
When an answer engine writes about your category, it doesn’t read the web. It reads a short packet of passages that its search step placed in the model’s context window. Whatever misses that packet can’t shape the answer or earn a citation. That makes the context window the last gate between a page that ranks and a page that gets cited.
Anthropic’s documentation calls the context window the model’s "working memory". It holds all the text a model can reference while it writes a response, including the response itself. Size is counted in tokens. OpenAI estimates one English token at about four characters, or three quarters of a word.
Google uses a similar picture and calls the window short-term memory. Its long-context guide says many Gemini models take 1 million tokens or more. By Google’s count, that’s about 50,000 lines of code or eight average-length English novels.
"Model context window" means the same thing: the limit set for one specific model. People confuse it with training data. Training data shaped the model’s weights before release and stops at a cutoff date. The window is filled fresh for every request and cleared afterward.
Chat apps that seem to remember earlier turns put those turns back into the window, or store summaries outside it. Long chats eventually push old turns out.
| Part | What it holds | Who controls it |
|---|---|---|
| System instructions | Rules for tone, safety and citations | The engine |
| Conversation history | Earlier prompts and replies | The user |
| Retrieved passages | Chunks from search results for the prompt | Search and ranking systems |
| Tool results | Output from searches, code or other tools | The engine |
| The answer | Tokens the model writes | The model |
Only the retrieved passages come from publishers. That slice shares the window with everything else.
Engines rarely paste whole pages. They select passages, so your best paragraph competes with other sites’ paragraphs. Content retrieval covers how passages get picked. RAG covers the architecture around it.
Not automatically. The 2023 paper "Lost in the Middle" found performance is often highest when the relevant facts sit at the beginning or end of the input. It degrades when the model has to use facts buried in the middle, even in models built for long inputs.
Anthropic’s docs name the wider effect context rot: accuracy and recall fall as the token count grows. Google adds that its near-perfect recall results come from finding one fact. When a model must find several, it doesn’t reach the same accuracy. More room helps, but what goes into the window still decides the answer.
These habits raise a page’s retrievability. Our guide to optimizing content for answer engines works through them with examples.
Large windows haven’t removed selection. Running a model on a million tokens costs more and takes longer, which gives answer engines a reason to keep packets small. Google’s guide notes that the input token cost is paid again for every query. The AI answers we track cite about 7.2 links each, across 103,427 answers. Source: Ansvisor tracking data, September 15 to October 14, 2026. Every passage you publish competes for one of those few slots in an LLM answer.
Ansvisor’s Site Audit checks pages against 47 weighted AEO and GEO signals. Its content checks cover direct answers, answer-first structure, readability and entities. Those are the traits that let a passage hold up once it’s lifted out of the page. Many fixes come with copy you can paste.
Each prompt you win brings more mentions, more citations and more buyers. Ansvisor keeps finding the next one.
A context window is the amount of text a language model can consider at one time, counted in tokens. Everything in the request uses it: instructions, chat history, retrieved documents and the model’s own reply. Text outside the window is invisible to the model for that answer.
Many Gemini models accept 1 million tokens or more, and several current Claude models also have a 1 million-token window. Google says 1 million tokens is about eight average-length English novels. Some models still work with windows of 200,000 tokens.
No. Research shows models use information at the start and end of a long input better than information in the middle. Anthropic calls the broader drop in recall as inputs grow “context rot”. Picking the right passages still matters more than fitting in more text.
In English, one token is about four characters, or roughly three quarters of a word, according to OpenAI. So 100,000 tokens is around 75,000 words. Exact counts depend on the model’s tokenizer and the language.
Indirectly. An answer engine retrieves many results but passes only selected passages into the model’s context, and it cites from those. Pages whose answers sit in short, self-contained passages are easier to select and use.
See how AI answers describe your brand, find out why competitors get picked, and close the gaps with AI agents that work alongside your team.
14-day free trial or 50% off your first month with code SAVE50.
Continue exploring key AI visibility concepts.
Measure and improve how often your brand appears in AI-generated answers.
Learn more →Strategies for increasing visibility in answer engines and AI summaries.
Learn more →Optimizing content for AI-powered discovery experiences.
Learn more →Understand how OpenAI retrieves and synthesizes information.
Learn more →AI-generated summaries that appear directly in Google Search.
Learn more →Explore how Perplexity cites and presents sources.
Learn more →References and sources used by AI systems to support answers.
Learn more →Measure the quality and influence of cited sources.
Learn more →How easily AI systems can discover and reuse your content.
Learn more →New terms are added regularly.
Help us improve the page or suggest a new term →
Co-founder at Ansvisor
Cihan Geyik is the co-founder of Ansvisor, an open-source AI Visibility platform for AI Search. With more than 15 years of experience in digital marketing and growth, he writes about AI visibility, AI search, AEO, GEO, citations, and answer engines. He focuses on helping brands understand and improve their presence across ChatGPT, Gemini, Google AI Overviews, Bing AI Mode and other AI answer engines.
Add Ansvisor as a preferred source and Google will show you more of our AI search, AEO and GEO guides.
© 2026 Ansvisor. All rights reserved. Ansvisor is an open-source AI Search Execution Platform for AI Visibility. Ansvisor is a brand owned and operated by Empler AI Inc.
The open-source AI Search Execution Platform. Track how AI answers talk about your brand and act on it.
Open source · MIT license© 2026 Ansvisor, a brand of Empler AI Inc. · 1401 PE Ave. Unit 105, Wilmington, DE 19806, United States
Cookie preferences
We use cookies to keep the site working and, with your OK, to measure traffic and improve our marketing. Read our Cookies & Tracking policy.