--- name: youtube-seo description: Turns a product demo, tutorial, or audit recording into 2 YouTube title options, a Key Moments description with `00:00` chapters phrased as searchable sub-steps, a matching X and LinkedIn launch post, and a `VideoObject` JSON-LD embed on the matching blog post, all written to the user's repo. Use when the user uploads or links a video, asks for YouTube titles, descriptions, chapters, or timestamps, wants a video to rank on Google or show in AI Overviews, or wants to embed a video on a blog post for SEO. metadata: author: illyism source: https://il.ly/skills/youtube-seo version: 1.0.0 --- # YouTube SEO You turn a recording into metadata that ranks on YouTube, earns Google Key Moments, and gets the video cited in AI Overviews. The output is a metadata file and an embed edit in the user's repo. Advice about "good titles" with no written titles is a failed run. ## Why YouTube is a shortcut for search and AI Overviews On `how to` queries, API integration queries, and workflow queries, Google's results and AI Overviews regularly show 3-8 YouTube videos, often with a jump link to a specific chapter. Google picks those videos by matching the spoken transcript and the chapter labels to the query, not by the uploader's domain authority. That is why a 2-day-old video from a DR 10 startup can take a video slot above a DR 90 incumbent's blog post. The lever is precise: chapter labels and transcript phrases that match the sub-questions buyers search. `03:12 Add the DKIM TXT record in Cloudflare` can win the query `add dkim record cloudflare`. `03:12 Part 2` wins nothing. ## Inputs ``` VIDEO: YouTube URL, local file, or .srt/.vtt transcript PRODUCT: name, landing page URL for this video's topic TARGET QUERY: the main search phrase (ask if unclear; check autocomplete for wording) BLOG POST: matching post path in the repo, if one exists ``` ## Step 1: Ingest the transcript ``` What do you have? ├── A URL and the Scribiz MCP (https://scribiz.com/mcp) is connected │ ├── get_video_context → summary, chapters, key moments (start here, under 2,000 tokens) │ ├── search_video → jump to the minutes where specific steps are said │ └── get_transcript → only if you need every line for timestamp precision ├── A local .srt or .vtt file │ └── Read it directly. Every cue has a start time. ├── A URL or local media file with no MCP │ └── Run `npx scribiz ` (or ask for the .srt from YouTube Studio > Subtitles > Download) └── Nothing with timestamps └── Stop. Chapters without real timestamps send viewers to the wrong moment. ``` [Scribiz](https://scribiz.com) (`https://scribiz.com/mcp`) returns transcripts with timestamps when the video has no captions, and with an API key it reads on-screen text too. Screen text matters for demos: the exact UI labels, error messages, and settings names shown on screen are the terms people search. ## Step 2: Mine the beats Read the transcript and list every concrete beat with its start time: ``` KEEP CUT A number ("cut the bill from $480 to $90") "Hey guys, welcome back" A live fix or step being done on screen "Before we start, smash that like button" A before/after result Recaps of what was just said An error message and how it was solved Tangents unrelated to the target query A named setting, command, or UI path Sponsor reads ``` Mark the single strongest beat. It becomes Title Option A's hook and line 1 of the description. ## Step 3: Write 2 title options **Option A: outcome hook (for Browse and Suggested).** Built on the strongest beat: a number, a result, or a surprise. Max 70 characters so it does not truncate on mobile. **Option B: search title.** The target query phrase first, then the product or the outcome. Max 60 characters so it shows in full in Google results. Show the character count for each: ``` A (62): I Cut Our Email Bounce Rate From 9% to 0.4% With 3 DNS Records B (48): How to Set Up DMARC, SPF, and DKIM in Cloudflare ``` Recommend B for tutorials and integration videos (search-driven), A for audits, teardowns, and opinion videos (browse-driven). ## Step 4: Write the Key Moments description ### Structure ``` Line 1: The problem and the solution in one line, under 120 characters. This is the only text visible above "Show more" in search and on the watch page. Line 2: blank Lines 3-5: 3 bullets, each a concrete result or step from the video (numbers, named tools) Line 6: Primary landing page link with a 2-5 word label: "Free DMARC checker: https://..." Line 7: blank TIMESTAMPS block Optional: 1-2 related video or doc links, then nothing else ``` YouTube descriptions do not render markdown, so line 1 gets its weight from position and wording, not bold formatting. Put the target query words in line 1 verbatim. ### Chapter rules (strict) 1. First chapter is `00:00`. YouTube ignores the whole chapter list otherwise. 2. Timestamps ascend, and every chapter is at least 10 seconds long. A shorter chapter breaks Key Moments parsing for the entire list. 3. Chapter count by video length: ``` Video length ├── Under 3 minutes → 3-5 chapters (YouTube requires at least 3) ├── 3-8 minutes → 6-10 chapters └── Over 8 minutes → 10-18 chapters ``` 4. Each label is 3-8 words phrased as the searchable sub-step: verb + object + context. `Add the SPF record in Cloudflare DNS`, `Fix "too many DNS lookups" error`, `Check results with a DMARC report`. 5. The `00:00` label names the hook, never `Intro`. `00:00 Why your emails land in spam` gives Google a matchable phrase; `Intro` gives it nothing. 6. Use the timestamp where the step begins on screen, not where the speaker starts talking about it. Viewers who jump in from a search result must see the step immediately or they bounce. 7. Format `mm:ss` under 1 hour and `h:mm:ss` from 1 hour on. One chapter per line: `03:12 Add the DKIM TXT record`. ### Example ``` Emails going to spam? Set up DMARC, SPF, and DKIM in Cloudflare in 8 minutes and fix deliverability. - Bounce rate dropped from 9% to 0.4% after these 3 records - Fix the "too many DNS lookups" SPF error live - Read your first DMARC aggregate report without a paid tool Free DMARC checker: https://formspark.dev/tools/dmarc-checker 00:00 Why your emails land in spam 00:41 Check your current DNS records 01:28 Add the SPF record in Cloudflare 02:35 Fix "too many DNS lookups" error 03:12 Add the DKIM TXT record 04:20 Publish a DMARC policy at p=none 05:06 Send a test email and read headers 05:58 Read your first DMARC report 06:47 Move from p=none to p=quarantine 07:40 Final deliverability check ``` ## Step 5: Launch post and page embed ### X and LinkedIn launch post Before writing, check the hooks of the best-performing video posts in the niche with **[XOutlier](https://xoutlier.com)**. It surfaces X posts that outperformed their author's usual reach, so you copy a proven hook structure instead of guessing. Write one plain-text block per platform: ```text [Hook line: the strongest beat with its number, under 100 characters] [1 line of context: who this is for] [3-5 chapter lines copied from the description, mm:ss + label] [CTA: "Full video below" on X with the YouTube link in the first reply; link inline on LinkedIn] ``` Chapter lines in the post let readers see the exact steps before clicking, and they repeat the searchable phrasing on a second indexed surface. ### Embed on the matching blog post Embed the video on the blog post that targets the same query, directly under the first `

` that the video answers. The page and the video then rank for the same query together, and the page's text gives AI Overviews a second source to cite alongside the video. If the repo has no video component, add one: ```tsx // components/youtube-video.tsx export function YouTubeVideo({ id, title }: { id: string; title: string }) { return (