Knowledge Management Trends for 2027: 5 Predictions We'll Check in December 2027
Read our five knowledge management predictions for 2027 on AI search, MCP, and self-hosting. Each has a test and a date, so you can see if we were right.
TL;DR
- This post makes five predictions about knowledge management in 2027, mostly about AI search, MCP (a way to connect AI tools to your wiki), and self-hosted wikis.
- Each prediction says what would prove it wrong.
- Each one has a check date in December 2027.
- The table below lists all five.
A prediction that can't be proven wrong isn't worth much. Take "AI will change how teams find answers." Nothing that happens next year could show it was wrong, so it tells you nothing.
The five predictions below are different. Each one says what would prove it wrong and gives a date when we'll check. We'll publish the scorecard in December 2027, whichever way it goes.
We picked the five from areas where Docmost has features: AI search, MCP, and self-hosting. You won't find market-size forecasts or borrowed survey numbers here. Where a claim needs data, the post says where the data comes from.
By the end you'll have five claims you can argue with, and a method for writing your own.
The Five Predictions and How We'll Test Them
|
# |
Prediction |
We're Wrong If |
Evidence We'll Use |
|
1 |
Most people who connect an AI tool to their wiki will untick Write, so the tool can read but not edit |
More than half of AI tools connected through the sign-in screen are approved with Write ticked |
What people approve when they connect in cloud workspaces, or a user poll |
|
2 |
AI search that ignores page permissions gets switched off |
Fewer than half of the teams that found AI search showing a page a user can't open turned AI search off |
User poll, plus a re-run of our AI assistant prompts (see below) |
|
3 |
Keyword search stays alongside AI search |
Keyword search falls below a quarter of searches in workspaces with AI search turned on |
Search usage split in cloud workspaces |
|
4 |
Self-hosted teams that add AI use a model they run themselves |
Most self-hosted teams using AI point at a cloud model |
Community poll |
|
5 |
Teams clean up old pages before connecting AI |
Fewer than half reviewed or archived pages before turning AI on |
User poll |
In the "We're Wrong If" column, amounts like "more than half" or "a quarter" are the points where we'll admit we got it wrong. We chose them ourselves before seeing any data. They didn't come from a study or a survey.
1. AI Tools Will Read Your Wiki Long Before They Edit It
Most people who connect an AI tool to their wiki will untick the Write box, so the tool can look things up but can't change anything. We expect that to still be true in December 2027.
MCP is an open standard that lets AI apps like Claude or ChatGPT connect to outside tools and data. A wiki can be one of those tools.
Why We Expect This
In Docmost, a workspace admin turns MCP on under Settings, then AI settings. After that, any member can connect their own AI tool. Most tools connect through a sign-in screen with two boxes, "Read" and "Write", and both are ticked by default. If you untick "Write", the tool can only read and search.
Also, a connected tool acts as the person who signed in, so it can't open pages that person can't open. Some tools can't use the sign-in screen and connect with an API key instead. A key can't be limited to reading. It can do anything your account can do through the API.
MCP is a paid edition feature, and the setup steps are the same on cloud and self-hosted. For the steps, see how to connect Docmost to Claude, ChatGPT, or Cursor with MCP.
Because Write is ticked by default, staying read-only takes a deliberate click. We think careful teams will make that click, since a tool that can change pages means someone has to check what it changed. Docmost's AI Chat has a similar setting. An admin can turn on Read-only mode, so AI Chat can draft content but can't save changes to pages. For any tool connected over MCP, you can remove access at any time under Settings, then API keys, then Authorized apps.
What Would Prove Us Wrong
If more than half of the AI tools connected through the sign-in screen are approved with Write ticked by December 2027, we're wrong. API key connections can't be limited, so we'll count them separately. We'll look at what people approve when they connect in cloud workspaces, and we'll ask users on self-hosted installs, where we can't see it.
2. AI Search That Ignores Page Permissions Will Get Switched Off
When a team finds that AI search shows a user a page the user can't open, we expect the team to turn AI search off instead of fixing it. By December 2027 we expect people to treat permissions in AI search as a given.
Why We Expect This
We think a search that shows someone a page they shouldn't see costs more trust than a search that misses a page. So we expect teams to switch AI search off first and sort out permissions after.
Docmost's AI search is called AI Answers. It only searches content you can open. It's a paid edition feature. A workspace admin turns it on under Settings, then AI settings, and the AI provider (OpenAI, Google Gemini, an OpenAI-compatible provider, or Ollama) is set up in the server's settings.
On the paid edition you can also restrict a single page, and a restricted page doesn't show up in search for people who can't open it. See wiki access control for large teams and how to search internal docs faster with Docmost.
What Would Prove Us Wrong
If a user poll shows that fewer than half of the teams that found AI search showing a page a user can't open turned AI search off, we're wrong. We'll also re-run the AI assistant prompts we describe below to see whether permission handling shows up in the answers. We treat that as a rough signal only, since AI assistants repeat what's written online, not what teams do.
3. Keyword Search Will Still Be in Your Wiki in December 2027
Teams will run AI search and keyword search side by side. People will ask AI search a question, and they'll use keyword search to find an exact error code, ticket ID, or policy number.
Why We Expect This
Asking a question and looking something up are two different jobs. A question may need details from several pages. A lookup needs the one page that has the exact thing you typed in. Keyword search still wins for exact matches like error codes, ticket IDs, and policy numbers (see AI search vs keyword search).
In Docmost, the two live in the same search window. Once an admin turns on AI Answers, a switch appears there. People turn it on for a question and leave it off when they're looking for an exact term.
