Best ChatGPT Prompts for SEO (2026)

AI models are genuinely good at pattern-matching tasks over structured data — exactly the kind of work that sits at the core of SEO. Grouping 500 keywords by intent, generating 10 title tag variants with specific character counts, mapping internal link opportunities across 30 URLs, generating JSON-LD markup for an article or product — these are high-volume, low-creativity tasks where a good prompt with the right constraints saves hours of work.

They are not good at tasks requiring real-world search data: they cannot pull live SERPs, they do not have accurate volume or difficulty figures, and their knowledge of current algorithm changes depends entirely on training cutoff. Use them for synthesis, structure, and generation. Use your SEO platform (Ahrefs, Semrush, Search Console) for actual data. The workflow that works: export your data from your SEO tool → paste it into the prompt → use the model to organize, analyze, or generate output → paste the result back into your CMS or sheet.

For more on the role prompts that get the best SEO outputs from AI, see role prompts for SEO specialists. For a head-to-head comparison of which model is best for specific SEO research tasks, see Claude vs Gemini for SEO research 2026.

Keyword clustering — grouping a flat list of keywords by topic, intent, and parent page — is one of the highest-value AI SEO tasks. It used to take an SEO analyst a full day to cluster 500 keywords. A well-structured prompt does it in under a minute with roughly the same accuracy for the easy cases (and flags the ambiguous ones for human review).

**Prompt 1 — Basic intent-based clustering:** ``` You are an expert SEO strategist. I will give you a list of keywords. Group them into clusters where each cluster represents a single searcher intent that could be satisfied by one target URL. For each cluster: (1) suggest a cluster name, (2) list the keywords, (3) label the intent as Informational / Navigational / Transactional / Commercial Investigation, (4) suggest a target URL slug. Keywords: [PASTE YOUR KEYWORD LIST HERE — one per line] Output as a markdown table with columns: Cluster Name | Intent | Keywords | Suggested Slug. ```

**Prompt 2 — Parent/child clustering with pillar mapping:** ``` Act as a senior content strategist building a topic cluster architecture. Given the keyword list below, identify: (1) 3-5 pillar topics broad enough to be cornerstone pages, (2) all supporting cluster keywords for each pillar, (3) for each cluster keyword, indicate whether it belongs as a subsection of the pillar or as a separate spoke page. Keywords: [PASTE YOUR KEYWORD LIST HERE] Output format: one section per pillar, with a table of cluster keywords, intent, and page type (pillar section vs spoke page). ```

**Prompt 3 — Deduplication and cannibalization check:** ``` You are an SEO auditor. Review the following keyword list and identify any groups of keywords that are so similar in intent that targeting them on separate URLs would cause cannibalization. For each cannibalization risk: list the conflicting keywords, explain why they conflict, and recommend which one URL should target all of them. Keywords: [PASTE YOUR KEYWORD LIST HERE] Be specific. Do not group keywords that are similar in topic but distinct in intent. ```

**Prompt 4 — Cluster labeling for a spreadsheet export (Gemini 2.5 Pro or GPT-5 recommended for large lists):** ``` I have a spreadsheet of keywords. For each keyword, output a single cluster label (2-4 words, title case) and an intent label (Info / Transactional / Commercial / Nav). Return results as a JSON array: [{"keyword": "...", "cluster": "...", "intent": "..."}]. No commentary, just the JSON. Keywords: [PASTE YOUR KEYWORD LIST HERE] ```

**Prompt 5 — Long-tail opportunity extraction:** ``` From the keyword list below, identify the 20 highest-opportunity long-tail keywords (4+ words, specific intent, likely lower competition). For each: (1) state the keyword, (2) describe the specific searcher intent in one sentence, (3) suggest what format the target page should take (listicle, how-to guide, comparison, tool/calculator, definition, etc.). Keyword list: [PASTE YOUR KEYWORD LIST] ```

SERP intent analysis means classifying a keyword — or a set of keywords — by what the searcher actually wants to accomplish, then cross-referencing that against what Google currently ranks. This informs content format, page structure, and whether you should target a keyword with a new page or optimize an existing one.

