ChatGPT is good at the thinking parts of SEO: grouping keywords by intent, outlining pages, writing titles and meta descriptions, and turning an audit into a to-do list. On its own, though, it has no keyword database, so any search volume or difficulty score it gives you is a guess. The fix is to connect a live data source, such as Manifold MCP, Semrush or Ahrefs, so ChatGPT works from real numbers and real search results.
This guide walks through a workflow you can run in one chat, from keyword research to a finished content brief.
Can ChatGPT do keyword research?
Yes, with a caveat. ChatGPT can brainstorm the words your buyers use, spot gaps in your topic coverage and sort ideas by intent. What it cannot do by default is tell you how many people search for each term, how hard it is to rank, or who ranks today. If you ask, it will often produce plausible-looking numbers. Treat those as placeholders, never as data.
There are two ways to get real numbers into the chat:
- Paste data in. Export keywords from your SEO tool as a CSV and upload it. This works, but the data is a snapshot and you repeat the export every time you need something new.
- Connect a data source through MCP. MCP (Model Context Protocol) lets ChatGPT call outside tools mid-conversation. Once connected, you can ask "what's the search volume for these 50 keywords?" and ChatGPT fetches the answer itself.
The second option is what turns ChatGPT from a writing assistant into an SEO research assistant.
Which SEO data sources work with ChatGPT?
Several, and the right one depends on what you already pay for.
- Manifold MCP is a hosted MCP server that gives ChatGPT, Claude and Cursor live marketing data. It is hosted, you need no API keys of your own, and you pay per call. Its SEO tools cover keyword ideas with volume, difficulty, intent and CPC, search results for any query, the keywords a domain ranks for, keyword gaps against a competitor, backlinks, traffic estimates and technical site crawls. Search and keyword data comes from DataForSEO. If you connect your own Google Search Console or Bing Webmaster Tools accounts in the Manifold MCP app, your agent can also read your own clicks, rankings and indexing, and those tools are free.
- Semrush has its own Semrush MCP, listed as an official ChatGPT app. Semrush says MCP access is included in its Semrush One and SEO Classic plans and is also available for API users. A good fit if your team already lives in Semrush.
- Ahrefs offers a hosted Ahrefs MCP server for users on a Lite plan or higher, with usage counted in Ahrefs API units. A good fit if Ahrefs is already your source of truth.
If you don't have an SEO subscription, a pay-per-call server is the cheapest way to start. If you already pay for Semrush or Ahrefs, their own integrations keep your numbers consistent with the tool your team already reports from.
How do I connect live keyword data to ChatGPT?
Here is the setup for Manifold MCP, taken from its docs. Other servers follow a similar pattern.
- In ChatGPT on the web, open Settings, then Security and login, and turn on Developer mode. ChatGPT needs it to add an MCP server by URL.
- Go to chatgpt.com/plugins, click Add at the top right, then Create MCP app.
- Name it Manifold MCP and paste the server URL:
https://mcp.manifoldmcp.com/mcp. Leave authentication on OAuth. - Confirm, then sign in when ChatGPT asks. A new workspace starts with 500 free credits and no card.
- In a chat, type @ followed by the app name you chose and ask a question.
On a ChatGPT Business workspace only an admin can use developer mode, so ask your admin to add it. In Claude, the same URL goes under Settings, then Connectors, as a custom connector.
What is a good ChatGPT SEO workflow?
Run these steps in order. Each one feeds the next, and ChatGPT keeps the context.
1. Find keyword ideas with real metrics
Start with a seed topic and ask for ideas with volume, difficulty and intent.
Use Manifold MCP to find 100 keyword ideas around "crm for startups" in the US.
Show volume, keyword difficulty, intent and CPC. Then group them into
topic clusters and mark which clusters look realistic for a site with
modest authority.
2. Check what Google actually shows
Volume tells you demand. The results page tells you what kind of content wins. Ask ChatGPT to pull the top results for your shortlisted keywords and describe the pattern: listicles, product pages, tools, forums, or an AI Overview on top.
For the 5 keywords I picked, get the top 10 Google results in the US,
including any AI Overview and the sources it cites. Tell me what type of
page ranks for each one and whether a blog post can realistically compete.
3. See what competitors rank for that you don't
A keyword gap is the fastest source of proven topics.
Run a keyword gap between example.com (us) and competitor.com. List the
top 50 keywords they rank for that we don't rank for at all, sorted by
volume, with their ranking URL. Group them by topic.
4. Turn the research into a content brief
Now ChatGPT does what it does best. Because the data is already in the chat, the brief reflects real search results rather than generic advice.
Write a content brief for "best crm for startups" using the SERP data
above. Include search intent, a recommended title and meta description,
an H2 outline phrased as questions, the subtopics every top result covers,
and one angle none of them cover.
5. Audit the page or site
Before publishing, or when refreshing old content, check the technical basics. Manifold MCP includes a free on-page snapshot for one URL and a paid technical crawl for a whole site.
Audit example.com for technical SEO: crawl up to 500 pages and list broken
links, duplicate titles and descriptions, and non-indexable pages. Then
give me a prioritized fix list with the 10 most important items first.
How much does it cost to use ChatGPT for SEO?
ChatGPT itself requires whichever plan gives you developer mode and connectors. The data source is the variable cost.
With Manifold MCP, every call has a published credit price. According to its pricing page, 100 keyword ideas with metrics cost 10 credits and the top 10 Google results for a keyword cost 1 credit. You start with 500 free credits, then top up from $10 (top-up credits never expire) or take Pro from $20 a month for 2,500 credits. Your agent can check a call's price before it runs and cap what any call spends.
With Semrush or Ahrefs, MCP usage is tied to your existing plan and its API unit limits, so check those vendors' pages for what's included.
What should you not use ChatGPT for in SEO?
- Inventing data. If no tool is connected, don't publish numbers ChatGPT gives you.
- Publishing first drafts unedited. Use ChatGPT for briefs, outlines and rewrites, then add your own expertise, examples and screenshots.
- Rank tracking from memory. ChatGPT doesn't remember yesterday's positions. Pull fresh positions each time, or use a scheduled rank tracker.
If your team prefers Claude, the same workflow carries over. See how to use Claude for SEO for a Claude-specific version.
Questions people ask
Is ChatGPT good for SEO?
ChatGPT is good for SEO tasks that involve reasoning and writing: clustering keywords, outlining pages, drafting titles, and summarizing audits. It is not a data source on its own. Connect a live SEO tool so its recommendations are based on real search volumes and real results.
Can ChatGPT give me accurate search volumes?
Not by itself. Without a connected data source, ChatGPT estimates. With an MCP server such as Manifold MCP, Semrush MCP or Ahrefs MCP connected, it pulls volumes from that provider's database and can show you where each number came from.
Do I need a Semrush or Ahrefs subscription to use ChatGPT for SEO?
No. Semrush and Ahrefs MCP access is tied to their paid plans, but a pay-per-call server like Manifold MCP needs no SEO subscription and no API keys of your own. You sign in with OAuth and pay only for the calls you make.
Can ChatGPT write SEO content that ranks?
ChatGPT can draft content quickly, but ranking depends on matching search intent, covering the topic better than current results, and adding real expertise. Give it the live SERP and competitor data first, then edit its draft with your own knowledge and examples.
Does this workflow work in Claude or Cursor too?
Yes. MCP is an open standard, so the same server URL works in Claude, Cursor and other MCP clients. The prompts in this guide work unchanged.