Best AI prompts for SEO, organized by task
Summary
- The best AI prompts for SEO name one job, the inputs that job requires, the limits the model must respect, and the format of the output.
- This page gives you copy-ready prompts for keyword clustering, search intent analysis, content briefs, outlines, titles and metas, internal linking, content refreshes, and source-backed answer blocks.
- Each prompt tells the model not to invent search volume, live rankings, URLs, or sources you did not provide.
- A short quality check after every template shows what to review before you use the result.
- ChatGPT, Claude, and Perplexity fit different jobs: drafting, long-document review, and source-backed research.
- Related templates live in the AirOps prompts library. A companion guide, linked after the templates, covers the method behind them.
The best AI prompts for SEO do a specific job, ask for the inputs that job requires, and tell the model how to format and check the result. They also tell the model to use only the search volume, citations, and rankings you supply.
This page collects copy-ready prompts for eight search and answer engine optimization (AEO) tasks. AEO is the work of making content easy for AI search systems to extract, cite, and trust. Each section includes the task, the inputs to provide, a complete prompt, a picture of a strong output, and a quality check.
AirOps works with marketing teams to turn prompt work like this into repeatable workflows across the channels that shape AI discovery. Use the templates on their own, or build them into a process you can run again with the same inputs and the same review steps.
How we picked these AI prompts for SEO
These templates were written to a fixed standard. This page makes no benchmark claims about them.
A prompt earned a place here when it includes all five of the following:
| Criterion | What it means in practice |
|---|---|
| Clear job | The prompt states one task and the decision the output should support. |
| Required inputs | You supply the export, URL list, draft, or sources the model needs. |
| Constraints | The prompt forbids invented metrics, URLs, and unsourced claims. |
| Structured output | The prompt asks for a table, outline, or other format you can review. |
| Built-in QA | The prompt asks the model to flag gaps, ambiguities, and missing inputs. |
Replace every placeholder in square brackets before you run a prompt. If an input is missing, say so in the prompt rather than asking the model to fill the gap.
Prompts for keyword clustering
Task: Group a keyword export by shared search intent so you can decide which queries belong on the same page, and which need their own URL.
Required inputs:
- A keyword export with the fields you actually have (query, and any volume, difficulty, or current-URL columns)
- Your market and language
- An optional list of URLs you already rank or plan to keep
For more clustering templates, see the topic cluster prompts for Claude. For the strategy behind the groupings, see the AI content strategy playbook.
Copy-ready prompt:
You are an SEO strategist grouping keywords by shared search intent.
Inputs:
- Keyword export: [PASTE KEYWORD EXPORT]
- Existing or planned URLs (optional): [PASTE URL LIST]
- Market and language: [e.g., United States, English]
Task:
Using the supplied keyword export, group keywords by shared search intent. Return a table with cluster, representative query, supporting queries, recommended page type, and rationale. Flag ambiguous groups for review. Do not invent search volume or infer live search results from keywords alone.
Output:
Return a markdown table with these columns:
- Cluster name
- Representative query
- Supporting queries
- Recommended page type (hub, supporting article, comparison, glossary, or other)
- Rationale
- Ambiguity flag (Yes or No)
- Review note
Constraints:
- Use only the fields present in the export.
- If search volume, difficulty, or SERP features are not in the export, write "not provided" instead of guessing.
- Place every keyword in exactly one cluster or in an "Unassigned / review" group.
- Do not force a cluster when two intents are equally plausible.
Quality check before you finish:
- Confirm that no metric in the table was invented.
- List any keywords you could not place with confidence.
- Call out clusters that might support more than one page type.Example output: A table with one row per cluster. Each row names a representative query, lists supporting queries from the export, recommends a page type, and explains the shared intent in one or two sentences. Ambiguous groups appear with the flag set to Yes, and leftover queries sit in an Unassigned group.
Quality check:
- Every query from your export appears once.
- No volume, difficulty, or ranking number appears unless it was in the file you pasted.
- Ambiguous groups are flagged in their own row.
