ChatGPT prompts for SEO are shortcuts for the mechanical half of the work, clustering keywords, drafting titles, outlining a brief, writing schema, so your time goes to the judgement calls a model can't make. This page is a copy-paste library: one prompt for each common SEO task, each followed by why it works and where it still needs a human. Use them as starting points, not finished answers. A prompt that produces plausible output fast becomes a liability the moment you can't tell when it's wrong, and on SEO tasks, it's wrong often enough to matter.
Two habits make every prompt below better. First, give ChatGPT context up front, what your business does, who you sell to, and where, because a prompt with no context returns generic output that fits nobody. Second, treat the output as a first pass to check, not a decision to ship. That second habit is the whole difference between using these well and getting quietly buried by your own plausible-looking content.
Keyword and topic research
Act as an SEO strategist. My business is [what you do, for whom, where]. List 20 topics my potential customers search for, grouped by where they are in the buying journey, just researching, comparing options, or ready to buy. For each, give the core phrase and two or three ways a real person might type or ask it.
Why it works: forcing the buying-journey grouping stops you getting one flat list dominated by high-level informational terms, and surfaces the commercial phrases actually worth a page.
Where it needs you: it invents demand it can't measure. Treat every phrase as a hypothesis and check real search volume and difficulty before you commit a page to it.
Search-intent mapping
Here are 15 keywords: [paste list]. For each, tell me the dominant search intent (informational, commercial, transactional, navigational), the page type that usually ranks for it (blog post, comparison, product or service page, tool), and in one line what a searcher wants to see first.
Why it works: intent decides page type, and getting it wrong is the most common reason a well-written page never ranks. This gives you a fast first pass across a whole list at once.
Where it needs you: the model guesses intent from the words, not from live results. For anything ambiguous or high-value, open the actual search results and confirm, what Google currently ranks is the only real answer.
Titles and meta descriptions
Write 8 title-tag options for a page targeting [keyword]. Each under 60 characters, leading with the keyword where it reads naturally, no clickbait, written for [audience]. Then write two meta descriptions under 155 characters that each contain the keyword and one concrete reason to click.
Why it works: loading the constraints up front, length, keyword position, audience, no clickbait, gets usable options instead of generic ones, and eight of them gives you real range to choose from.
Where it needs you: character counts drift, so verify the lengths yourself, and make sure each title matches what's actually on the page. A title that oversells wins the click and loses the trust.
Content briefs and outlines
Build a content brief for an article targeting [keyword], search intent [X], for [audience]. Include: the one-sentence promise of the page, a suggested H1, an H2/H3 outline where every heading is a real question a reader would ask, the subtopics a thorough piece must cover to be complete, and 5 FAQ questions. Keep headings specific, not generic.
Why it works: a brief built around questions maps directly onto how both readers and AI assistants consume a page, and the "must-cover subtopics" list doubles as a completeness check.
Where it needs you: the model doesn't know what's already ranking. Compare its outline against the pages currently winning the query and add what it missed, it drifts toward the obvious and skips the angle that would set you apart.
On-page optimisation
Here's my draft for a page targeting [keyword]: [paste]. Without padding it or stuffing the keyword, tell me: where the main topic isn't clear in the first 100 words, any question in my headings the body doesn't actually answer, passages too long or vague to be quoted cleanly, and three specific improvements. Critique it, don't rewrite it.
Why it works: asking for critique rather than a rewrite keeps your voice and hands you decisions to make, instead of a blander, homogenised version of your own page.
Where it needs you: it will sometimes invent problems to look useful, or push a keyword density that reads badly. Take the notes that are true; ignore the ones that would make the page worse to read.
Internal linking
Here is a list of my site's pages with their titles and target keywords: [paste]. Suggest internal links between them. For each, give the source page, the destination, and a natural anchor phrase. Prioritise links that pass authority to my most commercial pages, and flag any page that looks orphaned.
Why it works: it spots linking opportunities across a long list faster than you can hold them in your head, and the orphan flag catches pages that nothing currently points to.
Where it needs you: it only knows the pages you pasted and can't see which links already exist. Confirm each suggestion is genuinely relevant and not already in place before adding it, an irrelevant internal link helps nothing.
Schema and FAQ markup
Generate FAQPage JSON-LD schema for this page. Here are the exact questions and answers as they appear on the page: [paste]. Return valid JSON-LD only, using the wording from the page verbatim.
Why it works: schema is fiddly, mechanical and easy to get syntactically wrong, exactly the kind of task worth handing off, and feeding it your real on-page Q&As keeps the markup honest.
Where it needs you: only ever mark up content that actually appears on the page, or you risk a manual penalty. Validate the output in a schema testing tool before it ships; a model can produce JSON that looks right and still fails validation.
Optimising for AI-search visibility
This is the job most prompt lists skip, and the one moving fastest. More of your buyers now put their question to an AI assistant and read the answer it writes instead of scanning ten links, and that answer names only two or three sources. Being one of them is a different task from ranking, and prompts help you audit and prepare for it.
Take the question a buyer in my category would put to an AI assistant, for example, "[realistic buyer question]". Answer it the way an assistant would, naming specific providers. Then tell me what makes those named sources quotable, and what a page would need to be included in that answer.
Why it works: it shows you the shape of the answer your buyers already see, who gets named and why, so you're optimising toward a real target instead of guessing.
Where it needs you: an assistant's live answer draws on the current web and shifts as pages are re-crawled; the model can't reproduce that from memory. Use this to understand the pattern, then check who actually gets cited today with a live assistant.
Rewrite the opening of each section of this page so it answers the section's heading in the first two sentences, as a standalone statement a source could quote directly, without losing my voice: [paste].
Why it works: assistants lift passages, not whole pages. A section that opens with a clean, self-contained answer is far likelier to be the one quoted than one that opens with throat-clearing.
Where it needs you: watch it doesn't flatten everything into the same rhythm or strip the specifics that make you worth citing. Keep the named examples and concrete numbers; those are what get quoted.
What prompts can't do, and why it matters now
Every prompt above speeds up producing the work. None of them close the loop. They can't tell you which of the pages you published actually earned traffic or enquiries, which to cut, and which to double down on, and that measurement, not the drafting, is the part that decides whether any of this paid off.
That gap is widening on purpose. As the same prompts spread, the volume of competent-looking content climbs for everyone, and output alone stops being an edge. What separates the businesses pulling ahead is precision, knowing which work is landing and moving spend toward it while everyone else keeps producing blind. The prompts get you moving faster; they don't tell you whether you're moving in the right direction.
Frequently asked questions
Are ChatGPT prompts enough to do SEO on their own?
No. Prompts speed up the mechanical work, research, briefs, drafts, schema, but SEO is decided by judgement and measurement: choosing the right targets, verifying the output, and knowing which published pages actually earn traffic and enquiries. A prompt can't tell you what's working; that's the part that determines results.
What's the single best ChatGPT prompt for SEO?
There isn't one. SEO is a set of distinct jobs, keyword research, intent mapping, briefs, on-page, schema, AI-search visibility, and each needs its own prompt with its own context. A library you adapt to your business beats one "magic" prompt every time.
Will content written by ChatGPT rank?
It can, but not on autopilot. Unedited AI output tends to be generic, occasionally wrong, and indistinguishable from everyone else's. Pages that rank and get quoted still need a human to add specifics, verify claims, and shape the piece around what's genuinely missing from the current results.
Is it safe to use ChatGPT to generate schema markup?
Yes, if you only mark up content that actually appears on the page and validate the output before publishing. Marking up content that isn't visible risks a manual penalty, and models can produce JSON-LD that looks correct but fails validation, so always test it first.