What Is a Context Window?

In one sentence

A context window is the maximum amount of text a model can hold in view at once — the prompt, retrieved documents, and its own response.

What Context Window means

It's measured in tokens, roughly three-quarters of a word each. Everything the model considers has to fit: your question, any retrieved passages, and the answer being generated.

Windows have grown substantially, but they remain finite and retrieval systems budget them carefully.

Why it matters for AI search

This is why only a handful of sources get cited. The system can't pass fifty retrieved documents into a limited window, so it ranks candidates and passes the top few.

It also rewards concise, self-contained passages: a tight answer costs fewer tokens than a rambling one, so more sources fit alongside it.

See it in action

Budgeting a context window across retrieved sources.

How a context budget gets spent

Token allocation
// click generate

The window, not the index, is why citations are scarce.

Concise self-contained passages cost fewer tokens, so more of them fit. That's a mechanical argument for tight writing rather than a stylistic one.

How to get it right

Writing for a limited window

  • Keep answers tight — a 60-word answer costs a fraction of a 400-word build-up
  • Front-load the substance so the useful part survives truncation
  • Make each section self-contained so it can be passed alone
  • Avoid repetition; it wastes budget without adding retrievable value
  • State facts explicitly rather than requiring inference across paragraphs

Common questions

Do bigger context windows mean more citations?

Not proportionally. Systems budget conservatively for speed and cost, so the practical number of cited sources stays small even as windows grow.

Does page length affect citation odds?

Indirectly. Retrieval selects passages rather than pages, so a long page is fine if its sections are tight and self-contained.

What is a token exactly?

Roughly three-quarters of an English word. A 750-word passage is about 1,000 tokens.

These come up alongside Context Window constantly.

Free 12-point check

Send Ron your site

Two fields. We’ll crawl your site as GPTBot, ClaudeBot and PerplexityBot, benchmark a sample of your category’s prompts, and send you what we find.

  • What each AI crawler actually receives from your site
  • Whether models resolve your brand as a real entity
  • A sample of category prompts and who gets cited
  • The three fixes we’d prioritise first

No commitment, no sales sequence. If we’re not a fit, we’ll say so.

We’ll pull your favicon so you know we found the right site.

We reply from a real address. No sequence, no newsletter.