Earlier this year, Claude told us our own website had canonical tag issues. We checked. It was wrong. Every page had the right one.
A week later it told us an ecommerce client had no product schema. Also wrong. The site had Product, Offer and review schema on every product page, and Google's Rich Results Test confirmed it.
I still use Claude for SEO every day. So does the rest of the team. It's been part of how we work for over a year now, and roughly 80% of our SEO analysis, keyword sorting and technical prep goes through Claude before a person takes over.
Both of those things are true at once, and that's really the whole post. Claude is the most useful thing we've added to our SEO work in years. It's also wrong often enough that nothing it says goes to a client until a person has checked it.
We first wrote this in April 2026 after testing more than 50 Claude Code SEO plugins. A lot has changed since, so this is the updated version.
Is Claude good at SEO?
Yes, at some of it. The short version:
It's good when the data is already in front of it and the job is to reason about it. It's unreliable when it has to go and fetch, render or check something on a live website by itself.
That one distinction explains nearly every good and bad result we've had.
Where it earns its keep
Pattern-spotting across big exports is the obvious one. Give it a few thousand rows of Search Console data and ask which queries are losing clicks while holding their position, and you'll get an answer in seconds. That used to be an afternoon for a junior analyst.
A real example. In October we wanted to know what AI search had done to the sites we manage. Claude pulled a year of Search Console data across our accounts, cut it by query length, question searches and price searches, and found that click-through on long searches had halved while impressions grew. That became Ranking better, clicked less. The analysis took an afternoon. Without it, I doubt we'd have done it at all.
Other things it does well for us:
- Sorting keyword lists by intent
- Finding internal linking gaps across a crawl
- Drafting schema for a new page template
- Redirect mapping for a site migration
- Comparing our page against the three or four pages that outrank it
Where it gets SEO wrong
The two examples at the top are the pattern: confident findings about a live site that turn out to be false.
Why does it happen? Claude works from what it managed to fetch, which isn't always what Google sees. It can miss anything added by JavaScript, anything behind a login, anything a CDN serves differently. And it knows what a broken site usually looks like, so sometimes it finds that pattern where it doesn't exist.
It also gets facts about your own business wrong if you let it. When we drafted our guide to SEO for accountants recently, the first version quoted our SEO retainers at $2,000 a month. That was our PPC price, pulled from an old internal document. I caught it on the read-through. Nobody else would have.
And a second AI doesn't fix this. We ran the Search Console post past a different AI model for a review. It came back with a nicely formatted table saying click-through on short searches had gone up. The data said it had fallen 30%. A review that reads well and gets a number backwards is worse than no review.
What's changed since April: connectors and skills
When we first wrote this, using Claude for SEO mostly meant pasting CSV files into a chat.
Two things have changed how we work.
Connectors. Claude can now talk to the tools directly. Ours is connected to Ahrefs and to Search Console, so instead of exporting a spreadsheet we ask a question and it pulls the live numbers. This matters more than it sounds. Most of the made-up findings we used to see came from Claude guessing about data it didn't have. When it's reading the real thing, it guesses a lot less.
Connectors do break. Our Screaming Frog one was down on the day we launched the new site, so we ran the crawl the old way from the command line. Have a plan B.
Skills. A skill is a saved set of instructions for one job, so Claude does it the same way every time. We've built them for technical audits, for Search Console reviews, and one that holds our writing voice so drafts don't come out sounding like every other AI post on the internet. There are public SEO skill packs around too. We've tried a lot of them. Most are fine as a starting point and too generic to use as they come.
The two we lean on most are ones we built ourselves: one that flags unusual movements in Search Console data, and one that checks every draft against our house style. As for public plugins, skip anything that promises a one-click technical audit from a URL. Without a real crawl underneath, it just makes up problems faster.
How to use Claude for your SEO
If you want to try this on your own site, here's the order we'd do it in.
- Give it real data first. A crawl export, your Search Console data, and the pages you're trying to beat. Paste them in or connect the tools. Don't ask it to "audit my site" from a URL alone.
- Ask narrow questions. "Which pages lost clicks but kept their ranking?" beats "how's my SEO?"
- Make it argue with itself. After any audit, ask: "Act as a sceptical SEO lead. Find evidence in the data that contradicts these findings." It catches a good share of its own mistakes this way.
- Check anything about the live site by hand. Canonicals, schema, redirects, robots.txt. View the source or use Google's own testing tools.
- Read everything before it ships. Especially prices, names, dates and numbers.
Here's how we split it:
| Let Claude do it | Check before acting | Keep a person on it |
|---|---|---|
| Keyword sorting and clustering | Technical audit findings | Strategy and priorities |
| Search Console pattern analysis | Canonical and schema checks | What to tell the client |
| Schema and content drafts | Redirect chains | Final sign-off on live changes |
| Competitor gap analysis | Any claim about a live URL | Judgement calls on risk |
Can Claude replace an SEO agency?
I'm obviously biased, so take this for what it is.
Not yet. If you're a business owner hoping Claude means you can skip hiring anyone for SEO, I'd wait. The mistakes it makes are exactly the ones a non-specialist won't spot.
If you already have an SEO person or a marketing manager, start today. It makes the boring parts faster and it makes some analysis possible that you'd never have had time for.
Inside One Egg it hasn't reduced our headcount. It's changed where the hours go. The team spends far less time wrestling with spreadsheets and far more on strategy and on fixing what the data turns up.
Simple rule: use it for everything, trust it for nothing until someone's checked.
If you'd like to see what this looks like on your own site, get in touch. And if your real question is how to show up when people ask ChatGPT or Claude for a recommendation, that's a different job, and it's what our AI SEO service does.
FAQs
Is Claude good at SEO?
For analysis, yes. It's fast and sharp when you give it real data like crawl exports and Search Console numbers. It's unreliable when it has to check a live website by itself, where it can report problems that don't exist.
How do I use Claude to do my SEO?
Give it your Search Console data, a crawl of your site and the competitor pages you want to beat. Ask specific questions, ask it to challenge its own findings, and check anything technical by hand before you change your site.
What are Claude SEO skills?
A skill is a saved set of instructions that makes Claude do one job the same way each time, such as a technical audit or a content brief. Public SEO skill packs are a reasonable starting point, but most need adapting to how you actually work.
Can Claude replace my SEO agency?
Not yet. It speeds up specific tasks, but it still makes confident mistakes on technical audits and on facts about your business. It works best alongside someone who can tell when it's wrong.
