SEO used to eat up a huge chunk of my client work hours. Not the strategic thinking — that part is worth paying for — but the mechanical stuff. Auditing title tags across 80 pages. Generating meta descriptions one by one. Pulling together keyword gap reports and turning them into content briefs. It was skilled copy-paste work that didn't require much judgment but still took hours every week.
Over the past year I've automated most of that with Claude Code. Not all of it — there are still judgment calls only a human should make — but I've cut the time I spend on repeatable SEO tasks by around 70%. Here's exactly how I'm doing it.
What I'm Actually Automating
Before I get into the how, it's worth being clear about what "SEO automation" means in my workflow. I'm not talking about auto-generating blog posts and hoping Google won't notice. I'm talking about the audit and optimization layer — the work that happens before content gets published or after a site goes live and needs ongoing maintenance.
The three areas where Claude Code has made the biggest difference:
- Technical SEO audits — crawling a site's HTML, flagging missing or duplicate meta tags, identifying heading hierarchy issues, checking canonical tags
- Meta tag generation — writing title tags and meta descriptions at scale, tuned to a client's target keywords and character limits
- Content brief creation — turning a keyword list into structured briefs with suggested headings, word counts, and competitor angles
Each of these used to be a standalone project. Now they're scripts I run in a Claude Code session.
The Technical Audit Script
For a new client, my first move is always a technical audit. I used to do this in Screaming Frog, export a CSV, and then spend an hour interpreting it and writing up the findings. Now I do it differently.
I built a script that takes a sitemap URL, fetches each page, and extracts the key on-page signals: title tag, meta description, H1, canonical URL, robots meta, and Open Graph tags. Claude Code reads the raw HTML for each URL and outputs a structured JSON object. Then a second pass scores each page against a checklist and flags issues.
The prompt that drives the second pass looks something like this:
You are an SEO auditor. For each page object in this JSON array, check: 1. Title tag: present, under 60 chars, contains primary keyword 2. Meta description: present, 120-155 chars, has a call to action 3. H1: exactly one, not duplicating title tag 4. Canonical: present and self-referencing (or correctly pointing elsewhere) 5. OG tags: title and description present Return a JSON array with the same pages, adding an "issues" array to each. Flag any problem with a severity of "critical", "warning", or "info".
The output is a prioritized list of fixes. I review it, remove anything that's a false positive, and send it to the client as a formatted report. What used to take three hours now takes about 25 minutes, most of which is the review.
Generating Meta Tags at Scale
This is where I've saved the most time on ongoing client work. A lot of my clients have e-commerce or service sites with dozens or hundreds of pages that were built without much attention to meta tags. Writing good ones manually is tedious. Writing bad ones automatically is worse than doing nothing.
The approach that works: I give Claude Code the page's H1, the primary target keyword, the page type (product, category, service, blog post), and any brand guidelines around tone. It generates a title tag and meta description. I spot-check a sample, and if the quality is consistent, I run the full batch.
The key constraint I always include: title tags must stay under 60 characters and include the keyword in the first half. Meta descriptions must be 130–155 characters and end with a soft CTA. Claude Code follows this reliably — much better than asking a junior hire to do it manually.
For a Vancouver-based retailer I work with, this process generated 140 optimized meta tags in a single afternoon. We saw measurable improvements in click-through rate within six weeks.
Content Briefs from Keyword Lists
The third major use case is turning raw keyword research into content briefs. A client will hand me a spreadsheet of 30 keywords they want to rank for. Turning each one into a usable brief — with a suggested title, target word count, outline, competitor summary, and internal linking opportunities — used to take 20–30 minutes per keyword. That's 10+ hours for a 30-keyword list.
Now I run it through a Claude Code workflow that:
- Takes the keyword and search intent classification (informational, commercial, navigational)
- Suggests a primary title and 2–3 alternative titles
- Proposes a logical H2 structure based on what the search results suggest users want
- Notes related keywords to weave in naturally
- Flags any existing content on the client's site that could be internally linked
I still review every brief before it goes to a writer. But the review takes 5 minutes instead of the 25 minutes it would take to build the brief from scratch.
What Claude Code Can't Replace
I want to be direct about the limits here, because I've seen people over-automate SEO and get burned.
Claude Code is not a replacement for a good keyword strategy. It can help you execute on a strategy, but deciding which keywords to pursue, understanding the competitive dynamics in a specific local market, and knowing when a client's site isn't ready to compete for a term — that still requires human judgment and experience.
It's also not a substitute for link building, which remains stubbornly manual in any ethical form. And it won't fix Core Web Vitals or server-side rendering issues on its own — you still need a developer for that.
What it does well is handle the volume work: the auditing, the writing, the structuring of information that follows a clear pattern. If the task has rules you can articulate, Claude Code can usually follow them faster than a person.
Getting Started If You Want to Try This
If you're an SEO consultant or agency owner who wants to test this, here's where I'd start:
- Pick one repeatable deliverable — meta descriptions are the easiest starting point
- Write out the rules you use when you do it manually (character limits, keyword placement, tone)
- Build a small test batch of 10 pages and compare Claude Code output to what you'd write yourself
- Refine the prompt until quality is consistent, then scale
The biggest mistake I see is trying to automate everything at once. Start with one workflow, get it working well, and then expand. That's how I built this out over about six months — one script at a time.
If you want to see how this could work for your specific client base, I'm happy to walk through it on a call. And if you have questions about whether any of this applies to your situation, the FAQ page covers a lot of the common ones.
The tools are here. The question is just which part of your workflow you want to give back to yourself first.