- Ten copy-paste Claude prompts for SEO, each built on the six-part structure that separates a working prompt from a wish: role, your real data, one task, constraints, output format, and a verify step.
- Claude's edge for SEO is context size and structure: paste whole Search Console exports or full pages, wrap them in labelled tags, and make Claude check its own output.
- When you find yourself reusing a prompt weekly, stop pasting and install it: every prompt here has a free downloadable skill that runs it automatically.
Most Claude SEO prompts fail the same way: they ask for an opinion when they could hand over evidence. "Suggest keywords for my plumbing site" gets you generic guesses. Paste your actual Search Console export and the same request returns analysis of demand you already have. The ten prompts below are built for that second mode, and every one follows the same six-part anatomy, so you can repair or extend them yourself.
What makes a Claude SEO prompt actually work?
A working Claude SEO prompt has six parts: a role, your real data, one task, explicit constraints, an output format, and a verify step. The role sets standards ("a technical SEO lead who has to defend this to a client"). The data is the biggest upgrade most people skip: Claude's context window comfortably holds full GSC exports, crawl files, or entire pages, and pasted evidence beats invented examples every time. One task per prompt keeps the output usable. Constraints state what Claude must not do (invent search volumes, pad with fluff, guess numbers). Format defines the exact shape you want back. And the verify step makes Claude audit its own answer or ask for missing inputs instead of hallucinating them. Claude responds especially well to labelled sections, so the prompts below wrap data in simple tags like <gsc_data>, which is how the anatomy becomes a habit.
The 10 Claude SEO prompts
1. Keyword landscape from your own Search Console data
You are a senior SEO analyst. I am pasting my Google Search Console query export below. Analyse the demand I already have instead of inventing keywords. <gsc_data> [paste your Queries.csv rows here] </gsc_data> Task: group these queries into topic clusters, and for each cluster report: total impressions, total clicks, impression-weighted average position, and the single highest-leverage query. Constraints: use only the numbers in the data. Do not estimate search volumes or difficulty; if I need those, tell me to validate in a keyword tool. Do not create a cluster with fewer than 2 queries. Format: one table (Cluster | Impressions | Clicks | Weighted position | Top query), then a 3-bullet read of the biggest opportunity. Verify: before answering, check the maths on one cluster and show the check. If the paste looks truncated, ask me for the rest instead of proceeding.
The upgrade over "suggest keywords": everything is measured. When this becomes weekly work, the GSC Quick Start skill runs the whole read automatically.
2. Search intent mapping that assigns page types
You are an SEO strategist mapping queries to pages. Here are my target queries: <queries> [paste 10-30 queries] </queries> Task: classify each query's intent (informational, commercial, transactional, navigational) and assign the page type that wins it (guide, comparison, product, category, tool, FAQ). Constraints: if a query is ambiguous, say so and give both readings rather than forcing one. No generic advice paragraphs. Format: table (Query | Intent | Page type | One-line rationale). End with any queries that should share one page instead of getting separate pages. Verify: flag every query where you are less than confident, so I can check the live results myself.
The last line matters: merged queries prevent you building three thin pages where one strong page wins. The Search Intent Mapper skill is this prompt, installable.
3. Information-gain outline from competitor gaps
You are a content strategist who refuses to write consensus content. I want to outrank the current results for "[target query]". <competitor_headings> [paste the H2/H3 headings from the top 3 ranking pages] </competitor_headings> <what_i_know> [2-5 bullets: your real experience, data, or contrarian view on this topic] </what_i_know> Task: build an outline that covers what ranking pages agree on in less space, and leads with what none of them have, drawn from my inputs. Constraints: every H2 phrased as a question. The unique material goes in the first third, never the end. If my "what I know" bullets are too thin to create real difference, say so plainly. Format: H1 + H2/H3 skeleton, with one line under each H2 stating what the section proves. Verify: mark each section [CONSENSUS] or [NEW] so I can see the gain at a glance.
This is the information gain play as a prompt: the [NEW] sections are why the page deserves to rank.
4. Title and meta rewrites that carry the query's grammar
You are rewriting SERP packaging. My page ranks for "[query]" at position [X] with a low click-through rate. <current> Title: [current title] Meta: [current meta description] </current> Task: write 5 title options and 3 meta options that carry the query's exact words near the front. Constraints: titles under 60 characters, metas under 155. No clickbait the page cannot keep. No pipes-and-brand-only titles. Format: numbered options, each with a one-line "why this earns the click". Verify: for each title, state the character count. Flag any option where the promise exceeds what my page delivers.
Check pixel widths (characters lie) in the free SERP Snippet Checker before shipping.
5. The de-AI editing pass
You are a ruthless line editor. Below is a draft that reads machine-written.
<draft>
[paste the draft]
</draft>
Task: edit it so a practitioner would believe a human wrote it, keeping every factual claim intact.
Constraints: delete every sentence that says nothing. Replace hedges ("can potentially help") with plain claims or cut them. Kill formulaic transitions ("In today's digital landscape", "It's important to note"). Do not add new facts. Keep my numbers exactly as written.
Format: the edited draft, then a short list of the patterns you removed so I stop writing them.
Verify: reread your edit and cut 10% more.6. FAQ and FAQPage schema from visible content only
You are a structured-data specialist. Here is my page content: <page> [paste the visible page copy] </page> Task: write a 4-6 question FAQ answering what a reader of this page still asks, then produce the matching FAQPage JSON-LD. Constraints: every answer must be supported by the page content or the inputs I gave you; mark anything unsupported as [NEEDS SOURCE] rather than inventing it. Schema text must mirror the visible answers exactly. Format: the FAQ as headings and paragraphs, then one JSON-LD block. Verify: confirm the JSON parses and every schema answer matches its visible twin.
