On this page
1. Group keywords into topics
Prompt: “Here is a list of keywords for BUSINESS. Group them by topic and by what the searcher wants. Name each group, and say which one page could cover each.” Paste the export, not a guess.
Check the groups against live results: if the top pages differ between two keywords, they probably need separate pages. See SEO keyword research.
Ask the model to flag keywords that look like questions and keywords that look like purchases, then handle those groups differently. Keep the volumes in the data you paste, and sort by volume before you start.
Save the prompt and the output next to the keyword export, with the date. When rankings change, you can see what you assumed, and you can rerun the prompt on a new export in minutes.
2. Classify search intent
Prompt: “For each keyword, label the intent as learn, compare, buy, or find a place, and say what kind of page matches it.” Add the page type that currently ranks if you have looked.
The model has no live results, so confirm by reading the top five for any keyword you care about. A page that mismatches the intent will not rank, however well it is written.
Ask for a one-line reason for each label so you can spot mistakes quickly. When two keywords sit between intents, check the live results to see what Google actually rewards.
Want this done in your account?
Connects your ad accounts and your site. Every change waits for your approval.
3. Draft a content brief
Prompt: “Write a brief for a page on TOPIC for CUSTOMER. Give the main question to answer in the first two sentences, the subheadings, what proof to include, and what a reader does next.”
Add what you know that competitors do not: prices, photos, first-hand details. A brief is only as good as the unique material you put in. See write an SEO blog post.
Ask the model to list what a reader would need to trust the page, such as prices, credentials, or a photo, and add whatever you have. Then give the brief to a writer, or use it yourself as a checklist.
5. List the questions customers ask
Prompt: “List 15 questions a CUSTOMER asks before buying PRODUCT, ordered by how early in the decision they come. Say which are objections.” Then answer them in the customer’s own words.
Compare against Google’s People Also Ask and your sales emails. FAQ markup is on FAQ schema.
Ask for the objections separately from the how-to questions, and answer each in two or three sentences. Keep the answers factual, and drop any question your page cannot answer well.
Add one or two answers to the page itself and keep the full list for a help page. Use the language customers use in emails and calls so the answers match their searches.
6. Suggest internal links
Prompt: “Here are my page titles and URLs. For the new page about TOPIC, suggest which existing pages should link to it and the anchor text, and which pages it should link to.”
Reject links that do not help the reader. Internal linking explains what a useful one looks like.
Give the model the list of titles with a short note on each, so it can choose well. Add the links by hand, and check that each anchor text describes the target page.
7. Refresh an old page
Prompt: “Here is my page and the search queries it gets impressions for. What questions does the page fail to answer? List sections to add and statements that are out of date.”
Pull the queries from Search Console. Do not let the model invent statistics to fill gaps, and keep only facts you can source.
Update the date only when you change something real. Replace old figures with ones you can source, remove sections that no longer apply, and add a section for each missing question.
8. Draft structured data
Prompt: “Write JSON-LD LocalBusiness markup from these facts: NAME, ADDRESS, PHONE, HOURS. Use only what I provided.” Validate it before publishing.
The prompts need your data pasted in. Sprites reads Google Search Console and GA4 itself, and runs keyword research, page audits, and site crawls, then proposes article and meta title updates as approval cards you edit and approve. See local business schema.
Check the markup in a testing tool and compare each field with the page. Mark up only what is visible on the page, since mismatches can be ignored or flagged.
Questions
Can ChatGPT do SEO?
It can speed up the writing and analysis tasks, such as grouping keywords, drafting briefs, and suggesting titles. It cannot see live rankings or your analytics unless you paste them in, so verify against real search results.
Is AI-written content bad for SEO?
Google says it rewards helpful content made for people, however it is produced. Unedited generic output tends to be thin. Add first-hand detail, accurate facts, and your own examples.
Do the same prompts work in Claude?
Yes. The prompts are plain instructions, so they work in ChatGPT, Claude, and other assistants. Claude handles long pasted documents well, which suits content refreshes and brief writing.
Is SEO dead in 2026?
No. People still search, and AI answers often draw on pages that rank. The work now includes being quotable by assistants. See get cited by ChatGPT.
What makes a good SEO prompt?
It names the role, the page or data, one job, the format, and what to avoid. A prompt with your real data and a narrow task beats a long generic one.