AI Guides › Getting Started
What "Context Window" Means And Why It's The First Thing To Understand
By Nigel Guy · 2 min read
New users hit the same confusing moment eventually: a long conversation
that was going well suddenly seems to "forget" something from earlier, or a
huge document gets summarised strangely, and nothing in the interface
explains why. The explanation is a single concept most people never get
told plainly, and once you have it, half of the confusing moments stop
being confusing.
The rule: a context window is the amount of the current conversation the
model can actually see at once — once something falls outside it, it isn't
remembered within that chat, it's gone from view entirely.
The plain explanation
- Think of it as a moving window over the conversation, not a memory.
Everything inside the window — the messages, any documents you've
pasted or attached, its own previous replies — is what it's actually
working from for the next answer.
- It's measured in the amount of text, not time. A long, dense
conversation can fill it faster than a shorter one spread over the same
number of messages, and a large pasted document can take up a
significant chunk on its own.
- When it's full, older parts drop out of view first, not the newest
ones. That's why a fact you mentioned at the very start of a very long
conversation can stop being "known" without anything being deleted —
it's still in the chat history you can scroll to, just no longer inside
what the model is drawing on for its next answer.
- This is different from memory. Memory (where the product has it)
persists specific facts across separate conversations deliberately; the
context window is about what's visible within one conversation, and
isn't a substitute for memory or a project's standing context.
What this means practically
- Restate anything critical if a conversation gets very long. If a key
constraint from message two still needs to apply at message two hundred,
say it again rather than assuming it's still "seen."
- Split genuinely unrelated big tasks into separate conversations.
Cramming everything into one long thread eats into the same window a
large, important document might need.
- Don't assume a long conversation is a photographic record it's
drawing on in full — for a very long thread, treat recent turns as
what's actually informing the current answer.
What to skip
Skip trying to memorise an exact number for how large a context window is —
this changes across models and updates often enough that a specific figure
here would be wrong by the time you read it. The mechanism (a window, not
a memory) is durable; the exact size isn't.
Guardrails
- Exact context window sizes vary by model and change over time — check
current documentation if the precise number matters for your task, don't
rely on a remembered figure.
- A full context window isn't a sign of a malfunction — it's an expected
limit, and restating key facts is the normal, working fix, not a
workaround for a bug.
- This is separate from usage limits (how much you can send in a day or
month) — a context window is about one conversation's working scope, not
your account's overall allowance.
All 751 AI guides · JulieMango plans from £17/mo