seo & ai search · playbook
ChatGPT SEO prompts: 27 copy-paste prompts and how to use them
27 copy-paste ChatGPT SEO prompts for keyword research, briefs, titles, schema, internal links, content refreshes and GEO, plus the guardrails that keep output accurate.
Key takeaways
- ChatGPT is good at structuring, clustering, drafting and critiquing SEO work, and bad at search volumes, difficulty scores and current rankings, which it does not have.
- Feed it your real data (Search Console exports, competitor headings, your page copy) and ask for structured output such as tables, CSV or JSON.
- Every prompt here has a fill-in-the-brackets context block; the more specific the context, the less generic the output.
- Validate schema in Google's Rich Results Test and check every fact, number and URL before publishing.
- Google does not penalize AI assistance as such, but scaled low-value content made primarily to manipulate rankings violates its spam policies.
The best way to use ChatGPT for SEO is as an analyst and editor working on data you give it, not as a source of search data. It does not have search volumes, keyword difficulty or live rankings, so pair it with Search Console, Google Keyword Planner, Ahrefs or Semrush for numbers, then use the prompts below to cluster, brief, rewrite, mark up and audit. Below are 27 copy-paste ChatGPT SEO prompts grouped by task, each with brackets to fill in, followed by the guardrails that keep the output safe to publish.
How should you use ChatGPT for SEO?
ChatGPT is strong at tasks where the inputs are in the prompt and the output is structure or language. It is weak where the answer depends on proprietary or live data.
| Good at | Bad at (use a tool instead) |
|---|---|
| Clustering keyword lists by intent | Search volume and keyword difficulty |
| Turning SERP research into a content brief | Knowing who ranks today (unless it browses, and even then it sees one result set) |
| Writing title and meta description variants | Predicting CTR |
| Drafting JSON-LD from page content | Guaranteeing schema is valid or eligible for rich results |
| Suggesting internal links from a page list you supply | Crawling your site |
| Critiquing a draft for gaps and extractability | Verifying facts and statistics |
Five rules that make every prompt below work better:
- Paste the data. Keyword exports, Search Console CSVs, competitor H2s, your current copy. Output quality tracks input quality.
- State the audience and offer. “B2B SaaS, selling to RevOps leaders at 50-500 person companies” beats “my website.”
- Ask for a format. Tables, CSV or JSON are easier to check and import than prose.
- Ask it to flag uncertainty. Add “Mark anything you are not sure of with [VERIFY].”
- Iterate. Follow up with “Cut this by 30%,” “Make option 3 more specific,” or “Which of these would you cut and why?”
A reusable context block to paste above any prompt:
Context:
- Site: [domain]
- Business: [what you sell, in one sentence]
- Audience: [role, company size, industry]
- Primary goal of this page: [signups / demo requests / leads / ranking for X]
- Tone: [plain, expert, no hype]
- Constraints: American English, no invented statistics, mark uncertain claims with [VERIFY]Keyword research prompts
ChatGPT cannot tell you how many people search a term. Use it to expand and organize ideas, then check every keyword in a tool with real volume data before you commit.
1. Seed expansion by intent
I sell [product] to [audience]. List 40 search queries my buyers might type into Google, grouped into: problem-aware, solution-aware, comparison, and brand/pricing. Use the words buyers use, not our marketing terms. Do not estimate search volume. Output as a table with columns: query, group, likely intent (informational/commercial/transactional).2. Cluster a keyword export
Below is a keyword export with volume and difficulty from [tool]. Cluster these keywords so that each cluster can be served by one page. Name each cluster by its best head term, list the keywords in it, sum the volumes, and note the dominant intent. Flag keywords that probably need their own page because the intent differs. Output as CSV.
[paste keyword, volume, KD rows]3. Find question keywords
For the topic "[topic]", list 25 questions a [audience] would ask before buying [category of product]. Phrase each exactly as someone would type or say it. Group by stage: learning, evaluating, deciding. I will check volumes separately.4. Mine Search Console for quick wins
Here is a Search Console export with query, page, clicks, impressions, CTR and position. Find:
1) queries in positions 4-15 with high impressions (top opportunities),
2) queries where the ranking page looks like the wrong page for the intent,
3) pairs of pages that appear to compete for the same query.
Explain each finding in one line and output three tables.