**Prompt 6 — Batch intent classification:** ``` You are an SEO specialist. Classify each of the following keywords by search intent using the standard taxonomy: Informational (learning), Navigational (finding a specific site), Commercial Investigation (researching before buying), Transactional (ready to buy/act). For ambiguous keywords, note the dominant intent and the secondary intent. Keywords: [PASTE YOUR KEYWORD LIST] Output as a table: Keyword | Primary Intent | Secondary Intent | Notes. ```

**Prompt 7 — Content format recommendation from intent:** ``` For each keyword below, analyze the likely SERP intent and recommend the best content format to rank for it. Format options: featured-snippet definition, step-by-step how-to, comparison table, listicle, in-depth guide, product/service page, tool or calculator, FAQ, video. Provide a one-sentence rationale for each recommendation. Keywords: [PASTE YOUR KEYWORD LIST] ```

**Prompt 8 — Competitor gap analysis from SERP data (paste SERP results manually):** ``` I am targeting the keyword "[TARGET KEYWORD]". Below are the titles and meta descriptions of the top 10 organic results. [PASTE TOP 10 SERP TITLES AND META DESCRIPTIONS] Analyze: (1) What subtopics do most ranking pages cover? (2) What angle or format dominates? (3) What is missing — what could a new piece cover that the current top 10 does not address? (4) What is the implied content depth (word count estimate) based on the titles? ```

**Prompt 9 — Intent mismatch detection:** ``` Review the following list of keywords and the URLs my site currently ranks for them. Identify any intent mismatches — cases where the current ranking URL serves a different intent than the keyword implies, which may be limiting ranking potential. Data: [KEYWORD] — [CURRENT RANKING URL] — [CURRENT POSITION] [KEYWORD] — [CURRENT RANKING URL] — [CURRENT POSITION] (continue for all keywords) For each mismatch: explain the disconnect and suggest whether to (a) optimize the existing page, (b) create a new page, or (c) accept the current ranking as good enough. ```

Title tags are one of the most direct on-page ranking signals. They also drive click-through rate — a well-written title can lift CTR 20-40% without changing your position. The prompts below generate multiple variants so you can test. For the full picture of prompt structure principles, see prompt engineering for content marketing.

**Prompt 10 — Batch title tag generation with character limits:** ``` You are an expert SEO copywriter. Write 5 title tag variants for the following page. Requirements: (1) Primary keyword appears within the first 60 characters, (2) Total length is 50-60 characters, (3) Each variant uses a different structural format (number list, how-to, question, comparison, direct statement), (4) No clickbait. No vague promises. Page topic: [DESCRIBE THE PAGE] Primary keyword: [PRIMARY KEYWORD] Secondary keywords to include where natural: [SECONDARY KEYWORDS] Brand name (append with ' | ' or ' — ' if it fits): [BRAND NAME] Output each variant on a new line with character count in parentheses. ```

**Prompt 11 — Title tag rewrite for low-CTR pages:** ``` My page is ranking position [POSITION] for "[KEYWORD]" but has a low CTR of [CURRENT CTR]%. The current title tag is: "[CURRENT TITLE]" Rewrite it with 5 alternatives. Goals: (1) Better match the searcher's intent, (2) Add a specific benefit or number where possible, (3) Stay under 60 characters. Note what problem each variant addresses in one sentence. ```

**Prompt 12 — Emotional/power word injection:** ``` Rewrite the following title tag to increase click-through rate by adding one emotional or power word that is still accurate and not misleading. Keep the keyword and character count. Provide 3 variants. Original title: [CURRENT TITLE TAG] Primary keyword: [KEYWORD] Max characters: 60 ```

**Prompt 13 — Year/freshness signal variants:** ``` I need 3 variants of this title tag that include a year or freshness signal (e.g., '2026', 'Updated', 'Latest'). The keyword must remain. Keep under 62 characters. Base title: [CURRENT TITLE] Keyword: [KEYWORD] ```

**Prompt 14 — Competitor differentiator titles:** ``` My competitors rank for "[KEYWORD]" with these titles: [LIST 5 COMPETITOR TITLE TAGS] Write 5 title tag alternatives for my page that differentiate from these — use a unique angle, format, or claim that none of the above use. Keep under 60 characters and include the primary keyword. ```

Meta descriptions do not directly influence rankings, but they are your 150-character pitch in the SERP. They affect CTR, and CTR is a measurable feedback signal. The goal is to match the description tightly to the searcher's intent, add one specific benefit, and include a soft CTA. See best ChatGPT prompts for marketers 2026 for related copywriting prompts.