- Recommended page types match how your site is actually organized.
Prompts for search intent analysis
Task: Classify queries by the job the searcher wants done, so you can match page type and content format to intent.
Required inputs:
- The query list or cluster you want classified
- Any notes you already have about audience or funnel stage
- The page types you are willing to create or refresh
Pair this with long-tail keyword prompts for Perplexity or ChatGPT prompts for enterprise SEO content teams when you still need a cleaner query list. For definitions, see the search intent glossary entry.
Copy-ready prompt:
You are an SEO analyst classifying search intent from supplied queries.
Inputs:
- Queries or clusters: [PASTE QUERIES]
- Audience or funnel notes (optional): [PASTE NOTES]
- Allowed page types: [e.g., guide, comparison, template, glossary, product]
Task:
For each query, identify the likely search intent using only the wording of the query and the notes provided. Do not claim to know the current SERP, AI answer, or ranking page unless that evidence is pasted below.
Output:
Return a markdown table with these columns:
- Query
- Primary intent (informational, navigational, commercial investigation, transactional, or mixed)
- Searcher job to be done
- Recommended content format
- Confidence (high, medium, low)
- What extra evidence would raise confidence
Constraints:
- If a query could support more than one intent, mark it mixed and explain both readings.
- Do not invent click data, SERP features, or competitor URLs.
- Do not infer that a query is "high volume" or "low competition" unless those fields are in the input.
Quality check before you finish:
- Count how many queries you marked mixed or low confidence.
- List any queries that need a live SERP or AI-answer review before a page decision.Example output: A table that labels each query, states the searcher's job in plain language, and recommends a format such as a how-to guide, comparison, or glossary entry. Mixed or low-confidence rows include a note about the extra evidence you still need, such as a live results review.
Quality check:
- Intent labels follow from the query wording.
- Mixed intents stay mixed until you inspect live results yourself.
- No competitor URLs, featured-snippet claims, or volume figures appear unless you supplied them.
Prompts for content briefs
Task: Turn a target query, audience notes, and source material into a brief a writer can draft from without guessing facts, sources, or on-page requirements.
Required inputs:
- Primary query and any supporting queries
- Audience, product, and point-of-view notes
- Sources the writer is allowed to use
- Examples of on-brand pages, if you have them
Use the content brief prompts for ChatGPT when you want more brief variations. For SEO content-writing prompts that sit downstream of the brief, see content rewriting prompts for Perplexity.
Copy-ready prompt:
You are a content strategist writing a brief for one page.
Inputs:
- Primary query: [PRIMARY QUERY]
- Supporting queries: [PASTE SUPPORTING QUERIES]
- Audience and product notes: [PASTE NOTES]
- Approved sources: [PASTE URLS OR EXCERPTS]
- On-brand example pages (optional): [PASTE URLS OR EXCERPTS]
- Business goal for the page: [e.g., rank for the query, earn AI citations, support a comparison journey]
Task:
Write a content brief the writer can follow without inventing facts or sources. The brief should cover search intent, the page's job, outline direction, evidence requirements, and on-page elements.
Output:
Use these headings:
1. Page job and search intent
2. Reader and promised outcome
3. Must-answer questions
4. Evidence and sources the writer may use
5. Outline direction (H2 and H3 suggestions only)
6. On-page elements (title direction, meta direction, internal link opportunities if URLs are supplied)
7. Open questions and missing inputs
Constraints:
- Use only the approved sources for facts, statistics, and product claims.
- If a claim would need a source you were not given, add it to "Open questions" instead of drafting the claim.
- Do not invent customer results, quotes, or statistics.
- Do not write the full article.
Quality check before you finish:
- Confirm every factual claim in the brief traces to an approved source or is marked as a gap.
- List any must-answer questions that the supplied sources cannot support yet.Example output: A brief with a stated page job, a list of questions the article must answer, a source list the writer is allowed to use, outline direction (not a finished draft), and an open-questions section for anything the inputs cannot support.
Quality check:
- The brief does not smuggle in unsourced statistics or customer results.
