Ask ChatGPT to draft a quote for a client, and you will get something generic: polite, plausible, and useless. Ask it again with your price list, your last three quotes, and a note on how you handle delivery, and you will get something you could nearly send.
Same tool. Same question. The difference is context: what it could see when it answered.
AI does not know your business
A general-purpose AI starts without any reliable knowledge of your business. It has read a huge amount of public and licensed material, but it knows nothing about your prices, your clients, your processes, or the way you like to write. It cannot know that a certain client always needs a longer lead time or that you never discount below a margin you keep in your head.
So when an answer feels vague, the tool is not being lazy. It is guessing because you gave it nothing to stand on.
Context is the skill
The teams that get real value from AI have learned one habit: give the tool what it needs before you ask. The relevant email thread. The old proposal that worked. The spreadsheet with this year's numbers. A line about who the reader is.
This is most of what people call prompting. Your prompt tells AI what you want. Context gives it the evidence it needs to do the job properly. A mediocre question with the right context beats a perfect question with none.
Context does not only arrive by pasting. Every tool has a handful of routes it can pick things up by, and they differ in one thing worth knowing: how often you have to do anything about them.
-
Always on
Standing instructions
What you tell it about your business once, in its settings. It reads that every time.
-
Per project
A project space
Files and instructions kept together for one piece of work, apart from your other chats.
-
This time
What you attach
The document, spreadsheet or picture you add to the message. Used for this chat only.
-
When needed
What it goes and finds
Your files, your past chats or the web, searched when your question calls for them.
The exact names change between tools, and not every tool has all four. The shape holds though: the top two rows are the same everywhere, and the work of getting good answers is mostly work on the bottom row.
The problem: context does not scale by hand
Pasting the right background into every chat works, but it is manual, repetitive, and depends on each person knowing where the good material lives. New starters do not. And the moment someone builds up good context in one tool, it is stranded there. Switch from ChatGPT to Copilot, and you start cold again.
Most businesses solve this with folders of boilerplate and a lot of copy and paste. It works, badly.
The Copilot advantage: your Microsoft 365 data, built in
This is why Copilot matters and why it feels different from a chat tool. The paid Microsoft 365 Copilot licence grounds it in your Microsoft 365 data: your email, calendar, documents, and Teams chats. In Outlook, it can use the current email thread without you copying it into the prompt.
It also respects your existing Microsoft 365 permissions: it can use only information you are entitled to access. That makes good permissions more important, not less, because Copilot cannot tell the difference between access someone genuinely needs and access they picked up accidentally years ago. If your SharePoint has been overshared for a decade, Copilot makes that content easier to find.
Its automatic advantage is strongest inside Microsoft 365. Knowledge that lives in people's heads, in other systems, or in another AI tool needs connectors, agents, and extra setup to reach it.
How the other tools tackle it
The other tools know context is the problem too, and each has its own answer. ChatGPT and Claude offer connectors to sources like Google Drive, SharePoint, and your email, so they can search your files when they answer. Both also let you set standing instructions about your business and group work into projects where you upload the documents that matter once.
These work, and in training, we show your team how to use them. But notice the shape of the fix: each platform still has its own connectors, its own instructions, its own projects, its own access model, and its own maintenance. The context you carefully fed into ChatGPT does not exist in Claude or Copilot.
The platforms are all improving at context. Each still creates its own version of it. The problem for a business is keeping that knowledge consistent, current, and under control across several AI tools at once.
Keeping context consistent
Until the platforms share context with each other, that work falls to you. Agree one set of standing instructions about your business, keep the source documents in one place, and give someone the job of keeping both current. In training, we show your team how to do this for the tools you use.
Written with Claude, then steered, cut and checked by me. I answer for it.
Explore Benchmark Lab’s free indices for personal AI use, organisational adoption and governance. All are in Research beta and can be used independently of training.