**Prompt 15 — Standard meta description generation:** ``` Write 5 meta description variants for the following page. Requirements: (1) 140-155 characters each, (2) Includes the primary keyword naturally, (3) States one clear benefit to the reader, (4) Ends with a soft call to action, (5) No keyword stuffing. No vague filler phrases. Page topic: [DESCRIBE PAGE] Primary keyword: [KEYWORD] Key benefit of the page: [WHAT DOES THE READER GET FROM THIS PAGE?] ```

**Prompt 16 — Meta descriptions for commercial/transactional pages:** ``` Write 5 meta description variants for a [product/service] page. Each should: (1) Lead with the core value proposition, (2) Include a differentiator (price, speed, guarantee, number of options, etc.), (3) End with a transactional CTA, (4) Stay within 155 characters. Product/service: [NAME] Key differentiator: [WHAT MAKES IT STAND OUT?] Primary keyword: [KEYWORD] ```

**Prompt 17 — Batch rewrites for low-CTR meta descriptions:** ``` Below are meta descriptions for pages that have lower CTR than expected. Rewrite each with a single improved variant. For each rewrite, note the specific problem with the original (too vague / missing benefit / too long / passive voice / no CTA) and how you fixed it. [PAGE TITLE] — [CURRENT META DESCRIPTION] [PAGE TITLE] — [CURRENT META DESCRIPTION] (continue for all pages) ```

**Prompt 18 — FAQ-intent meta descriptions:** ``` This page answers the question: "[TARGET QUESTION KEYWORD]". Write 5 meta descriptions that are structured as a direct answer preview — i.e., they begin with the answer (or a summary of it) so the searcher immediately sees value in clicking. Stay under 155 characters. Include the question keyword. ```

**Prompt 19 — A/B test pair generation:** ``` Create 2 meta description variants for an A/B test. Variant A should emphasize [BENEFIT 1]. Variant B should emphasize [BENEFIT 2]. Both must include the keyword "[KEYWORD]", stay under 155 characters, and end with a CTA. Highlight the key difference between the two approaches. ```

A content brief is the document a writer uses before drafting a page. A good brief covers: target keyword and intent, recommended structure (H2s and H3s), word count range, competing pages to beat, questions to answer, and internal/external linking guidance. AI models are genuinely excellent at generating first-draft briefs from keyword + SERP data you paste in.

**Prompt 20 — Full content brief from scratch:** ``` You are a senior content strategist. Generate a full SEO content brief for the following target keyword. The brief should include: 1. Target keyword and 5 semantically related secondary keywords 2. Search intent analysis (what the searcher wants to accomplish) 3. Recommended page title (H1) 4. Recommended H2 structure (6-10 sections) with a one-sentence description of what each section should cover 5. Recommended word count range 6. 5 questions the page must answer to satisfy the reader fully 7. Content format recommendation (guide, listicle, comparison, etc.) with rationale 8. 3 external sources worth linking to 9. 2-3 internal link opportunities (describe the type of page, not specific URLs) Target keyword: [YOUR KEYWORD] Target audience: [WHO IS READING THIS?] Site niche: [YOUR NICHE/INDUSTRY] ```

**Prompt 21 — Content brief from competitor analysis:** ``` I want to outrank the following pages for "[TARGET KEYWORD]". Here are their H2 structures: Page 1: [LIST H2s] Page 2: [LIST H2s] Page 3: [LIST H2s] Create a content brief that: (1) Covers all topics they cover (table stakes), (2) Adds 3-5 sections or angles they are missing, (3) Recommends the optimal order for the sections, (4) Suggests word count that justifies a new page vs. consolidating with an existing one. ```

**Prompt 22 — Brief for a featured snippet opportunity:** ``` This keyword currently shows a featured snippet in Google: "[KEYWORD]". The current snippet text is: "[PASTE CURRENT SNIPPET IF VISIBLE]". Create a content brief specifically optimized to capture this featured snippet. Include: (1) The exact paragraph format and length that typically wins for this type of query, (2) H2 wording recommendations that signal the answer directly, (3) What the full page needs around the snippet paragraph to establish authority. ```