- Must-answer questions map to the primary and supporting queries.
- Outline direction is specific enough to draft from, and short enough that it does not become the article.
Prompts for outlines
Task: Build a page outline that matches intent, leads sections with answers, and leaves room for evidence you already have.
Required inputs:
- The brief, or the primary query plus audience notes
- Must-answer questions
- Sources or examples the outline should make space for
- Any sections you already know you need (FAQ, comparison, process steps)
Start from the blog post outline prompts for ChatGPT if you want additional outline formats.
Copy-ready prompt:
You are an SEO editor creating an outline for one page.
Inputs:
- Brief or page job: [PASTE BRIEF]
- Primary query: [PRIMARY QUERY]
- Must-answer questions: [PASTE QUESTIONS]
- Approved sources or examples: [PASTE SOURCES]
- Required sections (optional): [e.g., FAQ, step-by-step, comparison table]
Task:
Create an outline that a writer can draft from. Lead each proposed H2 with the answer that section should state first. Map each must-answer question to a specific heading.
Output:
Return:
- Working title options (3)
- H1 recommendation
- An outline with H2 and H3 headings
- For each H2: the question it answers, the answer-first sentence to open with, the evidence to include, and estimated depth (short, medium, or long)
- Notes on tables, lists, or FAQ questions to include
- Gaps that the supplied sources cannot fill
Constraints:
- Do not write full paragraphs of article copy.
- Do not add sections that require facts you were not given.
- Do not invent statistics, quotes, or case studies to justify a section.
- Keep headings in sentence case unless the brief specifies another style.
Quality check before you finish:
- Confirm every must-answer question appears under a heading.
- Flag any heading that cannot be supported by the approved sources.Example output: Three title options, a recommended H1, and a heading tree. Each H2 includes the question it answers, a suggested opening sentence, and a note about the evidence that belongs there. Unsupported ideas appear as labeled gaps.
Quality check:
- Headings mirror real questions.
- Every must-answer question has an owner heading.
- The outline does not assume sources, screenshots, or data you still need to gather.
Prompts for title and meta options
Task: Produce title and meta description options that name the outcome, include the primary query naturally, and stay inside length limits.
Required inputs:
- Primary query
- Page job and promised outcome
- Any words you must include or avoid
- Character limits you want to respect
For more meta-specific templates, see the ChatGPT meta description prompts.
Copy-ready prompt:
You are an SEO editor writing title and meta options for one page.
Inputs:
- Primary query: [PRIMARY QUERY]
- Page job and promised outcome: [PASTE]
- Words to include: [LIST]
- Words to avoid: [LIST]
- Title character limit: [e.g., 50 to 60]
- Meta description character limit: [e.g., 140 to 160]
Task:
Write title tag and meta description options that state what the reader gets. Include the primary query once in the title and once in the meta description when it fits naturally.
Output:
Return a table with these columns:
- Option
- Title
- Title character count
- Meta description
- Meta character count
- Notes (intent match, differentiation, risk)
Provide 5 options. Then recommend one title and one meta description, with a one-sentence reason.
Constraints:
- Do not use clickbait, fake urgency, or invented results (no "tested," "proven," or made-up percentages).
- Do not start a meta description with "Learn," "Discover," or "In this article."
- Do not repeat the exact same wording across options.
- If a required word will not fit the character limit, say so instead of overflowing.
Quality check before you finish:
- Recount every title and meta description.
- Confirm the primary query appears naturally, not as stuffed repetition.Example output: A five-row table of titles and metas with character counts and a short note on each option. A final recommendation names one title and one meta, and explains why that pair states the page's outcome most clearly.
Quality check:
- Character counts are accurate when you paste the text into your own counter.
- The primary query appears once in the recommended title and once in the recommended meta, without awkward repetition.
- No option promises a result you cannot defend.
Prompts for internal linking
Task: Suggest internal links from a draft to real URLs you already have, using descriptive anchor text and a clear reason for each link.
