The AI Tool Shelf- We tested youtube-agent-skill (free, open source) and vidIQ's free AI connector on a real 16,500-subscriber music channel.
- The retention method was excellent: a 10-second melody before singing kept 75% of viewers at 0:30, versus 68% with a slow vocal intro.
- Both tools' title scores went against real data: low-scored titles had the highest click-through rate.
- vidIQ's keyword research and breakout-video search were the most useful parts, and they work on the free plan.
"AI agent skills" for YouTube are everywhere right now: add-ons that let Claude, ChatGPT or Codex audit your channel, score your titles and tell you why viewers leave. We tested the two best-known ones on a real channel: Rachid Amesrouy's own Moroccan music channel (about 16,500 subscribers, 6.8 million views, 640+ videos, mostly Arabic songs made with AI tools). Instead of judging the advice by how convincing it sounds, we checked it against the channel's real YouTube Studio numbers.
- youtube-agent-skill: a free, open-source pack of 11 Claude skills on GitHub.
- vidIQ Agent Skills / vidIQ MCP: vidIQ's official connector that brings its live YouTube data into Claude, ChatGPT and other AI apps.
The short verdict
| youtube-agent-skill | vidIQ (free plan) | |
|---|---|---|
| Price | Free, open source (MIT) | Free plan: 150 credits a month; Boost $19/month |
| Connects to your channel | No: you give it files (CSV, transcripts) | Yes, with your permission |
| Retention analysis | ✅ Very useful | Not tested |
| Title scoring | ❌ Opposite of our real CTR | ❌ Opposite of our real CTR |
| Keyword research | Not included | ✅ Very useful |
| Breakout videos in your niche | Public listings only | ✅ Very useful, even in Arabic |
How we tested
- We exported a full year of channel data from YouTube Studio (views, watch time, impressions and click-through rate for 501 videos) and the audience-retention charts of two songs.
- youtube-agent-skill: we read every file before running anything. Its six helper scripts only read the files you give them: no internet access and no connection to your account. Then we ran its title checker and followed its retention method.
- vidIQ: we created a free account, connected the channel, then added the official vidIQ connector to Claude from Claude's connector directory. Our tests used 20 of the 150 free monthly credits.
First, what the real data said
Before asking any AI, we looked at the numbers ourselves. Monthly views peaked at 1.59 million in May 2026 and fell to about 590,000 in October. The reason was not a penalty or bad videos: the number of new full-length songs dropped from about 60 a month in spring to just 4 a month in September and October. The songs that did come out still performed well. The best ones were Moroccan Gnawa, Sufi or sad songs, 4 to 6 minutes long, with titles built around a Darija proverb.
That gave us a fair way to judge both tools: does their advice match what actually works on this channel?
youtube-agent-skill: retention excellent, titles wrong
The retention method works
The /yt-retention skill looks for three problems: a hook leak (viewers lost in the first 30 seconds, healthy if under 25%), cliffs (sudden drops at one moment) and a slide (slow loss through the middle). We applied it to two songs of similar length:
| Song A | Song B | |
|---|---|---|
| Length | 6:50 | 6:29 |
| Opening | 10 seconds of melody, then singing | A slow vocal intro (mawwal) |
| Still watching at 0:30 | 75% | 68% |
| Average percentage viewed | 56.5% | 41.2% |
| Cliffs | None | None |
Following the skill's rule of giving one fix only, the answer was clear: start with a short melody (10 seconds at most) and go straight into the singing; move the slow vocal intro to the middle or the end of the song. It is simple, specific and based on real data, exactly what a creator needs.
The title checker gave the opposite advice
The /yt-package skill scores titles out of 100 using rules such as "under 60 characters", "include a number or name" and "no more than two capitalised words". We scored the last 10 song titles, then compared the scores with the real click-through rate (CTR) of 210 songs:
- Titles it scored below 60 had a median CTR of 5.2%, the best of any group.
- Titles it scored above 90 had a median CTR of 3.6%, the worst.