**Prompt 23 — Content refresh brief for an existing page:** ``` I have an existing page ranking position [POSITION] for "[KEYWORD]". The page was last updated [DATE]. Below is its current H2 structure: [LIST CURRENT H2s] Create a content refresh brief that: (1) Identifies sections to update (stale data, outdated tactics), (2) Identifies gaps to add, (3) Identifies sections to cut or consolidate, (4) Estimates the impact of each change on ranking potential. Keep the output actionable — specific, not generic. ```

**Prompt 24 — Topic cluster spoke-page brief:** ``` I am building a topic cluster. The pillar page covers: "[PILLAR TOPIC]". This brief is for a spoke page targeting: "[SPOKE KEYWORD]". Create a brief for the spoke page that: (1) Clearly delineates scope so it doesn't cannibalize the pillar, (2) Specifies exactly where to link back to the pillar (with anchor text suggestions), (3) Identifies 2-3 other spokes to cross-link to, (4) Recommends the content depth appropriate for a spoke vs. a pillar. ```

Internal linking is one of the most consistently under-executed on-page SEO tactics. Most sites link reactively — when a writer happens to remember another page. AI models can systematically map link opportunities from a content inventory, which turns a reactive process into a structured one.

**Prompt 25 — Internal link opportunity map from a URL list:** ``` You are an SEO specialist. Below is a list of pages on my site with their target keywords. For each page, identify 3-5 natural internal link opportunities from the other pages in the list. For each opportunity: (1) Source page (where the link goes from), (2) Target page (where the link points to), (3) Suggested anchor text, (4) One sentence of context explaining where in the source page the link belongs. Page list: [URL] — [TARGET KEYWORD] [URL] — [TARGET KEYWORD] (continue for all pages) Output as a table: Source Page | Target Page | Anchor Text | Placement Notes. ```

**Prompt 26 — Anchor text variation generator:** ``` I need to link to a page targeting the keyword "[TARGET KEYWORD]" from multiple locations across my site. Generate 15 anchor text variations that: (1) Sound natural in body copy, (2) Vary between exact match, partial match, and descriptive anchors, (3) Avoid over-optimization. Group them by anchor type. ```

**Prompt 27 — Orphan page linking plan:** ``` The following pages on my site have no internal links pointing to them (orphan pages). For each, suggest 3 existing pages on my site that should logically link to it, with anchor text suggestions. Base your suggestions on the target keywords provided. Orphan pages and their keywords: [URL] — [TARGET KEYWORD] [URL] — [TARGET KEYWORD] Existing site pages that could be sources: [URL] — [TARGET KEYWORD] [URL] — [TARGET KEYWORD] ```

**Prompt 28 — Hub page internal link structure:** ``` I am creating a hub page for the topic cluster "[CLUSTER TOPIC]". The hub page should link to 8-12 spoke pages. Below is my current list of spoke pages: [LIST OF SPOKE PAGE URLs AND TITLES] For each spoke page: (1) suggest where on the hub page it should be linked (which section, what context), (2) write the anchor text, (3) write a 1-2 sentence teaser description. Format this as a ready-to-use outline for the hub page author. ```

Structured data (schema markup) helps Google understand page content and can unlock rich results — star ratings, FAQs, how-to steps, article metadata, breadcrumbs. Writing JSON-LD by hand is tedious and error-prone. AI models generate valid JSON-LD quickly, and GPT-5 has strong schema.org knowledge from its training data.

**Prompt 29 — Article schema:** ``` Generate valid JSON-LD schema markup for the following article using the Article schema from schema.org. Use all applicable properties. Article title: [TITLE] Published date: [DATE] (ISO 8601) Modified date: [DATE] Author name: [NAME] Author URL: [URL] Publisher name: [ORGANIZATION] Publisher logo URL: [URL] Page URL: [CANONICAL URL] Description: [META DESCRIPTION] Output only the JSON-LD block, wrapped in <script type="application/ld+json"> tags. No commentary. ```

**Prompt 30 — FAQ schema from a Q&A list:** ``` Generate valid JSON-LD FAQPage schema markup for the following FAQ. Each question and answer must be included. [Q1]: [A1] [Q2]: [A2] [Q3]: [A3] (continue) Output only the JSON-LD block wrapped in <script type="application/ld+json"> tags. Ensure answers are plain text (no HTML inside acceptedAnswer). ```