Required inputs:
- The draft, outline, or section that needs links
- A URL inventory (title, URL, and a one-line description if you have it)
- Any linking rules (maximum links, words you do not use as anchors, pages to avoid)
Use the internal linking prompts for Claude when you want more linking patterns.
Copy-ready prompt:
You are an SEO editor recommending internal links from a draft to an approved URL inventory.
Inputs:
- Draft, outline, or section: [PASTE CONTENT]
- URL inventory: [PASTE TITLE, URL, AND SHORT DESCRIPTION FOR EACH PAGE]
- Linking rules: [e.g., maximum 8 links, no "click here," do not link the same URL twice]
Task:
Recommend internal links that help the reader and clarify topical relationships. Use only URLs from the inventory.
Output:
Return a table with these columns:
- Destination URL
- Destination title
- Suggested anchor text
- Sentence or section where the link belongs
- Reason for the link
- Priority (high, medium, low)
Then list inventory URLs you considered and rejected, with a one-line reason.
Constraints:
- Never invent a URL, title, or redirect.
- If no inventory URL fits a passage, say "no match" instead of guessing.
- Write descriptive anchor text that matches the destination topic. Do not use "click here," "read more," or a bare URL as the anchor.
- Do not recommend linking the same destination twice unless the rules allow it.
Quality check before you finish:
- Confirm every recommended URL appears in the inventory exactly as written.
- Confirm each anchor would make sense if read without the surrounding sentence.Example output: A table of recommended links, each tied to a sentence or section in the draft, plus a short rejection list that shows why other inventory URLs were skipped. Every destination URL matches the inventory character for character.
Quality check:
- You can open every recommended URL and recognize the page from the suggested anchor.
- No URL was invented or "helpfully" completed.
- Anchor text describes the destination, and the same URL is not repeated.
Prompts for content refreshes
Task: Decide what to update on an existing page, based on the current copy and the evidence you provide, without rewriting the whole article by default.
Required inputs:
- The current page URL and full copy (or a complete outline plus key passages)
- What you know about performance or freshness (optional, and only if it is real)
- New sources, product changes, or questions the page should now answer
- Sections that must stay the same
Use the content rewriting prompts for Perplexity for more refresh patterns, and the SEO audit report prompts for Perplexity when you need a wider page review first. For broader refresh planning, see the content refresh strategy guide.
Copy-ready prompt:
You are an SEO editor planning a content refresh for one existing page.
Inputs:
- URL: [PAGE URL]
- Current copy: [PASTE FULL COPY]
- Known performance or freshness notes (optional): [PASTE ONLY VERIFIED NOTES]
- New sources or product changes: [PASTE]
- Questions the page should now answer: [PASTE]
- Sections that must stay: [LIST]
Task:
Propose a refresh plan that updates the page without discarding what still works. Recommend keep, revise, replace, add, or remove for each major section.
Output:
Return:
1. A section-by-section table with columns: Section, Action (keep, revise, replace, add, remove), Reason, Evidence needed, Priority
2. A list of claims that look outdated or unsourced
3. New questions the page still cannot answer with the supplied sources
4. A recommended refresh scope (light edit, substantial update, or rewrite) and why
Constraints:
- Do not invent traffic, ranking, citation, or conversion changes.
- If performance notes were not provided, do not infer that the page is winning or losing.
- Do not invent new statistics to "freshen" the page.
- Do not rewrite the full article unless the supplied evidence shows the current structure cannot do the job.
Quality check before you finish:
- Separate observed issues in the copy from issues that would require data you were not given.
- List every claim you flagged as unsourced.Example output: A section table that says what to keep, revise, add, or remove, plus a list of unsourced or dated claims and a recommended scope. The plan flags gaps that need new sources and leaves replacement statistics out.
Quality check:
- The plan is tied to the current copy and the new inputs you provided.
- No traffic, ranking, or citation change is stated unless you supplied it.
- A rewrite is recommended only when the current structure cannot carry the updated job.
