YouTube SEO Title Generator (calibrated on channel data)

Produces ONE recommended title + 2 A/B alternates for a video idea, scored against an evidence-based rubric. Titles are in Spanish (channel language) unless the user requests otherwise.

Channel benchmark (empirical — source of truth for these rules)

Channel totals (last period): 3,582 thumbnail impressions · 5.1% CTR · 10,394 views · 0:37 avg view duration · 69.4 watch hours.

What the data proves

1. Conflict + specificity wins views. Top-7 titles (1,188–1,835 views) all contain drama (EXPLOTA, EMPUJÓ, obligar, humillar), a score/number (3-3, 0-2), or a bold stake (descenso por ley, dato que nadie vio).
2. Same event, better title = 13× views: "¡HUMILLADOS! Querétaro le gana 0-2 a Chivas en su casa" (1,339 views) beat "Cantó el GALLO: Querétaro humilla a Chivas en su casa" (103 views). Score + caps exclamation outperformed the softer phrasing.
3. Passive/descriptive titles die: "El partido se le complicaba a México frente a Colombia" → 4 views, 0% CTR. "El niño que salvó..." → 7 views. No stake, no conflict, no caps.
4. Emoji prefixes hurt: "🔴 Así quedó el debut..." → 18 views, 2.6% CTR (the 🟣/🔴 convention reads like clickbait spam).
5. ALL-CAPS power word works only with substance: EXPLOTA, HUMILLADOS, EMPUJÓ, GALLO (in caps with score) succeeded; "ABANDONARON" flopped (16 views) because the claim had no names/score attached.
6. CTR peaks on curiosity gaps: "El dato que nadie vio..." got 66.7% CTR (3 impressions — directional only). "Ditta EMPUJÓ..." combined curiosity + stake and delivered the best volume CTR (6.4% over 2,057 impressions).
7. Search still needs the entity front-loaded: highest-impression titles carried searchable nouns (Ditta, árbitro, Liga MX, Rafa Márquez, México, Chivas). Feed-driven views hide this, but impressions come from search/discovery.

Title rules (apply in order)

1. Keyword first (chars 0–40): start with the primary search entity — matchup ("México vs Perú"), player ("Gilberto Mora"), competition ("Liga MX Jornada 10"), or event. Never bury it mid-title.
2. One power word in ALL CAPS: a single verb/noun of conflict or emotion (ABUCHEADO, EXPLOTA, HUMILLÓ, RECHAZÓ, RENUNCIAR, TRIPLUTE, DEBE GANAR). Max 1–2 words; never the whole title.
3. Attach a concrete detail: final score, money figure, date, club, or role ("1-1", "12 millones", "3 de octubre", "Real Madrid").
4. Create the stake/curiosity gap: what it costs, what nobody saw, what happens next ("puede PERDER EL PUESTO", "el dato que nadie vio").
5. Length: 45–70 characters (YouTube cuts search display ~70; aim ≤ 60 so the full title survives mobile). Absolute max 100.
6. No emoji prefix, no hashtags in the title (hashtags go in the description). No ALL-CAPS sentences, no deception — the title must match the video (channel AVD is 0:37; broken promises kill retention).
7. Honest clickability: if two candidates score equal, prefer the one whose promise the script actually delivers.

Process

1. Read the topic/video idea: script hook, key facts, keywords.
2. Generate 5 candidate titles covering different angles (conflict, score+result, curiosity gap, stake/pressure, SEO-plain).
3. Score each 0–2 on five axes (max 10):
   • SEO — primary keyword in first 40 chars, entity people search
   • Power — ALL-CAPS conflict word with substance
   • Specificity — score/number/names/date present
   • Stake — curiosity gap or consequence
   • Format — 45–70 chars, no emoji/hashtag spam
4. Pick the highest scorer; make sure 2nd/3rd differ in angle so they are usable as A/B thumbnail-test variants.
5. Output the block below (in Spanish) and, when inside a research session, append it as section 6 of the topic file.

Output format

## 6. Título SEO para YouTube
**Título recomendado:** <best title> (<n>/10)
**Alternativo A:** <angle> — <title> (<n>/10)
**Alternativo B:** <angle> — <title> (<n>/10)
**Keywords para descripción:** <3-5 search phrases, comma separated>
**Por qué funciona:** <1-2 lines citing the data rules it hits>

Full upload package (title + description + tags + keywords)

Every video produced by the video-idea-research playbook needs the complete metadata block, not just the title. After the title block above, output:

## Paquete YouTube
**Título:** <recommended title>
**Descripción:** <2-4 lines: hook line first, summary of what the video shows,
1-2 questions to drive comments, 3-5 hashtags at the end>
**Tags:** <8-15 comma-separated tags: entities (player/club/matchup),
competition, format words ("resumen", "goles", "polémica"), Spanish + English>
**Keywords:** <8-15 search phrases: how people actually search —
"<player> hoy", "<team> vs <team>", "liga mx jornada N">

Rules:

• Title: the rubric above — keyword front-loaded, one ALL-CAPS power word, concrete detail, stake, 45-70 chars, no emoji/hashtag spam, honest promise.
• Description: first line repeats the title's promise with a twist (it shows in search); 2-4 short lines; a question to farm comments; hashtags last.
• Tags/keywords: mix narrow (exact matchup) + broad (competition) + format words; include the English spelling of names when the audience is bilingual.
• Backend seed option: generate_metadata() in Backend/gpt.py (title / description / 6 keywords / social post) — treat its output as a first draft and rewrite with these rules; the rubric-scored title wins.

Quick examples (before → after, using channel evidence)

• "🔴 Así quedó el debut de Rafa Márquez al frente de México" (18 views) → "El debut de Rafa Márquez CASI FUE UN DESASTRE con México" (stakes + caps)
• "Cantó el GALLO: Querétaro humilla a Chivas en su casa" (103 views) → "¡HUMILLADOS! Querétaro le gana 0-2 a Chivas en su casa" (1,339 views)
• "El partido se le complicaba a México frente a Colombia" (4 views) → "México vs Colombia: el empate que CASI se vuelve desastre del Tri"