Over the last year, I have been experimenting with different AI tools and automations for SEO work.
I have used Claude, Perplexity, ChatGPT, Gemini, and Grok. I also tried a few local LLMs. Overall, I spent the most time using Perplexity.
One thing I found useful was using Perplexity Pro inside the Comet browser. I could open a page, ask it to understand the intent and context of the URL, and then rewrite title tags and meta descriptions.
The output was usually basic and still required editing, but if a page had a poorly written title or meta description, it was a good starting point.
Using ChatGPT Projects
Later, I started using ChatGPT Projects and Claude Projects.
I uploaded project documentation and Markdown files so the AI could understand the context of the website and give more relevant answers. Having project-specific context available made the responses much more useful compared to starting from scratch every time.
Connecting Google Search Console to Claude Code
After that, I started experimenting with Claude Code using terminal tools like Windsurf (now Devin AI).
One of the first things I tried was connecting Google Search Console so I could directly ask questions about Search Console data.
Technically, it worked.
Claude Code could analyze the data and answer questions, but I still found myself preferring to work directly with exported data through BigQuery or Search Analytics for Sheets for most routine analysis.
I don’t fully trust AI-generated analysis unless the workflow is very clearly defined and the task is complex enough to benefit from automation and I have thought about all the edge cases beforehand.
Automating Weekly KPI Reporting
Another workflow I tried automating was weekly KPI reporting.
Every week I maintain reports for clients and stakeholders using data from:
- Google Search Console
- Bing Webmaster Tools
- GA4
- Microsoft Clarity
- Lead data from CRMs like Hubspot, Zoho, Salesforce for lead data analysis
- Ahrefs
- Server log data (The analysis was done on this data once a month but I still looked at patterns in server hits each week)
Updating all of that manually every week takes time.
To automate the process, I connected:
- Google Search Console API
- Bing Webmaster API
- GA4 API
- Ahrefs API
- Google Sheets API
Since I did not have admin access to every account, I used OAuth authentication wherever required.
Once everything was connected, I could simply specify a date range and Claude Code would pull the data and populate the reporting sheets automatically.
A few metrics still required manual work. For example, branded versus non-branded traffic was not directly available through some of the sources I was using.
Apart from those exceptions, most of the reporting process became automated.
Using SEO Machine for Content Audits
Another interesting project was using SEO Machine by Greg Hewitt.
Repository:
https://github.com/TheCraigHewitt/seomachine
I connected one of my personal WordPress website using Application Passwords, connected Google Search Console and Bing Webmaster data, and then connected Claude Code with Playwright MCP.
This allowed Claude Code to open a browser, inspect pages, analyze content, and review SERPs.
Using that setup, I audited content on my website, identified pages that needed improvement, and updated content where required.
I would not describe this as fully automated.
It still required a lot of oversight.
Claude Code could identify issues and provide recommendations, but I still had to review the output and decide what changes should actually be made.
The Reality of AI Content Automation
One thing I learned quickly is that AI automation sounds easier than it actually is.
When Claude Code updated content, I regularly ran into issues such as:
- Formatting problems
- Missing schema
- Poor content structure
- Overly long sentences
- Inconsistent outputs
A lot of time was spent refining prompts, adjusting workflows, testing changes, and fixing mistakes.
The SEO Machine repository provided a useful framework, but I modified it heavily for my own workflow.
Using that setup, I was able to update around 20–26 articles in a month.
Could I have updated more?
Probably.
But I also became more skeptical whenever I saw people claiming they had automated entire content teams and were updating hundreds of pages every month using AI.
My experience was that meaningful automation requires a lot of process refinement before it becomes reliable.
Automating Internal Linking
I also used Claude Code for internal linking.
First, I asked it to go into planning mode and build an internal linking structure across the site.
That required several rounds of refinement and approval.
Once I was satisfied with the plan, I asked it to implement the changes directly on the site.
Using Application Passwords, Claude Code was able to update internal links on both staging and live environments.
It also helped identify pages that were showing a “Crawled – Currently Not Indexed” status in Google Search Console.
For those pages, I reviewed the content, updated it, and improved internal links where necessary. This helped a lot with getting pages indexed.
Why I Prefer Claude Code
There are multiple ways to build these workflows.
You can use MCP servers, connect directly through APIs, or use other automation frameworks.
Personally, I prefer Claude Code in the terminal because it gives me more visibility into what is happening.
Before using any of these workflows, I spent a lot of time studying the SEO Machine repository and reviewing similar GitHub projects.
Most of the value came from understanding how the systems worked and then modifying them to fit my own requirements.
Early Results
After implementing some of these workflows across websites I manage, I noticed improvements in content quality, internal linking consistency, and overall visibility.
In one project, impressions started increasing a few months after content updates and interlinking improvements were rolled out. However, I would still consider this an ongoing experiment rather than proof that the workflow itself caused the growth.
For now, my conclusion is simple:
AI can automate parts of SEO workflows, but getting reliable results requires much more process design, testing, and oversight than most people realize.