**Prompt 31 — HowTo schema:** ``` Generate valid JSON-LD HowTo schema for the following process. Include all steps. If I have provided images or tool names, include them. Title: [HOW-TO TITLE] Description: [ONE SENTENCE DESCRIPTION] Total time: [ESTIMATED TIME] Steps: 1. [STEP NAME] — [STEP DESCRIPTION] 2. [STEP NAME] — [STEP DESCRIPTION] (continue) Output only the JSON-LD block wrapped in <script type="application/ld+json"> tags. ```

**Prompt 32 — BreadcrumbList schema:** ``` Generate JSON-LD BreadcrumbList schema for the following page hierarchy. Level 1: [HOME] — [URL] Level 2: [CATEGORY] — [URL] Level 3: [SUBCATEGORY] — [URL] Level 4: [THIS PAGE TITLE] — [URL] Output only the JSON-LD block. No commentary. ```

**Prompt 33 — Schema audit of existing markup:** ``` Review the following JSON-LD schema markup for errors, missing recommended properties, and any issues that might prevent rich results. For each issue: (1) identify the problem, (2) show the corrected version. [PASTE YOUR EXISTING JSON-LD HERE] Reference schema.org definitions and Google's structured data documentation. Be specific about which properties are required vs. recommended by Google. ```

Topic cluster architecture — building a pillar page + network of interlinked spoke pages around a broad topic — is how modern SEO sites build topical authority. Planning the architecture before writing means you avoid duplication, cover the full topic surface, and build internal linking from the start. This is also where best ChatGPT prompts for copywriters 2026 and AI for content marketing workflows intersect.

**Prompt 34 — Full topic cluster map from a seed keyword:** ``` You are a senior content strategist. Build a complete topic cluster architecture for the seed keyword "[SEED KEYWORD]". Deliverable: 1. One pillar page: topic, recommended title, target keyword, target word count 2. 10-15 spoke pages: topic, recommended title, target keyword, word count, intent type, relationship to pillar 3. A mapping of which spokes link to which other spokes (cross-linking plan) 4. 3 keywords that are too broad for spoke pages and should be separate pillars in the future Site niche: [NICHE] Target audience: [AUDIENCE] ```

**Prompt 35 — Gap analysis for an existing cluster:** ``` I have an existing topic cluster around "[PILLAR TOPIC]". Here are the spoke pages I've already published: [LIST OF SPOKE TITLES/KEYWORDS] Identify: (1) Subtopics within "[PILLAR TOPIC]" that my cluster does not yet cover, (2) For each gap, assess whether it warrants a standalone spoke page or a section on an existing page, (3) Rank the gaps by estimated search volume potential (high/medium/low based on topic breadth). ```

**Prompt 36 — Cluster keyword extraction from a broad topic:** ``` Given the broad topic "[TOPIC]", generate a comprehensive list of cluster keywords organized by subtopic. For each keyword: (1) list the keyword, (2) classify it (pillar / spoke / FAQ / tool page), (3) estimate intent (Info / Commercial / Transactional). Aim for 40-60 total keywords. Do not include branded keywords or geographic variants. ```

**Prompt 37 — Content calendar from a topic cluster:** ``` I have a topic cluster with the following pillar and spoke pages: [LIST PILLAR + ALL SPOKES] Create a 12-week content publication calendar. Order the pages so that: (1) The pillar publishes by week 4, (2) Supporting spokes that the pillar links to are published before the pillar, (3) High-authority spoke pages publish first to build internal link equity for the pillar. Output as a week-by-week table. ```

**Prompt 38 — Topical authority gap vs. a competitor:** ``` My site is trying to rank in the niche "[NICHE]". My competitor "[COMPETITOR DOMAIN]" outranks me on multiple cluster keywords. Based on typical topical coverage in this niche, identify: (1) What types of pages a well-rounded authority site in this niche should have, (2) Which of those page types I am probably missing (I'll confirm with my own audit), (3) Which types to prioritize first based on search intent alignment. Do not make up specific URLs or traffic figures. Reason from first principles about what a comprehensive resource in this niche covers. ```

All prompts in this guide work across the major frontier models, but behavior differs enough to be worth noting. GPT-5 follows structural formatting instructions (tables, JSON, numbered lists) with the highest consistency — it is the most reliable for schema markup generation and bulk output tasks where exact format matters. It tends to be verbose on section descriptions; use a character-limit instruction if you need tight output.

