How can my team find answers in our company documents without asking me?
Your team can find answers without asking you once the files you already have are organized so one file governs each topic and old versions are marked instead of deleted. Then the AI they already use can answer with the file it came from, and when two files disagree or the answer isn’t there, it says so and hands the question to a person.
Plenty of your team already asks AI first. In an October 2026 survey of 669 US workers who use AI regularly, Clutch found that 45% go to AI first with a work question, ahead of coworkers at 16% and managers at 9%.
The same survey found 67% still trust a coworker over AI for knowledge specific to their company. The AI doesn’t have your company’s knowledge, so the question still ends up with you.
I’m Ian Titus and this is the work I do for companies of 10 to 150 people in Franklin, Tennessee and around Nashville. If you’d rather start with your own situation, the free assessment takes about 15 minutes and I’ll reply in writing.
What it looks like when it works
It looks like a question going to a folder instead of to you, and an answer that names the file it came from. I’ve built that three times where you can check it yourself, on my own operation, on a fictional property company’s drive, and in the agent on this site.
I ran the method on my own operation before I sold it to anyone. That was 3.7GB of records and 804 files moved with nothing deleted, and every move logged so I can walk it backward. It read everything first, did a dry run, then moved.
Then I built a property drive seeded with the problems real drives have. It had nine leases, a rent roll three months out of date, the same lease in three places under three names, and an HVAC clause its own amendment contradicts. It’s a fictional company, built so I could publish the method, and the case study says so.
Against twelve questions I wrote by hand before the system saw them, it went 12 for 12. The two that count are the two it refused to answer because the files couldn’t settle them, and when asked who pays for HVAC at one property it quoted both clauses and named who has to decide.
The third one you can test right now. The agent on this site answers from these pages and names the page each answer came from. When the answer isn’t on any page, it refuses. Before anyone got the link it answered 17 of 17 test questions the way the key expected and held up across 4 rounds of someone trying to talk it out of a refusal.
How to set it up so your team stops asking you
Find the questions that come to you, find where each answer lives, decide which file governs when two disagree, and write that ruling where the AI reads first. Then point the AI you already pay for at it and test it against answers you know before your team relies on it. It’s six steps and the first one takes about a week.
The time is worth getting back. Atlassian’s own State of Teams 2025 survey of 12,000 knowledge workers found teams waste 25% of their time just searching for answers.
1. Find the questions that come to you
Keep a tally for a week of every question that lands on you. Then ask the people doing the work three things. What do they double-check by hand because they don’t trust it, what do they know that isn’t written down anywhere, and what goes wrong when they’re out?
Each answer points at a step that only works because somebody remembers it. Start with the files behind those questions, the ones in use now. The archive stays where it is and can come in later.
2. Find where each answer lives
For each question, find where the answer actually is. It might be a file on the shared drive, an email thread, a spreadsheet one person keeps, or nowhere but someone’s head.
Write down every place two files give different answers. That list is usually short and it’s where the real pain sits.
3. Decide which file governs, and mark the rest
For every pair that disagrees, write down which one counts and why. A signed lease beats an unsigned draft and a current lease beats a rent roll that’s out of date.
Then put the state in the filename, so a spreadsheet carries its date and the word STALE instead of being called current and being wrong for three months. Superseded copies go into a folder marked so nothing cites them. Nothing gets deleted, because deleting is how you lose the one copy that turns out to matter.
4. Write the index the AI reads first
At the top of the folder goes one plain document that says what’s where and which file governs when two disagree. It’s written for a person first, so anyone on your team can read it with no AI involved.
That index is where the judgment lives. Because the rulings are in a file, the AI inherits them and the answer doesn’t change with how someone words the question.
5. Point the AI you already pay for at it
Use whichever one your team already runs, whether that’s Microsoft Copilot, ChatGPT, Claude or something that ships next year, and hold it to one rule. If there’s no source, you don’t get an answer, and every answer names the file it came from.
