- Ten ChatGPT prompts for SEO that fix the reason most prompt lists disappoint: they ask for opinions instead of handing over evidence. Every prompt here has a slot for your real data and a constraint list that blocks the failure modes.
- ChatGPT has no live search-volume data. These prompts make it analyse what you paste (Search Console rows, competitor headings, crawl exports) and tell you what to validate in a real tool.
- Set your business context once in Custom Instructions, then run the prompts. Prefer Claude, or want these as installable workflows? See Claude Prompts for SEO and the free skills library.
Search "ChatGPT prompts for SEO" and you get fifty-item listicles where every prompt is a variation of "act as an SEO expert and suggest keywords". Run one and you get the same generic output everyone else gets, because the prompt gave ChatGPT nothing to work with. The fix is structural. A prompt that carries your data, your constraints, and a defined output shape returns work you can ship. This set is built on the prompt collection our parent agency published, upgraded with the structure that current prompt-engineering guidance actually recommends.
How do you write a ChatGPT prompt for SEO?
A ChatGPT prompt for SEO works when it has six parts: role, real data, one task, constraints, an output format, and a verify step. Role sets the standard the answer is judged by. Real data is your Search Console export or your page copy pasted in, and it is the difference between analysis and astrology. One task keeps the output focused. Constraints block known failure modes: invented search volumes, filler paragraphs, promises your page cannot keep. Format defines the shape (a table, a list, a JSON block) so you paste the result instead of reformatting it. The verify step tells ChatGPT to check its own work, or to ask you for missing inputs instead of guessing, which is where most hallucinated "data" comes from. Each prompt below is that anatomy applied to one SEO job.
The 10 ChatGPT SEO prompts
1. Cluster the demand you already have
Role: You are a senior SEO analyst who works only from evidence. My data (Google Search Console query export): [paste your Queries.csv rows] Task: Group these queries into topic clusters. For each cluster give total impressions, total clicks, impression-weighted average position, and the highest-leverage query. Constraints: Use only the numbers I pasted. Do not estimate search volume or keyword difficulty; tell me to validate new targets in a keyword tool instead. No cluster smaller than 2 queries. Format: A table (Cluster | Impressions | Clicks | Weighted position | Top query), then 3 bullets on the biggest opportunity. Before answering: check your maths on one cluster and show the check. If my paste looks cut off, ask for the rest first.
2. Map intent to the page type that wins
Role: You are an SEO strategist assigning queries to pages. My queries: [paste 10-30 queries] Task: Classify each query's intent (informational, commercial, transactional, navigational) and the page type that wins it (guide, comparison, product, category, tool, FAQ). Constraints: If a query is ambiguous, give both readings instead of forcing one. No advice paragraphs. Format: Table (Query | Intent | Page type | One-line why). Finish with queries that should SHARE one page rather than each getting their own. Before answering: flag every low-confidence row so I can check the live results myself.
3. Outline against the ranking pages' gaps
Role: You are a content strategist who refuses to produce consensus content. Target query: "[query]" The current top 3 pages' headings: [paste their H2s/H3s] What I have that they do not: [2-5 bullets of your real experience, data, or contrarian take] Task: Build an outline that compresses the consensus and leads with my unique material. Constraints: Every H2 is a question. My unique material goes in the first third. If my bullets are too weak to differentiate, say so bluntly. Format: H1 + H2/H3 skeleton, one line under each H2 saying what the section proves. Tag each section [CONSENSUS] or [NEW].
Why the tags matter: the [NEW] sections are the information gain, the reason your page deserves the ranking.
4. Titles and metas that say the searcher's words back
Role: You are rewriting SERP packaging for click-through rate. The page ranks for "[query]" at position [X]. Current title: [title]. Current meta: [meta]. Task: 5 title options and 3 meta options carrying the query's exact words near the front. Constraints: Titles under 60 characters, metas under 155. No promise the page cannot keep. No title that is just keywords and a brand pipe. Format: Numbered options with character counts and a one-line "why this earns the click". Before answering: flag any option whose promise exceeds the page.
Then confirm the pixel truth in the SERP Snippet Checker: characters approximate, pixels decide.
5. Strip the AI tells from a draft
Role: You are a ruthless line editor.
My draft:
[paste it]
Task: Edit so a practitioner believes a human wrote it, preserving every factual claim.
Constraints: Delete sentences that say nothing. Replace hedges with plain claims or cut them. Remove formulaic transitions ("In today's fast-paced world", "It's worth noting"). Add no new facts. Keep my numbers exactly.
Format: The edited draft, then the list of patterns you removed so I stop producing them.
Before finishing: reread and cut a further 10%.6. FAQ plus schema, from visible content only
Role: You are a structured-data specialist who marks up only what exists. My page copy: [paste the visible content] Task: Write a 4-6 question FAQ a reader of this page still needs answered, then the matching FAQPage JSON-LD. Constraints: Every answer must be supported by the pasted content; mark anything unsupported [NEEDS SOURCE] instead of inventing it. The schema text mirrors the visible answers word for word. Format: FAQ as headings + paragraphs, then one JSON-LD code block. Before answering: confirm the JSON is valid and each schema answer matches its visible twin.
