Ask Claude "give me a viral YouTube video idea" and you'll get something like "10 productivity hacks that changed my life". It sounds fine. It's also what 10,000 other people got.
The problem isn't Claude. It's that Claude has no idea what is working on YouTube this week. So we gave it that data.
Short answer: the fastest way to go viral with Claude is to stop asking it to invent and start asking it to read: which small channels just got 50x their usual views, on what topic, with what title format, and what viewers asked for in the comments. Then let it adapt that to your channel.
Why "viral" is a number, not a feeling
At Returbe we track 1,552,869 videos from 36,037 channels, updated every day (23,268 new videos in the last 24 hours). About 203,000 of them are outliers: videos with at least 3x their channel's usual views.
That ratio is the only definition of viral that helps you. A 2-million-view video on MrBeast's channel tells you nothing. A 900,000-view video on a channel with 1,140 subscribers tells you the topic and the format did the work, not the audience. That's something you can copy.
Step 1: ask Claude what's breaking out in your topic
Once Claude is connected to Returbe (2 minutes, see the MCP setup), you just ask in plain words. Here's a real answer from November 19, 2026:

Look at the top line: "People Are Using Claude for Wild Things Now", 871,198 views on a channel that usually gets 640. That's 871x. A generic tutorial on the same topic would have done 640.
Step 2: find the pattern, not the video
One outlier is luck. Five outliers with the same shape is a pattern. Ask Claude:
"Group these outliers by title format and tell me which formats repeat across different channels."
Behind that, Claude calls title_formats, which tracks 16,442 title templates and how much each one lifts views. Some that are producing outliers right now:

"Something Strange Is Happening to [X]" is a good example: 5 videos from 5 different channels, 4.3x boost on average, and Dylan Davis got 268,000 views (33x) with "Something Strange Is Happening to People Who Work With AI".
Step 3: check the demand isn't a fluke
Before you film, ask Claude to check the topic with keyword_demand (what people type in YouTube's search bar, by country) and search_videos (who already covered it, in which languages, with how many views). If 40 channels covered it last week with a median of 800 views, move on. If 3 did and one got 200K, go.
Step 4: read what viewers want next
Claude can pull the most liked comments of the outlier (video_comments) and its transcript (video_transcript). That's where the next video hides: "Can you do this for X?" with 3,000 likes is a video idea with proof of demand.
Step 5: pick the title with data, not taste
Write 3 to 5 titles and ask Claude to score them. title_score compares each one to close titles in the database and how they performed. It isn't a real CTR, but it beats choosing the one that "sounds best".
Connect Claude to Returbe in 2 minutes
25 tools, the same database as the app, included from the Basic plan.
A full example: one prompt, one video idea
Here's a prompt that runs the whole workflow:
"I run a 4K-subscriber channel about AI tools. Find the long-form videos about Claude from the last 30 days that got 20x+ their channel's usual views. Group them by angle, check which angle has the fewest videos for the most views, read the top comments of the best one, and give me 3 titles in the formats that work right now."
With the data from this month, Claude lands on the "concrete surprising result" angle (a lockpicking robot at 689x, a 3D animation at 120x) rather than tutorials, and proposes titles in the "Something Strange Is Happening to…" and "I Tried… Using ONLY Claude" formats, both of which have recent outliers behind them.
What Claude still can't do for you
- Make the video good. The data gets you the right idea; retention is still your job.
- Predict. Outliers show what already worked. Trends move: the GTA 6 wave went from 62 videos a week to 2,507 in two months, and median views dropped as it got crowded.
- Replace your judgment on fit. An idea that works on a 1K channel about storms won't work on your cooking channel.
About the data
This article uses real data, not estimates. Every number comes from Returbe's own database of 1,552,869 YouTube videos and 36,037 channels, which tracks views day by day. The figures were pulled on November 19, 2026. The outlier score is a video's views divided by its channel's usual views (median of its recent uploads). Media outlets, AI-generated story farms, music channels and low-CPM markets are excluded unless stated. The screenshots show real answers from the Returbe MCP inside Claude, laid out for readability.
Can Claude make a YouTube video go viral?
Not on its own. Claude doesn't know which videos are performing this week, so its ideas are generic. Connected to a YouTube database through MCP, it can find videos that beat their channel's usual views by 10x or more and base your idea, format and title on them.
What is the Returbe MCP?
A connector that gives Claude 25 tools to read Returbe's YouTube database: outlier videos, competitors, niches, title formats, keyword demand, comments and transcripts. It works in Claude.ai, Claude Desktop and Claude Code, and is included from the Basic plan.
What counts as viral on YouTube?
We use the outlier score: a video's views divided by its channel's usual views. 3x is a real outlier, 10x is rare (under 4% of long-form videos). A 1,000-subscriber channel getting 50,000 views is more interesting to copy than a 5-million-subscriber channel getting 2 million.
Find your next video ideas
Returbe spots the videos doing 3x, 10x, 50x their channel's average, every day.
Start for free