Claude Opus 4.8 from Anthropic excels at nuanced analysis tasks — SERP intent classification, content brief rationale, and competitor gap analysis where you want the model to reason through ambiguity rather than just label things. For tasks where you paste in a competitor page and want a strategic read of it, Claude tends to produce more actionable observations. Its JSON output is clean but benefits from an explicit 'no commentary' instruction.

Gemini 2.5 Pro from Google DeepMind has a longer context window that makes it particularly well-suited for large keyword clustering tasks — pasting 500+ keywords at once is more reliable here than with the other models. Its structured outputs for keyword tables are clean. For schema markup, prefer GPT-5 or Claude as Gemini occasionally invents non-existent schema properties. A detailed model comparison across SEO tasks is in Claude vs Gemini for SEO research 2026.

**Prompt 39 — Model selection decision prompt (use this when you're unsure which model to use for an SEO task):** ``` I need to complete the following SEO task: [DESCRIBE YOUR TASK]. My output requirements are: [FORMAT, LENGTH, PRECISION REQUIREMENTS]. I have access to GPT-5, Claude Opus 4.8, and Gemini 2.5 Pro. Which model should I use for this specific task and why? What are the top 2-3 risks with each model for this type of output? ```

Individual prompts are useful. A chained workflow — where the output of one prompt feeds into the next — is where AI SEO work gets genuinely powerful. Here is the sequence that goes from a seed keyword to a ready-to-brief spoke page in under 30 minutes:

**Step 1:** Run Prompt 6 (intent classification) on your keyword list to sort by intent. **Step 2:** Run Prompt 2 (topic cluster map) on the Transactional + Commercial keywords to identify your pillar structure. **Step 3:** Run Prompt 20 (full content brief) for your highest-priority spoke page, using your pillar keyword and the cluster map as context. **Step 4:** Run Prompt 29-33 to generate schema markup for the page type. **Step 5:** Run Prompt 25 (internal link map) once you have 10+ pages live to plan cross-linking.

**Prompt 40 — Full workflow orchestration prompt (GPT-5 or Claude Opus 4.8):** ``` You are an SEO strategist running a complete workflow for a new piece of content. I will give you a target keyword. Complete the following steps in order, clearly labeling each step: 1. Classify the search intent 2. Recommend a content format and word count 3. Generate a draft H1 and 3 title tag variants (under 60 characters each) 4. Generate a meta description (under 155 characters) 5. Outline an H2 structure (6-8 sections with one-sentence descriptions) 6. List 5 questions the content must answer 7. Suggest 3 internal link targets (describe the type of page) and draft anchor text for each 8. Generate FAQ schema for 3 likely FAQs about this topic Target keyword: [YOUR KEYWORD] Site niche: [YOUR NICHE] Target audience: [YOUR AUDIENCE] Do not skip steps. Be specific, not generic. ```

This single prompt is the 80/20 version of this entire guide — it does not replace the specialist prompts above for high-volume batch work, but it is the fastest way to go from keyword to a complete brief + metadata + schema in a single model call. For cost efficiency across all of these tasks, see our AI Prompt Cost Calculator to estimate model costs before running batch jobs.

**Mistake: Not giving the model your keyword list — just a topic.** If you ask 'what keywords should I target for a blog about project management tools', you get a hallucinated list with no real volume data. Always export your keyword list from Ahrefs/Semrush first, then paste it into the prompt. The model's job is to organize and label your data, not to invent data from scratch.

**Mistake: Leaving out output format instructions.** 'Analyze these keywords' produces a wall of text. 'Analyze these keywords and output a table with columns: Keyword | Intent | Cluster | Recommended Slug' produces something you can paste into a spreadsheet. Always specify output format, especially for bulk tasks.

**Mistake: Asking for schema markup without specifying the type.** 'Generate schema for my blog post' can produce Article, BlogPosting, TechArticle, or NewsArticle — they are different schemas with different property sets. Specify exactly which schema type you want and which properties are mandatory. If you're unsure, run Prompt 33 to audit existing markup first.

**Mistake: Using the model's output as final copy without a human read.** AI-generated title tags, meta descriptions, and content briefs are first drafts. They are fast and structurally sound, but they frequently produce generic benefit statements ('learn everything you need to know') that a human editor would cut immediately. Budget 10-15 minutes per batch to review and sharpen the outputs. See how to write better prompts: 15 rules for prompt structures that reduce this generic output problem.