[paste CSV]5. Map keywords to existing pages
Here is a list of my URLs with their titles and H1s, and a list of target keywords. Assign each keyword to the single best existing page, or mark it "NEW PAGE" if nothing fits. Flag any page that would end up targeting two different intents. Output as a table: keyword, assigned URL or NEW PAGE, reason.
URLs:
[paste]
Keywords:
[paste]Content brief prompts
6. Brief from competitor headings
Target keyword: "[keyword]". Below are the H2 and H3 headings from the top 5 ranking pages. Build a content brief that covers every subtopic most of them cover, plus three angles none of them cover well. Include: recommended title, one-sentence direct answer to open the page, H2 outline as questions, what each section must answer, suggested tables or lists, and questions to answer in an FAQ. Do not invent statistics; list data I should source instead.
[paste competitor headings with URLs]7. Search intent check
For the query "[keyword]", describe the likely search intent, the format that would satisfy it (guide, list, tool, comparison, template, product page) and what a searcher would consider a complete answer. Then tell me if my planned page "[planned title and angle]" matches that intent, and what to change if not.8. Information gain
Here is my draft outline for "[keyword]". Assume competing pages already cover the basics. Suggest 8 additions that would give a reader something they can't get elsewhere: first-hand steps, decision rules, worked examples, templates, original comparisons or data we could collect. For each, say what input I'd need from my team.
[paste outline]9. Turn an expert interview into a brief
Below is a transcript of an interview with our [role] about [topic]. Extract: their specific opinions, numbers, examples, mistakes they warn about and steps they follow. Then propose an article outline for "[keyword]" built around those first-hand insights. Quote them only where the transcript supports it.
[paste transcript]Title tag and meta description prompts
10. Title variants
Write 10 title tag options for a page targeting "[primary keyword]". The page offers [what's on the page]. Rules: include the primary keyword near the start, aim for roughly 50-60 characters, no clickbait, no "Ultimate guide". Include character counts. Then pick the best 3 and explain why.11. Meta descriptions
Write 5 meta descriptions for this page. Each must: summarize what the reader gets, include "[primary keyword]" naturally, name 2-3 specific things the page covers, and stay around 140-160 characters. Show character counts. Do not repeat the title.
Title: [title]
Page summary: [3-5 bullet summary]12. Fix low-CTR pages
These pages rank in positions 1-10 but have low CTR. For each, I've included the current title, meta description and top 5 queries. Diagnose why the snippet underperforms for those queries, then write a new title and meta description for each. Output as a table: URL, problem, new title, new meta.
[paste rows]For the full process behind this prompt, see the Search Console CTR optimization guide.
Schema markup prompts
Always run generated markup through Google’s Rich Results Test or the Schema Markup Validator, and only mark up content that is visible on the page.
13. Article schema
Generate JSON-LD Article schema for the page below. Use only information present in the page content I provide. Include headline, description, author (Person with name and URL), publisher (Organization with name and logo URL), datePublished, dateModified and mainEntityOfPage. Output only the script tag.
[paste title, author, dates, URL, logo URL]14. Product or SoftwareApplication schema
Create JSON-LD SoftwareApplication schema for [product]. Use these facts only: [name, category, operating system or "Web", pricing, rating if we have genuine reviews with a source]. If a required or recommended property is missing from my facts, list it instead of inventing a value.15. BreadcrumbList and Organization
Generate JSON-LD for (1) an Organization with name, URL, logo and sameAs links [paste profiles], and (2) a BreadcrumbList for the URL [URL] with this path: [Home > Section > Page]. Output two separate script tags.16. Audit existing schema
Here is the JSON-LD from my page and the visible page text. Check: syntax errors, properties that do not match visible content, deprecated or unnecessary types, and missing recommended properties for this type. Note that Google no longer shows FAQ or HowTo rich results, so flag those as descriptive only. Output a list of fixes.
[paste JSON-LD and page text]Internal linking prompts
17. Find link opportunities
Here is a list of my published pages with URL, title and a one-line summary. For the new page "[title, URL, summary]", suggest the 5-8 existing pages that should link TO it and the 3-5 pages it should link OUT to. For each link, give the exact sentence or anchor text where it would fit naturally. Avoid exact-match anchors on every link.
[paste page list]18. Topic cluster structure
These are all my pages about [topic]: [paste list]. Propose a hub-and-spoke structure: which page should be the hub, which are spokes, which pages overlap and should be merged, and what spokes are missing. Output as an indented list.19. Anchor text audit
Below is an export of internal links (source URL, target URL, anchor text). Find: generic anchors ("click here," "read more"), targets with only one or two inbound links, and targets where many anchors use the same exact text. Suggest better anchors for the 15 most important fixes.
