← Back to blog

Can AI (ChatGPT, agents) do SEO for you? What it can and can't do

ChatGPT, Claude, Gemini and a growing set of AI agents can now write pages, cluster keywords and spot broken schema in seconds. That raises a fair question: why pay a person, or an agency, for SEO at all?

The short answer is that AI is very good at the production work and weak at the judgement work. This post splits SEO into tasks, shows which ones AI handles well, and uses our own workflow as the example. If you want the content side done for you, see our AI search content service.

Is there an AI that can do SEO?

Yes, several AI tools can do parts of SEO well, but no single AI can do all of it on its own. Large language models can research, draft and check. They cannot know your business, prove your claims or persuade other sites to mention you, and those are the parts that move rankings and AI citations.

SEO is a bundle of very different jobs: research, planning, writing, technical fixes, link and mention building, and measurement. AI speeds up some of these a lot. Others still depend on access, relationships and judgement that a model does not have.

What AI does well, what needs a human, and what it gets wrong

Can ChatGPT do SEO?

ChatGPT can do useful SEO work if you give it good inputs: a keyword list, your product facts and a clear brief. It can draft pages, write meta descriptions, suggest FAQs and produce schema. It cannot check live rankings on its own, verify its own claims or see how other AI engines describe you.

Good ways to use ChatGPT for SEO

  • Turn a list of buyer questions into a page outline, with one H2 per question.
  • Rewrite an existing page so the first two sentences answer the main question.
  • Draft FAQ answers from your own notes, then check every fact.
  • Write JSON-LD for Organization, FAQPage or Product schema. Our free schema generator does the same without a prompt.
  • Summarise a crawl report into a ranked fix list.

Where ChatGPT goes wrong

  • Made-up numbers. Ask for search volume or market size and it may give you a confident figure with no source. Always pull numbers from a real tool or report.
  • Generic copy. Without your proof and examples, it writes the same page your competitors have. AI assistants have little reason to quote a page that adds nothing new.
  • Out-of-date facts. It may describe your product, pricing or competitors as they were some time ago.

There is a second, separate question: can you do SEO for ChatGPT, so that ChatGPT recommends you? That is answer engine optimization, and we cover it on our ChatGPT optimization page.

Can AI agents do SEO?

AI agents can run multi-step SEO tasks on their own, such as crawling a site, finding pages with missing schema, drafting fixes and opening a pull request. They still need a human to set the goal, approve changes and handle anything that involves judgement, other people or real-world proof.

Agents are the step beyond a chat window. Instead of answering one prompt, an agent can read files, call tools, check its work and keep going. For SEO that means it can take a whole job, like "add FAQ schema to every service page and validate it", from start to finish.

Our own workflow, as an example

We build this website with Claude as a coding and writing agent, and the process shows where the line falls.

  1. A human writes the brief. Our founder wrote the site brief and a brief for every priority page and post: target keyword, the questions to answer, what must be included and what proof is needed.
  2. The agent drafts to strict rules. Claude drafts each page as structured content, following the brief, our style rules and our schema model. Where a fact, number or client result is needed and we do not have it, the rule is to write a placeholder such as "PROOF NEEDED" rather than invent one.
  3. Scripts check the draft. Automated checks confirm the answer-first paragraph length, heading order, FAQ length, link targets and banned words before a draft is saved.
  4. A human reviews and fills the gaps. We add real numbers, screenshots and examples, check every claim, and decide what gets published.

The agent saves a great deal of production time. It does not decide what to say, and it is not allowed to make up proof. [PROOF NEEDED: screenshot of the workflow, and time saved per page if measured]

What an agent should not do unsupervised

  • Publish pages straight to your live site without review.
  • Change robots.txt, canonicals or redirects. One wrong line can remove pages from search or block AI crawlers. Our technical AEO checks cover these settings.
  • Send outreach emails in your name.
  • Add claims, prices or client names that nobody has checked.

Give agents narrow, testable jobs and a clear stop point where a person approves the result.

Can AI do SEO optimization?

Yes, AI can optimise existing pages: tightening titles, rewriting intros answer-first, adding FAQs, fixing headings and generating schema. Results depend on what you feed it. Optimisation that only rewords a page rarely moves rankings; adding real information, structure and proof does.

A simple AI optimisation pass

  1. Pick one page and the three to five questions it should answer.
  2. Ask the AI to rewrite the opening so the first sentence answers the main question in plain words.
  3. Ask it to add one H2 per question, each with a direct answer first.
  4. Add your own proof: a number, an example, a screenshot or a quote you have permission to use.
  5. Generate and validate schema, then check the page loads fast and renders without JavaScript.
  6. Re-check the page in search and in AI answers a few weeks later.

Steps 4 and 6 are the ones AI cannot do for you, and they are the ones that make the difference.

The risk of publishing raw AI output

When many companies publish AI drafts on the same topic, the pages start to look alike. Search engines have little reason to rank one over another, and AI assistants have little reason to cite any of them. The pages that stand out carry something a model could not have produced alone: first-hand numbers, screenshots, real prompts and answers, or a view from someone who has done the work.

Which AI is best for SEO?

The best AI for SEO depends on the task. General assistants such as ChatGPT, Claude and Gemini are all capable for drafting, outlining and schema. Agent tools suit repeatable technical work. Dedicated SEO platforms with AI features suit research, because they hold real search data that a chatbot does not.

  • For drafting and editing: use whichever general assistant follows your style rules most reliably. Test two or three on the same brief before you commit.
  • For technical work across many pages: use an agent that can read your codebase or CMS and run checks, with a human approving each change.
  • For keyword and competitor data: use an SEO platform with real data. Do not trust volumes from a chatbot.
  • For checking AI visibility: use a tracking tool or an audit that runs prompts across engines. Our post on what is an AEO audit explains how that works.

So should you let AI do your SEO?

Let AI do the production work and keep humans on strategy, proof and outreach. That split is how we run our own AI search content service: AI drafts and checks, people decide and verify. It is faster than writing by hand and safer than publishing raw AI output.

If you want to know whether your current pages are being picked up by AI assistants at all, book an AI Visibility Audit. We run your buyers' questions across ChatGPT, Perplexity and Google AI Overviews and show you exactly where you appear.

Need GTM creative shipped at AI speed?

One team for content, decks, landing pages, SEO/GEO, and video.

Get started