
The Zoom watermark problem
One more issue gets flagged, though it is deliberately left for a separate, dedicated video: the persistent on-screen artifact that shows up throughout a recording when Zoom is used without adjusting its settings. The advice given here is practical rather than prescriptive. If you don't know how to remove it, take a screenshot of the issue and ask your favorite language model how to fix it. Zoom itself is treated as an entirely acceptable recording tool for fellows without access to more advanced software like OBS Studio, but leaving a visible watermark artifact sitting in the video for its entire runtime is called out as unacceptable, regardless of what tool was used to record it.
Key takeaways
- A real YouTube description must be complete enough that the media team can copy and paste it without editing; a one-line placeholder does not qualify.
- Fellows can use Claude to draft a title and description by describing their video in a prompt, but must read and verify the output themselves before submitting.
- Every submission should include a link to the project's code, and a LinkedIn link is recommended as a way to market yourself during the fellowship.
- Low or muffled audio can usually be fixed with a simple amplification or compression effect in your editing software before you export and submit.
- Zoom is an acceptable recording tool, but visible watermark artifacts left in the final video are not; ask a language model for help removing them if you're unsure how.
Who this is for
This video is aimed squarely at Humanitarians AI Fellows preparing video submissions for the program, and at anyone producing tutorial or presentation video who wants a quick standard to measure their own descriptions and audio against before hitting submit.
Full transcript(auto-generated, with timestamps)
[0:01]Okay, Professor Barrett here. Um, more just basic basic stuff for volunteers. So, these are actual emails. I've anonymized them um that we got for people who did videos. This is not a YouTube description. The instructions are explicit here. We even give you a little prompt here that you can put into Claude. You can take whatever you said about your video. You can put it in Claude and it'll give you a description. Once it gives you a description in a title, read it. You're the one who knows whether this is correct or not, not the media. Of course, you need to include a link to the actual code.
[0:55]Should also put a link to your own LinkedIn, that's optional. But the reason why you're in this program, it's a bridge between when you graduated and when you're getting that real job. So why not market yourself? But this is unacceptable. What is that? It has to be that the media team can just simply take that, cut and paste that, take that, cut and paste that. Do not expect the media team to write these things for you. That's your job. Let's also look at the sound. So, these are the videos. So, one of them's here. It's your job. Just There's like 10 videos on this already. Open it up. Look
[1:55]At it. So, the sun is okay. It's not great, but what we want to do is we just want to go to effect, volume, compression, amplify, and take it up a bit so it's more audible. And then save it again. It's that simple. But do not expect the media team to do that for you. Or if let's take a look at the other one just for the just, you know, to see if people are listening. Okay, this is probably borderline acceptable. It looks like good job uh video number two person. This is this is what you would expect to see. So again, do not expect the media team to do this
[3:06]Stuff for you. You made the video. Another thing to look at, and I'll do a separate video on this, is to not have these little This thing is in the whole video. I'll make a another video on how to remove that. But if you want to remove that, you just take a screenshot, then ask your favorite language model, how do I remove this? So, I understand that, you know, you're a recent graduate. you don't have fancy uh uh fancy software to record stuff. So, Zoom is okay. OBS Studio is better if you want to learn that. But if you're going to use Zoom,
[4:20]Look at the Zoom settings and see how can we remove this thing from the Zoom settings. Okay, so that's it. Uh, I'll make another video on removing that from the Zoom settings because lots of people use Zoom and that's okay. But this is not okay. It's not okay to do a Zoom thing and just have that that sitting there. Okay, so that's it. I'll talk to you uh soon.
More videos
2:08Bridging the Pixel Gap in Browser Automation.
2:23How One Narrow Safety Rule Can Make an AI Less Safe Everywhere Else.
2:04Why splitting a chunk from its document makes it retrieve for the wrong question
4:20Three You Can Take Back. One You Can't.
2:21Why a 50-turn agent pays for the same screenshot 35 times unless it caches the pixels
1:53