Prompts for source-backed answer blocks for AI search
Task: Draft short, citation-ready answer blocks that AI search systems and readers can extract, with every claim tied to a source you supplied.
Required inputs:
- The question the block should answer
- Approved sources (URLs and the relevant excerpts)
- Entity names you want stated clearly (product, category, audience)
- Any claims you are not willing to make
Copy-ready prompt:
You are an AEO editor writing a source-backed answer block for AI search.
Inputs:
- Question to answer: [PASTE QUESTION]
- Approved sources, including excerpts: [PASTE URLS AND EXCERPTS]
- Entities to name clearly: [e.g., product name, category, audience]
- Claims to avoid: [LIST]
Task:
Write an answer block that a person or an AI search system can extract and cite. Lead with the answer. Support it with evidence from the approved sources only.
Output:
Use this structure:
1. Direct answer (1 to 2 sentences)
2. Supporting points (3 to 5 bullets)
3. Source list (each point mapped to a supplied URL or excerpt)
4. Entity recap (the names and relationships a reader should not have to infer)
5. Unverified items (anything the question asks that the sources do not support)
Constraints:
- Every factual claim must point to an approved source.
- If the sources are silent, move the point to "Unverified items" instead of filling the gap.
- Do not invent statistics, quotes, customer results, or study names.
- Do not mention competitors unless the supplied sources support a factual comparison.
- Keep the direct answer self-contained so it still makes sense when quoted alone.
Quality check before you finish:
- Map every sentence in the direct answer and each supporting bullet to a source.
- Reread the direct answer on its own and confirm it answers the question without extra context.Example output: A two-sentence answer, three to five supporting bullets, a source map that pairs each point with a URL or excerpt you pasted, and a short list of questions the sources cannot answer yet. Entity names appear in full so the block does not depend on implied context.
Quality check:
- You can trace every factual sentence to a source excerpt you provided.
- The opening answer can stand alone if an AI system quotes only that passage.
- Unverified items carry a clear label.
When to use ChatGPT vs Claude vs Perplexity
The prompt structure stays the same across models. Choose the model that matches the job.
| Model | Use it when the job is | Go to these templates |
|---|---|---|
| ChatGPT | Drafting, rewriting, and producing structured options from context you paste | ChatGPT prompts for enterprise SEO content teams |
| Claude | Reviewing long documents, comparing a draft against a brief, and following detailed constraints | Best Claude prompts for AI SEO |
| Perplexity | Research that needs live web sources and citations you can open and check | Perplexity AI SEO prompts |
Paste your own inputs in every case. A live-web model can still invent a neat summary if you do not ask it to quote and link the source. A drafting model can still stay accurate if you give it the source set and forbid unsupported claims.
For the research step behind these templates, see the prompt research glossary entry. For courses and practice, go to AirOps University.
FAQ
How complete do my inputs need to be before I run an SEO prompt?
Paste the raw material the job depends on: the export, URL list, draft, or source excerpts. The prompts above are written to use those inputs and to stop when the inputs run out. Running them on a topic name alone invites the model to guess.
Should I run the same SEO prompt in more than one model?
Use a second model when the job changes, such as moving from research to drafting, or from drafting to a long-document review. Running the same prompt in three models and averaging the answers is optional, and this page makes no claim that it improves results. Compare each output against the quality check.
How do I keep AI SEO prompts from inventing data?
State the ban in the prompt, then inspect the output for numbers, URLs, quotes, and customer results. If a figure was not in your export or source set, delete it or replace it with a clearly marked gap. The quality checks on this page exist for that review.
AirOps for SEO prompt workflows
AirOps is an enterprise AEO platform and expert partner for marketing leaders, connecting AI search intelligence, coordinated action across channels, and measurement tied to growth.
Repeatable SEO prompt work means running the same job with the same inputs, constraints, and review every time you cluster keywords, brief a page, or refresh an answer block. AirOps helps teams turn that prompt work into a system they can repeat across owned content and the other channels that shape AI discovery. Bring the templates on this page into a workflow with clear owners and QA, then measure whether the published evidence is doing the job.