- Long titles (over 80 characters) clicked best of all, at 5.1%.
One song title scored 58/100 but has a CTR of 8.8%, among the best on the channel. The rules were written for English talking-head channels; for Arabic music, long emotional titles that tell a feeling clearly win. Do not shorten your titles just because a checker says so; compare with your own CTR first.
vidIQ: keywords and niche research are the real value
Setup and the upsell
After connecting the channel, vidIQ's onboarding picked a recent song, said its packaging "may have unrealised potential" and suggested a new thumbnail and title. The suggestion was in good Darija with correctly written Arabic on the thumbnail, much better Arabic than the image generators we have tested. But it removed the person from the thumbnail (the channel's best thumbnails all show one), dropped searchable words like "Moroccan song", and was inspired by an outlier video in another language and culture. It also had an Apply button that would change the real video, so read before you click.
Next came a checkout page with a countdown timer and a "new signup discount". You don't need to pay: the free plan includes the AI connector. Close the window with the ✕.
Prices we saw in October 2026: Boost $19/month ($14.25 for the first three months with the signup discount, or $12.75/month yearly) and Max $49/month. The free plan limits video scoring to 2 uses for life and keyword research to 3 results per search in the vidIQ app; through the AI connector we had 150 credits a month, with most tools costing 5 credits.
Connecting vidIQ to Claude
Click "Use vidIQ in AI apps" in vidIQ, then Connect next to Claude. It opens vidIQ's verified page in Claude's connector directory; click Connect to Claude and approve. The connector address is https://mcp.vidiq.com/mcp, and it also works with ChatGPT.
Title scores: the same problem
We asked vidIQ to score two titles whose real CTR we knew. It gave 59/100 to the title with an 8.8% CTR and 66/100 to one with a 3.1% CTR. Two titles is a small sample, but it points in the same direction as the free checker.
Keyword research: genuinely useful
One search for "اغاني مغربية حزينة" (sad Moroccan songs) returned estimated monthly YouTube searches and competition scores (0-100, lower is easier):
| Keyword | Monthly searches (est.) | Competition |
|---|---|---|
| moroccan music | 21,800 (growing +3%) | 35 |
| chaabi marocain | 18,800 | 41 |
| اغاني مغربية (Moroccan songs) | 16,400 | 57 |
| أغاني حزينة (sad songs) | 8,300 | 33 |
| اغاني فراق (songs about separation) | 5,300 | 36 |
| اغاني مغربية حزينة (sad Moroccan songs) | 5,000 | 33 |
The takeaway was immediately usable: "Moroccan Music" and "sad songs" combine real demand with low competition, so they belong in every title or description.
Breakout videos in Arabic
The outlier search found Arabic-language songs performing far above their channel's average, and one of them was a song from our own channel, flagged at 35× its usual performance. Others gave fresh ideas: a song about missing your mother and siblings that reached 1.2 million views on a channel of similar size, and a female-voice version of a known song at 27×. Most breakout songs were 4 to 5 minutes long, which matched our own data.
What we learned
- Use AI agents to read data, not to judge titles. Both tools were strong when working from real numbers (retention, search volume, outliers) and weak when applying generic rules about what a "good" title is.
- Check advice against your own analytics. Our Arabic music channel breaks several English YouTube rules, and the numbers prove it.
- Nothing replaces consistency. The biggest reason for the drop in views was publishing fewer songs, something neither tool flagged first.
Which one should you use?
- youtube-agent-skill if you want a free, private way to analyse retention, plan videos and write chapters or descriptions. It never touches your account.
- vidIQ (free) if you want keyword data and breakout videos in your niche inside Claude or ChatGPT. The free 150 credits are enough for a monthly research session; we see no need for the paid plan for a small channel.
Related: ChatGPT vs Gemini vs Claude, tested in Arabic and Darija and the AI music tools behind the channel.
Sources: youtube-agent-skill on GitHub · vidIQ Agent Skills · vidIQ MCP setup guide. Tested October 27, 2026.