Tooling
What an SEO can actually do with Claude Code
Claude Code is a coding assistant that runs in a terminal and can read and write files in a folder on your machine. That description puts most SEOs off, which is a shame, because the work it is good at is the work SEOs spend their week on: reading large exports, comparing two versions of a page, and doing the same careful check across 200 URLs without getting bored on URL 40.
It is a file reader before it is a code writer
Point it at a folder of crawl exports and ask a question in plain English. Which pages lost their canonical between the two crawls. Which titles are duplicated across templates. Which redirect rows resolve to a different status than the map claims. It reads the files, does the comparison and shows its working, and you can ask it to save the answer as a spreadsheet.
The difference from a chat window is that nothing has to be pasted in. A 40MB crawl export is a file path, not a context window problem.
Migration QA is the strongest use
A migration QA pass is a long list of small checks, each of which is easy and none of which is interesting. Request every old URL and record the response. Compare the canonical on staging against the canonical on live. Find the pages where the H1 changed. Check that the staging disallow did not ship. Every one of those is a script, and you do not have to write the script yourself.
It is honest about what it did, if you make it be
The risk is the same as with any assistant: it will tell you the check passed when it did not really run. Two habits fix most of that. Ask for the command it ran and the output, not a summary. And ask it to write the check as a file you keep, so the next time you run it you are running the same check rather than hoping for the same behaviour.
Where it is weaker
It has no opinion about your market and no access to your analytics unless you give it some. It will not tell you which of two strategies is right. It is also confident in the wrong direction sometimes, which matters more when the output is a recommendation than when the output is a list of URLs you can spot check.
Starting points that are worth the hour
Give it a folder of exports and ask it to find the difference between two crawls. Give it a sitemap and ask it to check every URL resolves. Give it a set of competitor pages and ask which entities they name that yours does not. All three are jobs you already do, and the value is in getting the boring 80% back.
Treat it as a very fast junior who never gets bored and never remembers what you told them last week. That framing gets you the benefit without the part where you stop checking.