Because the structure lives in your folders, you can change AI tools later and the same questions come back with the same answers.
6. Test it, then run it beside you for two weeks
Write ten or twelve questions you already know the answers to, including two or three it should refuse because the files can’t support an answer. Check each answer against the file it cites, and check that it refuses the ones it should.
Then run it beside you for two weeks. Questions still come to you, you answer them, and you compare your answer to the system’s. When two weeks go by without a difference that matters, point your team at it.
If you’d like someone to do steps one and two with you, that’s the audit. It’s about an hour of your time, in person in Franklin or by video, $1,250, and the operating map is in your inbox within 48 hours. The first step is the free assessment.
My employees keep asking me the same questions. How do I stop being the bottleneck?
You stop being the bottleneck when the answers people bring to you live somewhere they trust as much as they trust you. Most of those answers already exist in a file or an email. What’s usually missing is the ruling on which version is right, and that ruling often lives only with you.
It adds up. An APQC survey of 982 knowledge workers back in 2021 found they spent 2.8 hours a week looking for or requesting information they needed. In a small company a good share of those requests go to the same one or two people.
The usual advice is to write your procedures down, and Waybook’s guide to getting out of the bottleneck starts the right way, by counting the questions you get before writing anything. Where I’d go further is the files you already have. Most of the answers are written somewhere, just in three versions with nobody saying which one counts.
The goal isn’t to take you out of decisions, it’s to take you out of lookups. A question about what the warranty covers should go to a file. A question about whether to honor it this time should still come to you.
Why AI gives wrong answers from a messy shared drive
AI gives wrong answers from a messy drive because it can’t tell which of two files is current, and it answers anyway. A rent roll named “current” that’s been wrong for three months reads as current to a new employee and to a model, and nothing in the answer warns you.
AI speeds up whatever is already there, including the broken parts. The person in accounting who knows which vendors to double-check is holding a step together with memory, and an AI reading the same folder doesn’t have that memory. It runs the broken version faster.
A messy drive also tends to overshare. Microsoft’s own guidance says Copilot grounds its answers in the data users already have permission to access, and the first step in its deployment blueprint is to remediate oversharing. On a messy drive, files are often shared wider than anyone meant, and the AI will use them.
On the property drive I built, the same lease sat in three places and an amendment contradicted the original. Pointed at that folder as it was, any AI would have answered with confidence from whichever copy it found first. Organized, it refused the question nobody had settled and named who had to settle it.
How new hires get up to speed when nothing is written down
New hires get up to speed faster when the answers they’d otherwise have to ask for are in files they can search and trust, and when what still lives only in someone’s head gets written down while that person is still here. The first part is organizing. The second part has to be asked for.
Owners describe the same pattern over and over. Everything runs through them, new people take months to get useful, and the reason is that nothing is written down where a new person would look.
The audit traces how the work actually happens, which is almost never how the written process says it does. Part of that is finding what your experienced people know that was never written down and deciding what should become a file and what should stay a judgment call.
On the property drive I built, the lease renewal dates lived in one person’s head, which means a renewal option lapses the week that person takes vacation. Once they were in the files, the folder could surface them every morning without anyone asking.
Do it while they’re here. You can document what somebody knows before they leave, and you can’t interview them six months after.
How to get it out of someone’s head, your own included, is the next note in this series.
Can AI search our company documents and answer questions?
Yes. Microsoft Copilot, ChatGPT, Claude and Gemini can all read company files and answer questions about them, and so can tools built for it. Whether the answers are right depends on the files underneath, and whether you can check them depends on the answer citing the file it used.
Your people are already doing a version of this. A 2026 PagerDuty survey of 1,250 office workers at companies with $500 million or more in revenue found 88% had shared work information with public AI tools. That’s big-company data, but I talk to people every day who do their jobs on AI accounts on their own phones.
So for a small company the real question is whether the AI is reading something organized enough to answer from, with a source you can open.