Run the output through the Schema Markup Validator, and read What Is Schema Markup if this layer is new.
7. Internal link opportunities from your page list
You are an internal-linking strategist. Here are my site's pages and the new page I just published: <pages> [paste URLs + titles, one per line] </pages> <new_page> [URL + 2-line summary of the new page] </new_page> Task: recommend 3-5 existing pages that should link TO the new page, with the exact sentence and anchor text for each placement, plus 2-3 pages the new page should link out to. Constraints: anchors match the target page's title language, not "click here". Only suggest links a reader would genuinely follow. No footer or sidebar suggestions. Format: table (From page | Suggested sentence with anchor | Why). Verify: check no suggested anchor competes with the target page's own primary query on the source page.
A new page with no inbound links is invisible; the Internal Link Checker finds the orphans this prompt should fix first.
8. Technical audit triage from a crawl export
You are a technical SEO lead triaging a crawl. Here is my crawl export: <crawl> [paste rows: URL, status, title, meta, canonical, indexability] </crawl> Task: group every issue by type, then rank the groups by likely traffic impact, not by count. Constraints: use only what is in the data. A 404 on a page nobody links to is not priority one. Say which issues need a human decision versus a mechanical fix. Format: prioritised list, each with: issue, affected URLs (count + 3 examples), impact reasoning, the fix. Verify: state which columns were missing from my paste that would change your ranking, so I can re-export properly.
The installable version is the Technical SEO Audit skill, and the Screaming Frog Analyser for full crawl files.
9. GSC quick wins: packaging fixes for striking distance
You are hunting quick wins. Here are my queries ranking positions 4-15: <striking_distance> [paste query, page, impressions, position rows] </striking_distance> Task: pick the 5 highest-leverage rows and prescribe the packaging fix for each: the title change, the heading to add, or the paragraph to answer the query directly. Constraints: leverage = impressions x closeness to page one, not raw position. Prescriptions must be specific enough to paste, not "improve the content". Format: for each of the 5: the query, the page, the diagnosis in one line, then the paste-ready fix. Verify: if two queries on the same page compete, say which one the page should commit to and why.
10. Digital PR angles from something true
You are a digital PR strategist who only pitches real stories. Here is what I have: <assets> [your data, survey results, unusual observations from your work, or a contrarian position you can defend] </assets> Task: generate 5 story angles a journalist would open, each anchored to my real material. Constraints: no fake surveys, no invented statistics, no "study reveals" framing unless the study exists. If my assets are too thin for a story, say so and tell me what data would make one. Format: for each angle: the headline a journalist would write, the hook sentence, the outlet type it fits, and which asset it uses. Verify: mark any angle that needs additional data collection before it can be pitched honestly.
Then the pitch mechanics live in the Digital PR for Link Building guide and the Data PR & Outreach skill.
How do you set Claude up so prompts work better?
The setup that improves every prompt on this page is a Claude Project with your context saved once. Create a Project, and in its instructions paste the things you currently retype: your site, your audience, your services, your tone rules, and the constraint that Claude never invents statistics. Every conversation in that Project inherits it, so the prompts above shrink to their task and data. Two more Claude-specific habits: wrap every paste in labelled tags the way these prompts do (Claude tracks long context better with labelled sections), and when output quality matters, ask Claude to critique its own draft before finalising. When a prompt becomes a routine, that is the moment to switch from pasting to installing a skill.
Claude prompts vs Claude skills: which do you need?
Prompts are for trying; skills are for repeating. A prompt is pasted per conversation and forgotten. A skill is the same intelligence packaged as a .md file that installs once and triggers automatically, with intake questions, edge-case handling, and consistent output. Everything on this page exists as a deeper skill in the free Claude SEO skills library (46 at last count), and the honest guidance is: run the prompt twice, and if you reach for it a third time, install the skill. These prompts also work in ChatGPT and Gemini with the tags swapped for headings; the ChatGPT-tuned set lives in ChatGPT Prompts for SEO.
FAQ
What are the best Claude prompts for SEO?
The best Claude prompts for SEO are the ones that hand Claude your real data: your Search Console export, your crawl file, your page copy. The ten on this page cover keyword clustering, intent mapping, outlines, titles, editing, schema, internal links, technical triage, quick wins, and digital PR, and each includes constraints and a verify step so the output is usable, not generic.
Can Claude do keyword research?
Claude can cluster, classify, and prioritise keywords brilliantly, but it has no live search-volume data. Use it to analyse the demand in your own Search Console export, then validate volumes for new targets in a keyword tool. Never publish a search volume Claude produced.
Do these prompts work in ChatGPT or Gemini?
Yes. Swap the XML-style tags for bold headings and every prompt runs in ChatGPT or Gemini. Claude's advantages are context size for big pastes and how reliably it follows labelled structure and constraints.
What is the difference between a Claude prompt and a Claude skill?
A prompt is text you paste into one conversation. A skill is a .md file you install once; it triggers automatically, asks for the inputs it needs, and produces consistent output every run. Prompts are for trying, skills are for repeating.
Sources: StudioHawk's ChatGPT Prompts for SEO (the basis this set upgrades), Anthropic's prompt engineering documentation.