AI Prompts for SEO Specialists (2026)

Replace every [BRACKETED] placeholder with your real data before sending — your actual keyword export, your actual top-ranking SERP notes, your actual page copy. An AI that clusters your real keyword list is doing useful work; one inventing keywords from a one-line topic is wasting your time and may hallucinate search volumes it cannot know.

Demand structured output. SEO work is tabular by nature, so ask for tables, defined columns, and consistent fields you can paste straight into a sheet or CMS. For repeatable briefs, lock the format into a reusable system prompt — see how to write a system prompt and our structured output schema design patterns.

Critically: a language model does not have live search-volume, ranking, or difficulty data. Never let it state a search volume, keyword difficulty, or 'this ranks #3' figure as fact — pull those from your real SEO platform (Search Console, Ahrefs, Semrush) and paste them in. Use the model for synthesis, structure, and language, not for metrics it can't observe.

**1. SERP-informed content brief** — "You are an SEO content strategist. Build a content brief for the target keyword [KEYWORD] with search intent [informational / commercial / transactional]. Using the competing-page notes I paste below, produce: the recommended H1, a suggested URL slug, a logical H2/H3 outline that covers the subtopics competitors cover plus gaps they miss, the questions to answer (from People Also Ask: [PASTE]), suggested internal links from this list: [PASTE], and the primary + 5-8 secondary keywords to include naturally. Do not invent search volumes or difficulty scores. Competing pages: [PASTE NOTES]."

**2. Outline that matches intent** — "Draft an article outline for [KEYWORD]. First state the dominant search intent in one line and the format the SERP rewards (listicle / how-to / comparison / definition). Then give an H2/H3 outline where each H2 maps to a sub-intent, with a one-line note on what each section must answer. Flag any section that would be thin and should be merged. Intent evidence: [PASTE SERP NOTES]." Turn this into a full outline with the Blog Post Outline Generator.

**3. Brief for a writer (handoff-ready)** — "Convert the notes below into a writer-ready brief: target keyword, intent, audience, word-count range, mandatory sections, must-include entities/terms, tone, 3 credible source types to cite, and 5 questions the piece must answer. Add a 'do not' list (no fabricated stats, no unsupported claims). Output as a clean brief a freelancer could execute without follow-up. Notes: [PASTE]."

**4. Cluster a keyword list by intent and topic** — "Cluster the keyword list below into topic groups, each representing one potential page. For each cluster give: a cluster name, the primary keyword, the supporting keywords, the search intent, and the suggested page type (pillar / cluster / product / FAQ). Keep clusters mutually exclusive. Do not estimate search volume or difficulty — only group what I paste. Keywords: [PASTE LIST]."

**5. Pillar-and-cluster map** — "From the clusters below, design a pillar-and-cluster (topic hub) architecture: name the pillar page, list the supporting cluster pages, and specify the internal-link direction (clusters link up to pillar, pillar links down to clusters). Flag any cluster that overlaps another (cannibalization risk) and recommend a merge or differentiation. Clusters: [PASTE]."

**6. Identify cannibalization risk** — "Review the page-and-keyword list below for keyword cannibalization — multiple pages targeting the same intent. Group the at-risk pages, explain the overlap, and recommend one action per group: consolidate, differentiate, or canonicalize. Base this only on the data I provide; do not assume rankings I haven't given you. Data: [PASTE PAGES + TARGET KEYWORDS]."

**7. Title tags and meta descriptions** — "Write 5 title tags and 5 meta descriptions for a page about [TOPIC] targeting [KEYWORD] with [intent]. Rules: title under 60 characters with the keyword near the front and one clear benefit; meta description 150-160 characters, keyword used naturally, ending with a reason to click — no clickbait the page can't deliver. Each option should test a slightly different angle. Return a table: Title | char count | Meta | char count | angle." Generate these directly in the SEO Meta Generator.

**8. FAQ schema content** — "Write 6 FAQ question-and-answer pairs for a page about [TOPIC]. Phrase each question exactly as a user would type or speak it into a search engine, and keep each answer to 2-3 concise, self-contained sentences suitable for a featured snippet and FAQPage schema. Use only accurate, verifiable claims; do not invent statistics. Topic context: [PASTE]."

**9. Snippet-targeted answer block** — "Write a featured-snippet-optimized answer for the query '[QUERY]'. Lead with a direct 1-2 sentence answer (40-55 words), then a short supporting paragraph, then a 3-5 item list or table if the query implies one. Match the snippet format Google currently shows for this query: [paragraph / list / table]. No fluff before the answer."

**10. Internal linking recommendations** — "Given the list of existing pages with their topics and target keywords below, recommend internal links for a new page about [TOPIC]. For each suggestion give: source page, anchor text (descriptive, keyword-relevant, not 'click here'), and a one-line reason it's contextually relevant. Suggest links in both directions where it makes sense. Pages: [PASTE LIST]."

**Bonus — on-page audit checklist** — "Audit the page content below against on-page SEO best practices: title and H1 alignment with intent, heading hierarchy, keyword usage (natural, not stuffed), internal/external links, image alt text presence, and content depth versus the query. Return Issue | Severity | Fix. Note where you'd need live tools (Core Web Vitals, index status) that you can't check from text. Page: [PASTE]."

Map the whole publishing pipeline with the Content Calendar Generator, and for the deeper research workflow see Claude vs Gemini for SEO research and our 10 Claude prompts for SEO blog production.

The single biggest SEO-AI mistake is trusting a language model for metrics it cannot observe. It has no live access to search volume, keyword difficulty, current rankings, or index status, and it will invent confident-looking numbers if you ask. Pull every metric from your real platform and paste it in; use the model only for clustering, structure, briefs, and language. Anything it states as a number must come from a source you provided.

Don't publish AI metadata or briefs unedited, and don't keyword-stuff because the model offered to. Ranking depends on the page content and intent match, not the metadata alone — generate angle variations within the character limits Google renders, then test in Search Console. And don't paste confidential client strategy or unpublished data into a public chatbot without checking the data policy. For the security angle of pasting external SERP content into prompts, see the prompt injection defense checklist.

For large keyword lists and multi-page clustering, a long-context frontier model handles the volume best — Claude Opus 4.8, GPT-5.5, or Gemini 3.5 Pro. For high-volume metadata generation where cost matters, an efficiency-tier model like Claude Haiku 4.5 or Gemini 3.5 Flash is plenty. If you need the model to actually fetch and cite live pages, a search-grounded answer engine like Perplexity or Gemini with grounding fits better than a plain chat model — see our how to choose an AI model guide. The table compares durable dimensions, not prices.