YouTube Competitor Analysis with Claude: 3 Workflows on Real Channel Data

Most YouTube competitor analysis stops at subscriber counts. That tells you who is big, not what is working. The useful questions are different: which videos beat the channel's own average, what do those titles have in common, and what do viewers keep asking for in the comments?

Below are three workflows we ran on live public data with Claude and the Captapi MCP server. Every number comes from a real run on 28 October 2026, including what each run cost.

Disclosure: Captapi is our product. The same workflows work with any tool that returns public YouTube data to Claude.

What you need

• Claude (claude.ai, Claude Desktop, or Claude Code) or another MCP-capable agent such as Cursor.
• The hosted Captapi MCP server: https://api.captapi.com/mcp. In claude.ai, add it under Connectors as a custom connector and sign in; there is no key to paste. In Claude Code or Cursor, pass your key in an Authorization header. Setup steps: captapi.com/docs/integrations.
• A free account (100 credits). All three workflows below together used 39 credits.

Workflow 1: Compare three competitor channels (15 credits)

Prompt

Use Captapi to compare these 3 YouTube channels: [channel 1 URL] [channel 2 URL] [channel 3 URL].
For each: subscribers, last 10 videos with views, and which video beat the channel's own average the most.
Then read the top comments on each channel's best video and tell me what viewers keep asking for.
Finish with 3 video ideas nobody in this group has made yet. Keep it to one comparison table plus short bullets.

Claude made 9 calls: channel details and recent videos for each channel, then comments on each channel's best video. The run took about a minute and a half.

Result (three channels in the "Claude for creators" niche)

The biggest channel had the smallest breakout. The two smaller channels each had a video that did three to almost four times their own average. Size and momentum are different things, and you only see it when every channel is measured against itself.

The comments added the part numbers can't:

• Kallaway's viewers asked for a video on conversion funnels, and separately for brand deals and sponsorships.
• Alex Christian's viewers asked where to get footage and how copyright applies.
• Michele Torti's viewers were stuck on deployment: hosting, OAuth and credentials for client automations.

Those asks became three video ideas none of the channels had covered in their last 10 uploads.

Workflow 2: Find a channel's outlier videos (2 credits)

Prompt

Use Captapi to get the last 30 videos from [channel URL]. Mark every video with at least 2x the channel's
median views as an outlier. Then tell me what the outlier titles have in common and give me 5 title formulas
I could reuse in my own niche. Use one small markdown table plus short bullets.

One call returned all 30 uploads. For Kallaway, the median was 59,176 views, so the outlier bar was 118,352. Six videos cleared it. The top one, "How Social Media Algorithms Actually Work (And How to Beat Them)", did 1,216,736 views, 20.6× the median.

Patterns Claude pulled from the six outlier titles:

• They promise the hidden mechanism. Four of six frame the topic as science or psychology ("actually works", "neuroscience", "psychology").
• They pick broad topics that apply to every creator: algorithms, storytelling, hooks, audience.
• A parenthetical adds a second payoff, such as "(And How to Beat Them)".

Use the median, not the average. One viral video inflates the average and hides every other outlier.

Workflow 3: List the tools your competitors talk about (22 credits)

Prompt

Use Captapi to summarize the last 5 videos from [channel URL]. For each: one-line summary and every tool or
app they mention. Then list the tools mentioned more than once. Use one small markdown table plus short bullets.

Claude first requested five summaries, saw that summaries don't name tools reliably, and switched to reading the five full transcripts. On Michele Torti's channel, Claude Code appeared in 4 of 5 videos; n8n, Retell AI and Skool in 2 each.

It also flagged its own uncertainty: auto-captions mishear brand names, so it listed which names it had inferred. Keep that habit when you run this on your niche.

What this can and can't tell you

• Public data only. You get views, titles, publish dates, comments and transcripts. You don't get watch time, CTR, impressions or revenue for channels you don't own. For your own channel, YouTube Studio is still the source.
• Small samples. Ten or thirty videos is directional. Say so when you share the results.
• Very new videos haven't collected their views yet. Leave out the last two or three days when you compute a baseline.

Run it on your niche

1. Pick three channels your audience also watches.
2. Run Workflow 1 to see who is breaking out and what their viewers want.
3. Run Workflow 2 on the channel with the biggest outlier multiple.
4. Turn the repeated comment questions into your next three videos.

Create a free Captapi account (100 credits, enough for all three workflows twice). Prefer reusable instructions? The Captapi agent skills package these workflows as outlier-posts, comment-mining and competitor-teardown (source on GitHub).