Validate the block in the free Schema Markup Validator; new to structured data, start at What Is Schema Markup.
7. Internal links a reader would actually follow
Role: You are an internal-linking strategist. My pages (URL + title, one per line): [paste] My new page: [URL + 2-line summary] Task: 3-5 existing pages that should link TO the new page, with the exact sentence and anchor for each, plus 2-3 outbound links the new page should carry. Constraints: Anchor text matches the target page's title language. Only links a reader would follow. No footer/sidebar placements. Format: Table (From page | Sentence with anchor | Why). Before answering: check no anchor competes with the source page's own target query.
8. Triage a crawl export by impact, not count
Role: You are a technical SEO lead defending priorities to a client. My crawl export: [paste rows: URL, status, title, meta, canonical, indexability] Task: Group issues by type, then rank groups by likely traffic impact. Constraints: Only what is in the data. A 404 nobody links to is not priority one. Separate mechanical fixes from judgement calls. Format: Prioritised list: issue, count + 3 example URLs, impact reasoning, the fix. Before answering: name the columns missing from my paste that would change the ranking.
9. Turn striking distance into packaging fixes
Role: You are hunting the cheapest wins on the board. My queries ranking 4-15 (query, page, impressions, position): [paste] Task: Choose the 5 highest-leverage rows and prescribe each one's packaging fix: the title change, the heading to add, or the paragraph that answers the query directly. Constraints: Leverage = impressions x closeness to page one. Prescriptions must be paste-ready, never "improve the content". Format: For each: query, page, one-line diagnosis, the fix ready to paste. Before answering: where two queries compete on one page, pick the winner and say why.
10. PR angles from something true
Role: You are a digital PR strategist who only pitches real stories. What I have: [your data, survey results, odd observations from your work, or a defensible contrarian position] Task: 5 story angles a journalist would open, each anchored to my material. Constraints: No fake surveys, no invented statistics, no "study reveals" without a study. If my material is too thin, say so and specify the data that would fix it. Format: For each: the headline a journalist would write, the hook sentence, the outlet type, the asset it uses. Before answering: mark any angle that needs data collection before it can be pitched honestly.
The campaign mechanics live in Digital PR for Link Building.
How do you set ChatGPT up for SEO work?
Put your context in Custom Instructions once, so every prompt starts warm. In ChatGPT's settings, describe your site, audience, services, and tone, and add the standing rule that it never invents statistics or search volumes. That removes the boilerplate from every prompt above. Second habit: keep one conversation per project rather than one giant thread, because quality degrades as unrelated context piles up. Third: when the output matters, ask "what would a sceptical editor challenge in your answer?" before you accept it. And know the hard limit: ChatGPT has no live keyword-volume data, so every new-keyword suggestion is a hypothesis to validate in a real tool, not a fact.
What comes after prompts?
After prompts comes packaging the prompt so you never paste it again. If you run these weekly, the paste-edit-repeat loop becomes the bottleneck. That is what our free Claude SEO skills solve: each is a .md file that installs the whole workflow (intake, constraints, output) into Claude, and most run in any LLM as a system prompt. The Claude-tuned versions of this page's prompts, with labelled data tags and Project setup, live in Claude Prompts for SEO. And the four single-job prompts this library started with (site scorecard, information gain, title tags, brand consistency) are on the AI SEO Prompts hub.
FAQ
What is the best ChatGPT prompt for SEO?
The best ChatGPT prompt for SEO is the one that carries your real data. A prompt with your Search Console rows pasted in and explicit constraints returns analysis you can act on; a prompt that asks ChatGPT to "suggest keywords" returns the same guesses it gives everyone. Start with the demand-clustering prompt at the top of this page.
Can ChatGPT do keyword research?
ChatGPT can cluster, classify, and prioritise keywords, but it has no live search-volume data and its volume estimates are inventions. Use it to analyse your own Search Console export, then validate any new targets in a keyword tool before you build pages for them.
Are ChatGPT prompts better than Claude prompts for SEO?
The anatomy is identical; the engines differ. Claude holds more pasted context and follows labelled structure and constraints very reliably, which suits big exports and strict formats. ChatGPT is the more common starting point and works well with the markdown-structured prompts on this page. Run whichever you have; the constraints matter more than the engine.
Is it safe to publish content straight from ChatGPT?
Not without an editing pass. Verify every factual claim, strip the AI patterns (prompt 5 exists for exactly this), and add the first-hand material only you have. Unedited output is consensus content, and consensus content has no reason to outrank its sources.
Sources: StudioHawk's ChatGPT Prompts for SEO (the collection this set upgrades), Anthropic's prompt engineering documentation.