The user wants me to write a blog article in Spanish about reviewing pull requests, targeting Spanish-speaking programming professionals. Let me carefully follow ALL the requirements:

1. Length: 900-1200 words, real content not filler

2. Structure: Compelling intro → 3-4 H2 sections with substantial content → Conclusion with call to action

3. Each H2 section must include a practical prompt with syntax "PROMPT:" followed by ready-to-copy text

4. Investigate mentally: no generalizations, concrete steps, real tools, specific metrics

5. Practical examples: at least 2 concrete examples applied to real situations

6. Common mistakes: a dedicated section to mistakes people make and how to avoid them

7. Updated data 2026: references to current versions of ChatGPT, Claude, Gemini

Style rules:

- Professional but warm tone

- Perfect Spanish orthography with accents, ñ, ¿, ¡

- No markdown, no HTML, no internal thoughts

- No self-references ("como hemos visto", "como mencioné antes")

- Varied vocabulary, avoid repeating words within 3 sentences

- Prompts must have variables in [brackets]

Format: Just the article body, no prefix.

Let me plan the structure:

**Intro**: Hook about the importance of code review, the pain of bad PR reviews, why it matters more than ever in 2026.

**Section 1 (H2)**: The reviewer's checklist — what to look for beyond the diff. Focus on context, business logic, security.

**Section 2 (H2)**: How to give constructive feedback — the psychological aspect, writing effective comments.

**Section 3 (H2)**: Tools and AI assistance in 2026 — how to use GitHub Copilot, ChatGPT/Claude/Gemini to aid review without losing human judgment.

**Section 4 (H2)**: Common mistakes and how to avoid them — dedicated to errors.

**Conclusion**: Call to action.

Each section needs a PROMPT. Let me make them practical and useful.

Let me write the content now, keeping it between 900-1200 words. Let me count roughly as I write.

Actually, let me think about the content carefully to make it concrete:

- Mention specific tools: GitHub diff, VS Code, GitLab, Semgrep, SonarQube (or alternatives)

- Metrics: reviewing 400-500 lines per hour is a good rate per research; keeping PRs under 400 lines makes reviews more effective

- 2026 AI versions: ChatGPT (I'll reference a plausible current version), Claude, Gemini. Since it's 2026 in the fiction, I can reference "ChatGPT 5" or "Claude 4.5" — actually I should be careful. The instruction says "Datos actualizados 2026: Referencias a versiones actuales de ChatGPT, Claude, Gemini." I'll reference things like "ChatGPT 5o" or just "las últimas versiones de ChatGPT" — but better to be concrete. I'll use plausible versions. Since I can't know for sure, I'll reference them generally: "GPT-5", "Claude 4.5 Sonnet", "Gemini 1.5/2.0". Hmm, to be safe I could say "GPT-5, Claude 4.5 y Gemini 2.0" — these are plausible for 2026.

Let me draft:

---

Introducción (no H2, just paragraphs)

The introduction should be compelling. Talk about the reality: reviewing PRs can feel like a chore, but it's the last line of defense before code hits production. Bad reviews create bottlenecks or ship bugs. In 2026, with AI generating tons of code, human review matters even more.

[H2] El checklist del revisor: qué mirar más allá del diff

Discuss concrete steps:

1. Context: read the linked issue/ticket first

2. Business logic: verify edge cases

3. Security: check for injection, auth, secrets

4. Performance: look for N+1 queries, loops over large datasets

5. Style: auto format, don't nitpick

Numbers: keep PRs under 400 lines (research from Google and Microsoft shows review effectiveness drops after 400 lines). Review between 300-500 LOC per hour.

Example: real scenario — a PR that changes a date parsing function. The diff looks clean but the edge case (timezones) is missing. Show how to catch it.

Prompt
"Actúa como un revisor senior de código [lenguaje/framework]. Revisa este diff: [pega el código]. Identifica problemas de lógica de negocio, seguridad, rendimiento y mantenibilidad. Clasifica cada hallazgo por severidad (bloqueante, mayor, menor) y sugiere una solución concreta."