What is a context window and why does it matter for your AI use?
When AI tools advertise "200K context" or "1M tokens", it sounds like spec-sheet noise. It is the single most important number for an SME owner doing real work.
When you read about a new AI model, the headline figures are usually the same: model name, benchmark scores, and "200K context window" or "1M tokens". Most business owners skip the last one. They should not. The context window is the most practical number on the page.
This article explains what a context window is, why it matters for real work, and how to use it well.
What is a context window, really?
A context window is the amount of text an AI can hold in its working memory at once. Everything you say to it. Everything it says back. Everything you paste in. All counted together.
AI measures this in "tokens". A token is roughly three-quarters of a word in English. So 100K tokens is around 75,000 words, or roughly a short novel.
When the conversation goes over the window, the AI starts to forget what was said earlier. Not gracefully. The earliest part of the conversation drops off the start, like an old hard drive overwriting itself.
What the numbers actually mean
Here is a rough translation for the sizes you will see in 2026.
- 8K to 32K tokens: small. Roughly 6,000 to 24,000 words. Fine for everyday emails, drafting and short documents. Older free-tier models often sit here.
- 100K to 200K tokens: large. Roughly 75,000 to 150,000 words. You can paste a full set of accounts, a long contract, a year of customer emails. This is where most paid SME work happens today.
- 1M tokens and up: very large. Roughly 750,000 words. You can paste an entire book, a full codebase, or hundreds of documents at once. Available on Claude, Gemini and a few others on premium tiers.
Why this matters for an SME
Three real-world implications for a small business.
1. Long documents. If you are an accountant who wants AI to summarise a 50-page set of accounts, you need a context window large enough to hold them. A 32K window probably will not. A 200K window almost certainly will.
2. Multi-document work. When you are paste in three contracts to compare, or four years of accounts, the totals add up fast. Plan for the combined size, not the largest single document.
3. Long conversations. If you have spent two hours in a single chat working through a complex problem, AI may have already forgotten the start. The signs: it asks you something you have already told it, or repeats itself. Time to summarise the conversation so far and start a fresh chat.
What to do when you hit the limit
Three practical patterns.
Summarise and restart. When a conversation gets long, ask AI to summarise everything important, paste that summary into a new chat, and continue. Most paid tiers handle this gracefully ("here is a summary of our work so far - continue from this point").
Chunk the work. If you have a 200-page document and a 100K window, split the document into halves. Process each half separately. Combine the outputs.
Use a bigger window. Sometimes the answer is just a different tool. Claude and Gemini both offer larger windows than ChatGPT on similar paid tiers. The tool comparison post has the practical detail.
Getting the most out of your window
A practical rule for everyday work: do not just paste in everything you have. Paste in what is relevant. Long irrelevant context wastes the window and sometimes confuses the AI.
If you are asking AI to draft a quote based on three previous quotes, paste in those three. Not your whole quote folder. Less context, better focus, faster reply.
And if you are doing the same kind of task often - drafting quotes, replying to a particular kind of enquiry, summarising a particular type of report - save the prompt with the right context structure as a template. The 30-day plan post covers this pattern.
The honest summary
Context windows are the AI equivalent of "how many tabs you can keep open in your head". They are bigger than they were a year ago and bigger again than the year before. For most North Wales SME work today, the 100K to 200K range is plenty. The skill is learning when you have run out of room and what to do about it.
If you are running into limits and want to think through tooling for your specific work, that is what a free discovery call is for.