My Experience With It
About a year ago, I was looking for simple AI use cases to boost productivity in my product team.
My biggest challenge?
I was constantly getting messages, emails, and questions on calls about details of my product that were already published on our internal product site.
And as my product grew in usage across the globe, I had users reaching out from 3 different time zones.
When I logged in, backlog of messages.
During the day, more questions.
Turning off my laptop. Ping.
I'm a strong believer that internal product managers should be close to their users.
But this was different.
Internal users don't have time to read documentation online. They want a straight answer from the source. Fast.
And we hadn't solved that yet.
Because as an integrations product, our documentation wasn't short. Integrations are all about those technical details.
So I found an AI chatbot solution that could access all of my existing documentation and, with some instructions, find the right answer for the user much faster than they could themselves.
I loved it. My product team loved it.
And the users were happier, because they became more self-sufficient in using our product.
What do they do & how they can help you with your documentation
Think of a documentation AI agent as a smart search layer on top of what you already have.
You don't need to write anything new.
It reads your existing content and answers questions based on it.
Here's where they can help:
→ User-facing documentation → general product info, pricing, access, setup, troubleshooting
→ Internal technical docs → architecture decisions, API specs, runbooks for your product team and ops
→ Onboarding materials → helping new team members or users find what they need without asking five people
The agent doesn't replace your docs.
It makes them actually usable.
Which Tool Should You Use?
It depends on where your documentation lives.
Here are the 5 most common enterprise setups:
1. SharePoint → Microsoft Copilot / Copilot Studio
You can create agents in Copilot Studio, but to publish to SharePoint you need M365 Copilot (~$30/user/month) or pay-as-you-go billing through Azure.
Most enterprises already have M365, so this is often the path of least resistance.
2. Confluence → Atlassian Intelligence / Rovo Agents
Available on Confluence Cloud Premium and Enterprise.
Rovo agents search across Confluence, Jira, and connected tools. If your org runs on the Atlassian stack, this is your best bet.
3. Notion → Notion AI (built-in since Notion 3.0)
AI is bundled into the Business plan ($20/user/month) and Enterprise.
No longer sold as a separate add-on. Agents can work across pages and databases.
4. Google Drive → Gemini + NotebookLM
NotebookLM is free for basic use.
The Plus version comes with Google Workspace Standard ($14/user/month). Great if your org already lives in the Google ecosystem.
5. Any platform → Custom GPT or Claude Project
Upload your docs to a ChatGPT Custom GPT or a Claude Project (both ~$20/month).
Works with any documentation. Most flexible, lowest barrier to entry. But requires manual uploads and updates.
One very important note: Before you commit, check with your IT team on licensing, security policies, and data handling. Enterprise environments have rules.
Let's Build One Together
I went with option 5. A Custom GPT using my own Notion templates as the knowledge base.
Why? It's the fastest way to show you the concept. And you can try it today.
What my agent does:
It's a documentation assistant for my free product management templates.
I have 11 templates covering things like your first 90 days on a new product, launch checklists, MVP ideation, testing for PMs, internal go-to-market, database bottlenecks, ROI calculations, and more.
Instead of messaging me with "Which template should I use?" or "How does the launch checklist work?" you can ask the agent.
It knows what's available, what each template covers, and can recommend the right one based on your situation.
Here's how I built it:
Step 1: Gather your docs (~10 min)
→ Export your documentation as PDFs or text files
→ Pick a focused set. Don't dump everything in at once
Step 2: Create a Custom GPT (~5 min)
→ Go to ChatGPT → Explore GPTs → Create
→ Give it a name and description
Step 3: Upload files and write instructions (~10 min)
→ Upload the docs you exported
→ Tell the GPT what it is, who it's for, and how it should answer
→ Example: "You are a documentation assistant for [product]. Answer questions based only on the uploaded files. If you don't know, say so."
Step 4: Test it (~5 min)
→ Ask it real questions your users usually send you
→ See where it gets it right and where it stumbles
→ Adjust instructions or add more docs as needed
That's it. About 30 minutes.
If your company uses SharePoint, the same concept applies. You'd use Copilot Studio instead. Same idea, different tool.
Here's the template instruction I used (steal it for yours):
You are a documentation assistant for [your product or resource name]. Your job is to help users find the right information quickly.
Rules:
→ Only answer based on the uploaded files. Never make things up.
→ If you don't know the answer, say so and suggest where the user can find more help.
→ Keep answers short and practical. No long essays.
→ If a user asks which [template/feature/tool] to use, ask 1-2 clarifying questions about their situation before recommending.
→ Be friendly, direct, and helpful.
You can copy this, swap in your product details, and you're good to go.
If your company uses SharePoint, the same concept applies. You'd use Copilot Studio instead. Same idea, different tool.
Want to see my documentation assistant in action?
🔗 Try it here: [LINK TO MY EXAMPLE GPT]
What I Learned
- You can build it very fast. The setup is not the hard part.
- Garbage in, garbage out. If your docs are outdated or confusing, your agent will be too.
- People have low trust in bots. Lots of bad experiences with support chatbots. So treat your AI agent like a feature. Iteratively improve it based on what fails.
- Upfront effort is small. Maintenance effort is real. Even though setup takes 30 minutes, keeping it useful is ongoing. Review what questions users ask, where it gets stuck, and which docs need updating.
- Measure the impact. Run a rough before-and-after comparison. Include it in your CSAT score if you can. See how it improves user experience and your own productivity.