Should a small business use a wiki, an AI search tool, or organize what it has?
For most companies of 10 to 150 people, organize the files you already have first, then let whichever AI you already pay for read them. A wiki is the right call when the knowledge isn’t written anywhere yet. An AI search tool is worth it once the files underneath can be trusted.
| Write a wiki or procedures | Turn on an AI search tool | Organize what you have, then let any AI read it | |
|---|---|---|---|
| What it asks of you | Someone writes every page and keeps it current | Connect your apps and trust the content underneath | About an hour for the audit, then decisions on which file governs |
| What happens to messy files | They stay messy, the wiki starts fresh beside them | The tool reads them as they are | Marked by state, superseded copies kept but uncitable, nothing deleted |
| When two files disagree | Whoever wrote the page decided | Depends on the tool, and Slack’s own guide says effective AI search requires managed knowledge | The index records which governs, and if nothing settles it the answer says so |
| How you check an answer | Read the page | Depends on the tool | Every answer names its file |
| Cost, as published in November 2026 | Waybook lists $99 a month for up to 20 members, billed annually | Varies by tool and seats. Layer3 Labs publishes $3,000 to $8,000 for a pilot and $200 to $600 a month to run one | $1,250 for the audit, then the build at one fixed price from the map |
| What you’re tied to | The wiki’s platform | The tool’s platform | Your own folders, with any AI reading them |
I’m biased toward the third column because it’s what I do. The honest tradeoff is that it’s slower to start than switching on a tool, and the first weeks are decisions about your files instead of software.
The columns aren’t exclusive either. If you’re a Microsoft company, Copilot is a fine reader for an organized SharePoint, and a wiki is a good home for the procedures nobody has written yet.
Who this works for, and who doesn’t need it yet
This works for companies of 10 to 150 people that run on documents and deadlines, where the answers live with a few experienced people and two files often disagree about which version is current. Owner-run trades and home services, professional services firms, and property and real estate companies are the usual fit.
If you’re a one-person shop and all the knowledge is yours, you don’t need this yet. Come back when you’re hiring.
If you’re closer to five people and want to try the smallest version yourself, the System Starter is a free folder template you can install on your own. It was built for a coach’s client files, but the shape is the same one this note describes, with your knowledge in files you own and a record you can walk backward.
If you’re well past 150 people with your own AI team, you probably want a firm with a bench, because I’m one person and there’s no bench behind me.
Questions owners ask
How long does it take to set this up?
The audit takes about an hour of your time, in person in Franklin or by video, and the operating map is in your inbox within 48 hours. When your information is reachable, the first working system is live inside 30 days after that.
When it's hard to reach, because it lives in people's heads or in software that doesn't export cleanly, it takes longer and the audit tells you which before you commit to anything. Add the two weeks of running it beside you, and from the audit to your team asking the system instead of you is usually inside two months when the information is reachable. I'd rather tell you the longer number up front than have you find it out halfway through.
What does it cost?
The audit is $1,250, fixed. The build is quoted at one fixed price from what the audit's map finds, so you see the number before anything starts and it doesn't run on an hourly meter. If you go ahead within 14 days of getting the map, the full $1,250 credits toward the build.
There's no subscription of mine and no seat count. The structure lives in your folders and the AI that reads it is whichever one you already pay for, so the running cost is mostly what you're already spending. The cost worth weighing it against is the time your experienced people spend answering questions a file could answer.
Do we have to move our files or switch software?
No. The audit touches nothing, and you keep the map whether or not you hire me again. The build works in the folders and systems you already have, whether that's SharePoint, Google Drive, Dropbox or a server in the closet.
When files do move, the method reads everything first, does a dry run, then moves, and every move is logged so it can be walked backward. On my own operation that was 3.7GB of records and 804 files with nothing deleted. Your team keeps working the whole time, because nothing stops while it happens.
Do we have to clean up years of old files first?
