What Fills a Context Window in Practice
Every piece of text you send to an AI takes up space in its context window. This includes your prompt, all the instructions you give it, and any background documents you provide. It also includes the AI's own responses to previous turns in the conversation.
Each character or word consumes part of this budget. Longer prompts mean less room for supporting documents. Longer conversations mean less room for new information. It's a continuous trade-off.
Why Trimming Navigation and Boilerplate Matters
Web pages often contain much more than just the main content. You'll find navigation menus, footers, sidebars, advertisements, and legal disclaimers. This extra text takes up valuable context window space without adding relevant information.
Removing these elements before sending content to an AI frees up room. It allows the AI to focus on the essential data you want it to process. This improves the chances of getting a relevant and accurate answer.
Signs an Answer Degraded Because of Overload
If an AI's answer seems confused, irrelevant, or incomplete, the context window might be full. You might see the AI ignoring key details from the middle or end of a long text. It might also hallucinate facts or make broad, generic statements.
Another sign is when the AI fails to follow specific instructions you gave it. This can happen if those instructions were buried in a large block of text. The AI may have prioritized other information or simply 'forgot' earlier details.
A Rule of Thumb for Batch Size
To avoid overload, aim to send only the most essential information. A good starting point is to limit each batch of content to what a human could reasonably read and understand in a few minutes. For most people, this is around 500 to 1,000 words of dense text.
If you have a very long document, summarize it first. Or, break it into smaller, thematic chunks. Send these chunks one at a time with clear questions. This approach helps the AI focus and gives you more control over the output.