Optimize YouTube and podcast metadata for search rankings

Produces a researched title, description, tags, and hashtags package for YouTube videos, podcast episodes, or short-form clips and Reels, or audits a live video or episode that isn't getting views and rewrites its metadata - covering six scenarios in total: three content types (standalone video, podcast episode, short-form clip) crossed with two states (new upload or existing/underperforming), each handled by its own mode. Two rules run through every mode: every keyword is researched via real web search rather than guessed or pulled from memory, and the output matches exactly what was asked - a request for just a title gets just a title, a request for the full package gets every numbered section, never more than what was asked for.

Use it when the user wants a title, description, tags, and hashtags package for a new upload, an audit of a live video or podcast episode that isn't getting views, show notes or timestamps or podcast episode metadata, or says something like "SEO my video," "audit my YouTube video," "write a podcast description," "why isn't my video ranking," "generate tags," or "fix my podcast SEO" - for a new, already-published, or short-form piece of content. It handles exactly one video, episode, or clip at a time, never an entire channel or playlist; if asked for something broader, the correct response is to ask which single item to start with, then restart the flow for that one item. It needs real web search and URL fetch tools to function at all, under whatever name the host environment gives them.

The flow runs in a fixed order: read the user's request and check scope first; classify content type and status, inferred from whether a URL, "no views," or "uploading" language was given, fetching live metadata immediately if a URL was provided rather than asking the user to repeat it, to land on one of six modes; find the target keyword, either confirmed from what the user already said or extracted from their topic and researched with a real search batch (the keyword itself, a year-qualified version, trending or reddit-style real phrasing, and for podcasts the guest's name), pulling 3-6 secondary long-tail keywords from that research and never fabricating search volume, view counts, or algorithm claims - a thin result set is treated as a legitimate signal of lower competition, not a failure; gather any remaining links, CTA goal, offer details, or podcast-specific info by scraping what's available first and asking one question at a time only when scraping and inference both dead-end, using an explicit placeholder marker rather than ever inventing a fake URL; then build only the sections the user actually asked for and run a quality checklist before sending. Titles across every mode follow one formula:

[Primary Keyword] : [Outcome or Benefit] + [Power Word / Number / Year]

held to 60-70 characters counted exactly, with the primary keyword in the first 4-5 words, one emotional hook, a year only when the research shows year-stamped titles are actually common, no all-caps except a single emphasis word, no misleading promise, and an A/B variant that uses a genuinely different hook rather than a reworded version of the same one, drawn from named patterns for how-to, list, story, question, geo, comparison, and podcast-guest framings.

Long-form modes generate 14-19 tags across seven required types (an exact-match primary keyword, broad umbrella topics, long-tail phrases from research, question-based phrasing, a branded or show-name tag, a year tag only when research supports it, and a geo tag for location-specific content), all lowercase except proper nouns, under 500 characters total, never repeating a phrase; short-form content instead uses 5-8 tags reused from its parent video where applicable. Hashtags run 5-8 for video and podcast content or 3-5 for short-form, with the first three chosen strategically since they surface directly below the title, placed only on the description's final line and never in the tags field, formatted in CamelCase, mixed across broad, specific, geo, and branded types, and varied rather than reused identically across every upload. The description itself is a fixed three-block structure: a roughly 80-120 word hook whose first sentence carries the target keyword naturally, a 4-6 paragraph body weaving in one secondary keyword per paragraph plus repeated geo mentions and a guest-bio paragraph for podcasts, and a footer with real resource links, a two-sentence subscribe CTA, platform-listen links for podcasts, and hashtags on the very last line, totaling 700-900 words for video, 800-1,000 for podcast, or 150-200 for a short. Chapters, required for the four long-form modes, are hard-capped at 6-10 markers with adjacent topics merged if there would be more, must start at 0:00 or YouTube ignores every chapter entirely, and use short 3-6 word action-oriented titles that actually reflect the content, marking guest intros and sponsor reads separately for podcasts. Geo-specific content adds location mentions 2-4 times in the description, location-qualified tags, at least one geo hashtag among the first three, and for bilingual channels one sentence in the local language alongside the English description. The six modes each have their own detailed output template: Mode A builds a full new-video package including title, A/B variant, description, five primary keywords, tags, hashtags, chapters, thumbnail overlay text, end-screen cards, a playlist SEO note, a pinned comment, and an end-screen script; Mode B audits an existing video with a scorecard across ten elements, detailed per-element findings, a fully rewritten metadata set, and a three-step post-fix action plan spanning immediate changes, a 48-hour follow-up, and a 7-day CTR check; Mode C and D mirror A and B for podcast episodes, adding guest bio and platform-link handling, sponsor-read placement in the timestamps, and leading the title with the guest's name whenever research shows people actually search for it; and Mode E and F mirror the same new-versus-existing split for short-form clips and Reels, with a tighter 150-200 word description, no chapters, a mandatory Shorts hashtag, and an optional cross-post caption for Reels or TikTok. A named list of specific failure modes calls out exactly the mistakes to avoid: generating tags from memory instead of running the research batch, inventing unverified numbers, hardcoding a stale year, shipping a description full of unfilled placeholders, missing the 0:00 chapter marker, skipping geo signals on local content, an A/B title that's just reworded rather than a genuinely different hook, and a generic pinned comment instead of one carrying a keyword and a value teaser.

Creators and channel managers who need a fully researched, correctly structured SEO package for a single video, podcast episode, or short-form clip, either building one from scratch before upload or diagnosing and rewriting one that already exists and isn't getting views.