How to use Claude and Semrush for an SEO audit

A technical SEO audit can leave you with dozens of issues to investigate. Working out which ones deserve attention can take longer than running the scan.

Semrush’s guide to auditing with Claude describes a way to work through those findings in a conversation. Semrush’s Site Audit crawls the website. Claude retrieves the results through Semrush’s Model Context Protocol (MCP) connection and helps explain them. You can ask about affected pages, possible causes, and which repairs to tackle first.

I’d use that conversation to settle on a few specific tasks someone can complete and check.

Start with a recent, useful crawl

You need a Claude account, Semrush access that includes Site Audit and MCP, available API units, and a completed audit. Claude cannot start a new crawl or change the audit settings through this connection. Do that in Semrush before asking Claude to analyze the results.

Check the crawl’s scope as well as its date. Semrush lets you set a page limit and choose how it discovers URLs, including following website links, reading a sitemap, or using an uploaded URL list. An audit that never reached your important pages cannot tell you much about their condition. Semrush’s configuration guide explains those settings.

As of October 16, 2026, Semrush One Starter, One Pro+, SEO Classic Pro, and SEO Classic Guru include 50,000 MCP API units that refresh at subscription renewal. Semrush One Advanced and SEO Classic Business require an added API units package. Check your account against Semrush’s current access requirements before setting up the connection.

Connect Semrush to Claude

Find Semrush in Claude’s app or connector directory, choose Connect, sign in to Semrush, and approve the requested access. Semrush’s developer setup guide currently directs users through Customize → Connect your apps. Its blog walkthrough shows Settings → Connectors, so the route may depend on the interface you see.

In a new conversation, make sure Semrush is enabled and try:

Use Semrush Site Audit to find the latest completed audit for [domain]. Identify the project, the audit’s completion date, and how many pages were crawled. Tell me if any of that information is unavailable.

Check that Claude found the right website and audit before continuing. If you have several projects for the same domain, identify the one you want it to use.

Tell Claude what matters to the business

An audit identifies technical problems. You still need to explain which pages matter most to your work.

For a local service business, that might mean the service pages, location pages, and appointment form. For a shop, it might mean product categories and the products you are actively promoting. Give Claude those priorities rather than expecting it to infer them from the domain name.

Then ask for a small repair list:

Using the audit already retrieved, recommend three tasks for this week. Our priority pages are [URLs]. For each task, include the reported issue, affected page count, example URLs, a suggested owner, and a rough effort estimate. Label suspected causes as hypotheses and explain what someone should inspect to confirm them.

Suppose an appointment page returns an error while twenty older articles lack meta descriptions. I’d investigate the appointment page first because customers need it to contact the business. The page count alone would give an incomplete picture of what matters.

Claude’s effort estimates also need a person to check them. A title change might take a minute in one CMS and require a developer to change a shared template in another.

Check the evidence behind the recommendation

Ask to see the URLs behind an issue, especially when Claude describes a problem as widespread. Semrush’s Site Audit API documentation shows that issue reports use pagination, with a default of ten results per page. Ten URLs in a response may be a sample of a much larger group.

You can make that distinction explicit:

For the highest-priority issue, show the reported total and up to ten affected URLs. State whether this is a sample or the complete list. Explain what the audit confirms and what still needs checking on the live website.

Open several examples yourself. If the same problem appears across many pages, ask whoever maintains the site to inspect the shared template or navigation component. One underlying change may address the whole group, but that needs confirmation before anyone edits hundreds of pages individually.

Keep an eye on API usage

A conversation can trigger several data requests. Semrush’s API reference lists 100 units for the latest campaign summary, 1,000 for a page-detail request, and 10,000 for an individual audit snapshot. The history endpoint charges 10,000 units per snapshot returned.

At that rate, retrieving five historical snapshots would consume a 50,000-unit allowance before other requests. Start with the latest audit and investigate specific issues. Ask Claude to work from information already retrieved when possible, and check your remaining balance in Semrush. Instructions in a prompt do not guarantee how many requests the assistant will make.

Finish with a change you can verify

Once you choose a task, have Claude draft a short handoff: affected URLs, the observed problem, the proposed change, and the check that will show whether it worked.

After the repair, inspect the pages and run another audit in Semrush. Keep the crawl scope comparable so that a lower issue count reflects a repair rather than fewer pages being checked.

For me, the useful outcome would be three verified repairs and a clear explanation of what changed. That gives the next audit something concrete to measure.