Fire Automations is my weekly newsletter where I cut through the hype and teach you when automation is worth it, what to automate, and how to implement it in real businesses
Thanks for reading - and let me know your thoughts down below.
Your docs are probably lying to you.
Not on purpose. But between random Google Docs, dusty PDFs, meeting notes, Slack decisions, and “final_v7_really_final” files, your team’s knowledge is scattered and basically unsearchable. The result is the same three people getting pinged for answers all day, projects restarting from scratch, and everyone quietly rebuilding context in their heads.
I’ve been fixing this for myself by turning messy documents into a single searchable knowledge base, with lightweight automation so it stays fresh. It is not complicated, but it does require doing a few things in the right order. If you skip the boring parts (naming, chunking, permissions), the AI layer will confidently hallucinate the wrong answer and you will stop trusting it.
Here’s the 9-step approach that actually holds up in real work.
1. Pick one “source of truth” home base
Before you touch AI, decide where the knowledge will live. Not where it is today, where you want it to live going forward. For most teams I’ve worked with, it is either Google Drive, Notion, Confluence, or SharePoint. The tool matters less than the rule: one default place where docs belong, and everything else is either a shortcut to it or gets archived.
If you skip this, you end up building an AI search layer on top of chaos, which just gives you faster chaos.
2. Define what counts as “knowledge” (and what does not)
A searchable knowledge base dies when you throw everything into it. You want the repeatable stuff: SOPs, processes, client notes, policies, product decisions, onboarding docs, templates, and FAQs. You do not want 10 versions of the same brainstorm doc or every meeting transcript ever.
The filter I use is simple: will someone realistically search for this in 30 days?
If yes, it belongs.
If no, let it live in the project folder and move on.
3. Do a ruthless doc triage in 45 minutes
This is the part everyone avoids, so it never gets done. Put a 45-minute timer on your calendar and do nothing but classify documents into three buckets: keep, archive, delete. You do not need perfection, you need momentum. When I do this with clients, we usually eliminate 20 to 40 percent of “active” docs immediately, which makes everything downstream faster. You can always recover something later, but you cannot build trust in search when the base layer is full of duplicates.
4. Fix naming and metadata so search has something to latch onto
AI search is not magic. It needs signals. Clean names beat clever names. I like a format like: Team or Client, topic, status, date if relevant. Then add a tiny amount of metadata somewhere consistent: owner, last updated, and what “done” means for that doc. This is the unsexy step that makes the whole system feel like it reads your mind later. If you want the quickest win, rename the top 25 docs people ask about constantly. You will feel it immediately.
5. Standardize your “evergreen” doc templates
If every SOP looks different, your knowledge base will feel random. I keep a handful of templates that force consistency: SOP, meeting decision, project kickoff, client brief, troubleshooting guide. The goal is not bureaucracy, it is predictable structure so both humans and AI can skim. This is where tools like Notion databases or Confluence templates shine, but you can do it in Google Docs too. Consistency turns search results into answers instead of scavenger hunts.
6. Convert PDFs and weird formats into clean text
If you have important knowledge trapped in PDFs, screenshots, or slide decks, fix that early. I usually run PDFs through a converter so the text is selectable and searchable, then copy the cleaned content into the main system with a quick summary at the top. If you have scanned PDFs, you may need OCR. The key is to avoid the trap of “we ingested it” when what you really did was store a document the AI cannot reliably read.
I learned this the hard way after spending an hour debugging why answers were wrong, and it was because the PDF text layer was garbage.
7. Choose your search layer: native, AI Q&A, or both
You have three practical options, and most teams end up with a combo.
Native search (fast, cheap, often good enough)
AI Q&A on top of docs (better answers, more setup)
Hybrid (native search for finding, AI for summarizing)
If you are already in Microsoft 365, Copilot can work well for document Q&A if your permissions and file hygiene are solid. If your team lives in Google Workspace, Gemini for Workspace can be a decent starting point. For a more “knowledge base” feel, tools like Notion Q&A or Confluence AI can be great.
The honest truth: if your docs are messy, all of these will disappoint you.
If your docs are clean, even basic native search feels like a superpower.
8. Add guardrails so the AI does not embarrass you
This is where trust is won or lost. Do not let AI answer from “everything” with no constraints. Start with one scope, like Operations or Customer Support, and only include the docs that are actually curated. Require citations or at least “show me the source doc” behavior, so people can verify. Also, be explicit about what the system should do when it is unsure: say “I don’t know” and suggest where to look next. A confident wrong answer is worse than no answer, because it teaches your team to stop using the tool.
9. Automate upkeep so it does not rot in 30 days
This is the step that makes it stick. Without upkeep, your knowledge base becomes a museum.
I like three lightweight automations that do not require a technical team:
A weekly reminder to doc owners to review anything not updated in 90 days
A “new doc intake” checklist (name it, tag it, assign an owner)
A monthly cleanup pass where you archive duplicates and stale drafts
If you use Zapier or Make, you can automate reminders based on file activity. If you are lighter weight, a recurring calendar event and a simple dashboard list works. I personally prefer the boring version that people actually do over the fancy version that breaks when one login expires.
Final thoughts
If you want your docs to become searchable knowledge, do not start with an AI tool. Start by making your information legible: one home base, fewer duplicates, consistent structure, and clear ownership. Then add AI on top as an accelerator, not a crutch. This week, pick one team folder, do the 45-minute triage, and clean up the top 25 most referenced docs. Once those are solid, the “searchable knowledge base” part gets dramatically easier.