[paste CSV]Content refresh prompts
20. Refresh audit
This article was published on [date] and has lost [X]% of clicks over [period]. Here is the article and the H2s of the 3 pages now outranking it. List: outdated sections, missing subtopics, sections that should be cut, places where a table or step list would help, and claims that need a fresh source. Prioritize by likely impact.
[paste article and competitor headings]21. Rewrite an intro to answer first
Rewrite this introduction so the first two sentences directly answer the query "[keyword]" in a way that could be quoted on its own. Keep it under 70 words. No throat-clearing, no "In today's world."
[paste intro]22. Merge cannibalizing pages
These two pages compete for "[keyword]". Page A: [paste]. Page B: [paste]. Recommend which should be the surviving URL and why, then produce a merged outline that keeps every unique section from both. List which URL should 301 redirect to which.GEO and AI search prompts
These help you write content that AI Overviews, ChatGPT and Perplexity can extract and cite. For the principles behind them, read the AEO / GEO content optimization guide.
23. Extractability review
Review this page as if you were an AI search system choosing passages to cite for the query "[query]". For each H2 section, tell me: does the first sentence answer the heading on its own, is there a specific number or named entity, and could the section be quoted out of context? Then rewrite the three weakest section openings.
[paste page]24. Question headings
Rewrite these H2 headings as the natural-language questions a buyer would ask an AI assistant, keeping the same section content. Keep each under 70 characters and sentence case.
[paste headings]25. Entity and positioning clarity
Read our homepage and product page copy below. In one sentence each, state: what category we are in, who we are for, and when someone should choose us over [competitor]. If you cannot answer any of these from the copy alone, tell me what is missing or ambiguous.
[paste copy]26. Build an AI visibility prompt set
I want to track whether AI assistants recommend [brand] for buyers in [category]. Write 30 prompts real buyers would ask ChatGPT, split into category discovery, problem/use case, comparison, alternatives and brand facts. Make them full questions with realistic context (team size, budget, tools they use).Use the output as the starting prompt set described in how to check your LLM SEO visibility.
27. Comparison table from your facts
Using only the facts below, build an HTML table comparing [our product] with [competitor 1] and [competitor 2] on: best for, pricing model, key features, integrations, limitations. Where I have not supplied a fact, write "not verified" rather than guessing. Then write a two-sentence neutral summary of who should choose each.
[paste facts with sources]What are the guardrails for ChatGPT SEO output?
Run through this checklist before anything ChatGPT produced goes live.
| Risk | What happens | Guardrail |
|---|---|---|
| Invented search volumes | ChatGPT has no keyword database; numbers it gives are guesses | Get volume and difficulty from Google Keyword Planner, Ahrefs, Semrush or Search Console |
| Hallucinated facts and statistics | Plausible numbers, studies or quotes that don’t exist | Require a source for every number; delete anything you can’t verify |
| Fake or broken URLs | Links to pages that don’t exist | Click every link before publishing |
| Outdated guidance | Model knowledge has a cutoff, and SEO features change (FAQ rich results, for example, stopped appearing in Google Search on May 7, 2026) | Check claims about Google features against Search Central docs |
| Invalid schema | Missing required properties or markup that doesn’t match the page | Rich Results Test or Schema Markup Validator |
| Generic content | Pages that restate what already ranks | Add first-hand experience, data, examples and opinions (prompts 8 and 9) |
| Scaled content | Hundreds of thin AI pages | Google’s spam policies treat scaled content made primarily to manipulate rankings as abuse, regardless of how it’s produced |
| Confidential data | Customer data pasted into prompts | Strip personal data; follow your company’s AI usage policy |
Google’s position, from its Search Central guidance on AI-generated content, is that it rewards helpful, original content however it is produced, and that using automation, including AI, with the primary purpose of manipulating rankings violates its spam policies. In practice that means ChatGPT can help you produce better pages faster, but a human with subject knowledge needs to own each page.
When should you use a dedicated tool instead?
Prompts are good for one-off analysis. Once the same task repeats weekly (keyword gap checks, title rewrites across hundreds of pages, technical audits, rank monitoring), a prompt in a chat window stops scaling. GTM Hacker’s SEO and GEO agents run those recurring jobs on real Ahrefs, Keyword Planner and site-crawl data and file the findings as tasks on your SEO board for you to approve.