No, and waiting for that is how this never starts. Begin with the files your team actually uses now, the ones behind the questions that come to you every week. The archive stays where it is, marked for what it is, and comes in later one piece at a time when a question needs it.
The mess is there for a reason. It's years of exceptions, amendments and workarounds that built up while people were busy doing their jobs, and a lot of it still matters. Nothing gets deleted, because deleting is how you lose the one copy that turns out to matter.
What happens when two documents disagree?
The ruling gets written down in the index the AI reads first. A signed lease beats an unsigned draft, a current lease beats a rent roll that's out of date, and an amendment sits next to its lease instead of quietly replacing it. Anything reading the folder inherits those rulings, so the answer doesn't depend on how someone words the question.
When nothing in the index settles it, the system says so. On the property drive I built to show the method, asked who pays for HVAC at one property, it quoted both clauses, explained why neither clearly governs and named who has to decide. A question nobody has decided yet should go to a person, and that's where it went.
Will the AI make things up?
It can, which is why the rule is that if there's no source, you don't get an answer, and why every build gets tested before your team relies on it. I write questions with known answers by hand before the system sees them, including the ones it should refuse, and score whether it answered right, cited right and refused when it should.
On the property drive it went 12 for 12, and the two that count are the two refusals. That doesn't make it perfect. What it does is make every answer checkable, because each one names the file it came from and anyone can open that file and look.
Is it safe to let AI read our company files?
The files stay in your folders. When the system reads one to answer a question, it goes through a business account whose published terms say your data isn't used to train the model, and you choose which AI company that is.
The bigger leak in most companies is people pasting company details into their personal AI accounts, where conversations can be used for training unless the person has turned that setting off. In a 2026 PagerDuty survey of office workers at large companies, 88% had shared work information with public AI tools. Giving your team a better place to ask is part of the fix.
Does this work with SharePoint, Google Drive or Dropbox?
Yes. If it holds files and folders, it's the same work, whether that's SharePoint, a shared Google Drive, Dropbox, OneDrive or a network drive. The method is about what's in the folders and which file governs, so the brand of storage doesn't change it.
The AI reading it can be Microsoft Copilot, ChatGPT, Claude or whatever ships next. The judgment lives in the files, so the same questions come back with the same answers when you change tools, and you're never locked into one company's assistant or into me.
What if the answer only lives in someone's head?
Then it has to be asked for, and the audit is where that starts. I trace how the work actually happens, which is almost never how the written process says it does, and part of that is finding what your experienced people know that was never written down.
Some of it should be captured as a short file in the right folder. Some of it is judgment that should stay with a person, and the system should route those questions to them instead of guessing. The timing matters, because you can document what somebody knows before they leave and you can't interview them six months after.
Do I need an IT person to keep it running?
No. Everything is plain files a person can open and read with no AI in the room, and the index is a document, not a database. Someone on your team who already keeps the shared drive in order can keep it current.
There's ongoing work if you want it, keeping the system tended as the business changes, because a structure nobody maintains grows back into a mess. The test of that work is that it shrinks over time, because the goal is a system your team runs without needing me.
Sources
- Clutch, "Workers ask AI the questions they won't ask coworkers," survey of 669 US workers, October 2026
- Atlassian, State of Teams 2025
- APQC, survey of 982 knowledge workers, November 2021
- Waybook, "How to stop being the bottleneck in your business"
- Microsoft Learn, data security guidance for Microsoft 365 Copilot (updated November 13, 2026)
- PagerDuty and Wakefield Research, shadow AI survey of 1,250 office professionals, June 2026
- Slack, "Find company knowledge faster"
- Layer3 Labs, "AI knowledge search for small business" (updated August 26, 2026)
Start with the free assessment
The fastest way to find out whether any of this fits is the free assessment. It takes about 15 minutes, in writing or on a call, and I’ll tell you plainly if the audit isn’t worth it for you.