How a Faceless Channel Made $39K in 3 Months With Claude and ChatGPT

In the opening minutes the video claims a faceless channel earned close to forty thousand dollars in about three months from Shorts alone, with no long-form and no face on camera. Three different faceless Shorts channels are shown side by side to frame this as a repeatable pattern rather than a one-off, and the promise is a step-by-step system you can copy with three tools.

VidIQ estimates for the last 90 days set the range: the first channel about $40,000 from 250 million views, the second close to $25,000, the third around $33,000, while one of them is also shown at 2 billion views and over $170,000 estimated revenue for the year. The three cases come from distinct niches, which the video uses to argue that the same mechanics travel across topics.

The why now section leans on three reasons and starts with a counterintuitive Google Trends read. Searches for YouTube automation and faceless YouTube have fallen sharply from 2023-2024 peaks to near a five-year low, so while the saturation narrative is loud, search interest is moving the other way. The implication is that perceived crowding and actual demand are diverging, opening room for newcomers.

The second reason is that Shorts can pay at scale. AIR Media-Tech analyzed 274 real channels with direct YouTube Analytics API access, covering over 3,000 channel-month points, and found Shorts RPM mostly between $0.07 and $0.20, with US-heavy audiences near $0.33. At that level 100 million views is about $33,000, which explains how 250 million maps to a $40,000-scale estimate, with the broader context that Shorts typically earns 3 to 14 percent of long-form per thousand views.

The third reason is that automation now covers the full stack. A couple of years ago almost every step was manual; today ChatGPT is assigned the channel identity, Claude the research and scripting, and a third tool the assembly. The video presents this split as a single pipeline and then walks through each piece live.

For niche selection the video recommends one lane and proves it: interesting-facts videos, short 30-second mini-docs built around a single story, person, event or surprising fact. Examples include a 180,000-subscriber history channel with 2, 4 and 7-million-view hits, a 320,000-plus science and space channel using the same format, and a third around strange facts and true stories, with the point that the template adapts to history, space, animals, tech, mystery or geography.

Step one is finding competitors, and three to five strong channels are enough. The goal is not to copy but to learn which topics get views, what the first seconds do, how long videos run, and which stories keep going viral. The practical method is simple: watch Shorts in your target niche, like, subscribe and comment selectively, and let the feed turn into a research engine that surfaces more viral examples in the same lane.

Step two turns those competitors into analysis inside Claude via the VidIQ connector. After installing the VidIQ Chrome extension and refreshing Claude, a YouTube Insights connection is authorized, then inside the chat the plus menu opens Connectors, VidIQ is chosen and competitor breakdown is selected. The channel identifier is copied from the competitor channel via More then Share Channel, pasted into Claude, the prebuilt prompt is added and sent, and Claude scans the catalog of Shorts to surface top performers.

When the breakdown returns, the chat is used to ask for the top five topics with links and summaries. Shown examples are familiar curiosity hooks: what the small hole in an airplane window does, what happens if an elevator cable snaps, why Formula 1 tires are completely smooth. Each has already drawn millions of views elsewhere, so the idea is to rework proven curiosity rather than guess from zero.

After choosing a topic, a fresh Claude chat turns it into a script. The on-screen prompt template is pasted together with the summary, and Claude is asked to rewrite the proven topic as a new roughly 30-second Short. The result arrives as a single voiceover text, then a second prompt maps that text scene by scene with voice, visuals and timestamps.

Build moves to RankReel, where everything lands on one edit track. The full script from Claude is pasted into RankReel, a natural fast-paced voice is selected and generated, and the audio drops onto the edit track. The sample opening heard in the demo explains that the small opening near the base of the window is not damage but a detail that helps keep comfort at high altitude, and the first seconds are played back in place.

Visuals are then covered clip by clip in two modes: find a matching stock shot or generate exactly the described shot with AI. The first scene description from Claude is generated as a 9 by 6 vertical video and added to the edit track with its native audio muted, the second description finds a good existing clip imported by link and set to 9 by 6 preview, with trimming at start and end to match Claude’s durations, and the loop repeats until the whole voiceover is covered. Captions close the edit: generate captions for the full voiceover, then style quickly with a preset, larger text and tweaks to font, shadow and color.

Before posting, the video covers channel setup. It suggests checking for an aged account you may have forgotten, arguing that history with YouTube is preferable to a brand-new creation, while a new channel should be warmed for a day or two by watching Shorts for about 30 minutes and interacting naturally. In YouTube Studio under settings then channel and feature eligibility, standard features are checked, a phone verification unlocks intermediate features, and advanced features can wait.

Branding is then handed to ChatGPT: a one-word name is brainstormed and Curio is chosen as the example, a profile image prompt with the name yields a simple readable icon, a banner is requested in the same colors and style with a single central line enlarged by about half, and a description prompt produces a tagline plus what the channel covers and what viewers should expect, ready to paste. The closing loop is to find what already works, make your own version, publish, measure and repeat with the same system.