How We Automated Our Entire Knowledge Base: AI Automation Blueprint for Your Business

If you run a SaaS business, an agency, or any company with a complex product, you know that maintaining an up-to-date, helpful Knowledge Base (KB) is a massive drain on resources. We used to struggle with keeping our documentation fresh, correctly formatted, and easily accessible to users. 

Recently, we completely overhauled our customer support and documentation workflow by integrating AI (Claude) directly with our Intercom setup. We want to share exactly what we did, why we did it, and how you can replicate this workflow to automate your own customer education processes.

At Serpstat, we use AI not only in our SEO workflows but also in customer education. As an AI-first SEO partner, we aim to reduce manual work across search marketing, support, and documentation processes.

We decided to move our user manual from our own admin panel to Intercom because it is much easier for the Intercom AI bot to get the data within its own ecosystem. This allows it to provide the best possible answers to our users directly in the on-site chat. Previously, managing the user manual inside our own admin panel was a tough challenge.

Here are the exact use cases and technical steps we implemented to automate documentation.

1. Data Extraction: Why We Chose Markdown Over JSON

To start, we needed to feed our existing documentation to the AI. Our developers exported the entire database in both Markdown and JSON formats.

How to apply this: We discovered that AI models process Markdown significantly better than JSON. With JSON, the AI has to spend processing power determining what each element represents, whereas Markdown's clear markup allows the AI to immediately understand the document's structure. If you are exporting data to train an AI on your company's documents, always opt for Markdown.

2. Structuring Content for Both Humans and LLMs

We fed the AI our old manuals, FAQs, and various use cases, instructing it to synthesize this data into comprehensive, unified articles for each specific tool. We designed the structure of these articles to perfectly mirror the Serpstat product sidebar, ensuring a consistent and logical user experience.

How to apply this: When generating articles, we created a highly specific template that benefits both users and AI agents: 

3. Mass Publishing via Direct Intercom API Integration

We used the Intercom MCP server — a pre-built connector that plugs Claude directly into the Intercom knowledge base without writing any custom scripts. Setup takes under 5 minutes: one config file, no code. Once connected, a single command like "Generate and publish all 11 articles for the Site Analysis section following the brand template" was enough — Claude handled writing, formatting, cross-linking, and publishing autonomously. We reviewed the final output.

How to apply this: Instead of writing scripts, we connected Claude directly to Intercom via API and gave it permission to modify the Knowledge Base. By doing this, we fully automated mass publishing. We simply gave the AI a command like, "Process the entire Site Analysis section," and it automatically generated and published all the nested pages in that category in a single request. All we had to do was review the final formatting.

4. Keeping Users in the Loop with In-App Contextual Help

One of the biggest conversion killers is forcing a user to leave the app to read documentation.

How to apply this: We mapped our newly generated articles so that they appear directly within the Intercom chat widget inside the Serpstat interface. If a user clicks on "Help", the exact article they need opens up right in the chat, allowing them to read and execute the task simultaneously without breaking their workflow.

Furthermore, we linked our help articles directly back to the actual tool reports — acting as internal landing pages — so users can jump straight into the action.

5. Brand-Safe Automated Localization

Translating technical documentation manually is tedious, but standard AI translation often ruins brand terminology.

How to apply this: We automated the translation of our English baseline articles into other languages simply by instructing the AI to process the existing Intercom database. To prevent the AI from generating awkward literal translations, we created a shared team "Skill" (a centralized reference document) containing our strict brand terminology. For example, the AI consults this document to know that it should never use direct translations for restricted phrases and knows exactly how to translate technical terms.

Our choice of the Claude Team plan significantly enhances this collaborative process by allowing us to share internal knowledge, specialized skills, and custom connectors across the entire organization. This centralized sharing ensures team-wide consistency; when a "Skill" is effective, it is immediately available for everyone to use, and if it requires improvement, the entire team has the visibility to identify and refine it.

By the way, we have a couple Serpstat skills for SEO reporting you might be interested in. Here is the Serpstat Skills GitHub repository and video tutorial:

6. The "Golden Reference" Self-Correction Loop

Prompt engineering can take hours, so we automated it with a lightweight alternative to formal eval frameworks.

How to apply this: We picked one of our existing articles that we considered the gold standard — perfect structure, tone, depth, and formatting — and embedded it directly into our Skill as a reference example. 

The Skill instructs Claude to compare every newly generated article against this reference along the same criteria: 

• structural completeness 
  
• tone consistency
• level of detail
• formatting patterns

If the new article deviates on any of these dimensions, Claude automatically rewrites it until it matches the quality of the reference. This gives you the benefits of a self-correction loop — consistent quality, no manual review of every output — without the overhead of building a numerical scoring rubric. The reference article sets the quality benchmark: it's a much cheaper way to encode "what good looks like" than writing explicit point-based metrics. 

7. Automated KB Updates from Task Trackers

Finally, a documentation base is only useful if it is up-to-date. Tracking every new feature release and updating the KB manually is exhausting.

How to apply this: We integrated our AI directly with our task tracker. Now, when our developers close a task or release a new tool, the AI automatically reads the task details, checks the relevant Intercom article, and generates a list of proposed edits. A human manager simply reviews and approves the list, and the AI automatically pushes the updates directly to the live Intercom article.

8. Content Consolidation

Intercom only supports custom subdomains (like help.serpstat.com), not subdirectories. Since we want the KB traffic to benefit our main domain for SEO purposes, we're planning to set up a reverse proxy so the final URL becomes serpstat.com/tutorial/ — the content stays on Intercom, but from search engines' perspective it lives on our main domain. This is a DevOps task we've already scoped with the team. 

How to apply this: We implemented 301 redirects for every old page, merging fragmented content and bot signals into a streamlined collection of high-authority, comprehensive pages. After cleaning server cash, we checked the ready-made redirects using Audit by Serpstat.

Conclusion

AI automation is no longer limited to customer support chatbots. With the right setup, it can become the foundation of an AI-powered knowledge base that stays structured, localized, and up to date with less manual work.

By connecting AI to your knowledge base management system, defining clear brand guidelines, and using a reliable reference article as a quality benchmark, you can turn scattered product information into consistent AI documentation. Add CMS or Intercom integration, task tracker monitoring, and automated review workflows — and your team gets a practical alternative to traditional automated documentation software.

The result is not just faster content production. It is a support system where users find answers in context, managers spend less time on repetitive updates, and your documentation keeps pace with product changes.