What Would Prove Us Wrong
If keyword search falls below a quarter of all searches in workspaces that have AI Answers turned on, we're wrong. We'll check this against search usage in cloud workspaces.
4. Self-Hosted Teams That Add AI Will Run Their Own Model
Among self-hosted teams that turn on AI features, most will point them at a model they run themselves by December 2027.
This is the riskiest prediction on the list. Docmost only collects basic counts from self-hosted installs (the version, and the number of users, pages, spaces, and workspaces), and teams can switch even that off. So we can't see which AI model anyone uses, and we'll have to ask.
Why We Expect This
One reason teams self-host is so their data never leaves infrastructure they control. You also set your own rules for access controls, encryption, and retention instead of inheriting a vendor's. Hosting, backups, upgrades, and security patching also become your IT team's responsibility.
In Docmost, AI features can use OpenAI, Google Gemini, an OpenAI-compatible provider, or a local model through Ollama. Ollama is a tool that runs AI models on your own hardware. If you self-host Docmost and use a local model through Ollama, the whole AI setup stays on your own servers. Both AI Answers and AI Chat list Ollama as a supported provider. AI Answers also needs a PostgreSQL database with a database add-on called pgvector, so a team that self-hosts should check its database first.
We think teams that chose self-hosting for control will want the same control over their AI. If you're still deciding whether to self-host at all, see self-hosted wiki software.
What Would Prove Us Wrong
If a community poll shows that most self-hosted teams using AI point at a cloud model, we're wrong.
5. Teams Will Clean Up Old Pages Before They Connect AI
Teams will review and fix their old pages before they connect AI, because AI answers come straight from their pages.
Why We Expect This
Any mess in your wiki shows up in the answers. So clean up first, then connect AI. For the full process, see using AI for knowledge management.
A named owner and a review date on each page are a simple way to keep pages current. In Docmost, page verification lets you assign verifiers to a page and set a repeating or fixed review date. It's an Enterprise edition feature. Docmost doesn't have a built-in reminder for review dates, so someone has to check which pages are due, for example with a recurring calendar slot. If you're not on Enterprise, put a short block at the top of each page with the owner, a backup, the last review date, and the next review date (see assigning page owners and review dates).
What Would Prove Us Wrong
If fewer than half of the teams that turned on AI search or AI Chat reviewed or archived pages first, we're wrong. One poll covers this and predictions 2 and 4.
What We're Recording Now So December 2027 Has an Answer
A prediction with no baseline can't be scored. This is what we record by mid-December 2026, and where each number will come from.
|
Prediction |
What We Record |
Source |
|
1 |
What people approve when they connect an AI tool (Read, Write, or both), and how many connect with API keys |
Cloud workspace data, or a user poll |
|
2 |
Whether teams running AI search have seen it show a page a user can't open, and what they did about it |
User poll |
|
3 |
The split between AI searches and keyword searches in workspaces with AI search on |
Search usage in cloud workspaces |
|
4 |
Which model self-hosted teams point their AI features at |
Community poll |
|
5 |
Whether teams reviewed or archived pages before turning AI on |
User poll |
|
All |
How often AI assistants mention self-hosting, MCP, and permission handling when asked buyer-style wiki questions |
A fixed set of about 25 prompts across ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews, re-run around January 31, 2027, and again in December 2027 |
Ask an AI tool the same question twice and you can get two different answers. So the baseline shows how things looked on the day we ran it, not where things are heading.
Frequently Asked Questions
- What Are the Biggest Knowledge Management Trends for 2027?
We expect five. AI tools will read your wiki long before they edit it, and AI search that ignores page permissions will get switched off. Keyword search will stay alongside AI search. Self-hosted teams that add AI will run their own model, and teams will clean up old pages before they connect AI.
- How Will We Know If These Predictions Were Right?
Each prediction has a "wrong if" line and a source of evidence, and we'll check all five in December 2027. We'll publish the results in a scorecard, whether we were right or wrong. The baseline we're recording now is what we'll compare against.
- What Is a Knowledge Management Process?
A knowledge management process is the set of habits your team uses to capture, organize, and share what it knows. A knowledge management system is the software those habits run on, and a wiki can be that software. For more, see what a knowledge management system is.
- What Is RAG, and Why Does It Matter for Wiki Search?
RAG stands for retrieval-augmented generation. The "retrieval" part finds the relevant content in your wiki, and the "generation" part is the AI writing an answer from it. That's why the quality of your pages sets the quality of the answers. For more, see what RAG means for internal documentation.
- How Do You Keep Wiki Pages From Going Out of Date?
Give each page a named owner and a review date, then check the dates on a schedule you'll keep. Docmost's page verification lets you assign verifiers and set a repeating or fixed review date. It's an Enterprise feature and doesn't send reminders, so put the check in your team's calendar. For setup, see assigning page owners and review dates.
- What's the Difference Between AI Answers and AI Chat in a Wiki?
AI Answers lives in the search window. You ask a question in plain words and get an answer drawn from your pages. AI Chat is a back-and-forth conversation about your workspace, and it can find and summarize pages, answer questions, and create or edit pages. Both need the paid edition, and an admin turns each one on separately under Settings, then AI settings.
Check Back in December 2027
We'll publish the updates in December 2027, whether we're right or wrong. If we're wrong, you'll find out while there's still time to change plans.
If you want to try AI search, MCP, or self-hosting on your own wiki, you can get a free trial license or contact sales.
If you're comparing wikis first, you can compare Docmost side by side with other wikis, and if you're bringing pages over, see the import and export guide.