Quick answer: A context document is a one to two page file that gives any AI tool the background it cannot infer on its own: who you are, who you serve, what you offer, how you sound, how you make decisions, and what you are working toward. It exists because AI models measurably lose context as conversations grow, and because model updates can reset what a tool appeared to know about you. Saved as a PDF or Word document, it becomes a durable artifact you re upload to ChatGPT, Claude, Muse, Gemini, or any other tool in a single message. The fastest way to build one is to have the AI interview you in voice mode, one question at a time: about 20 minutes for a working version, about 90 minutes for a comprehensive version that can power an AI assistant acting on your behalf. Update it every three months, or whenever your business objectives change.
The Problem: AI Does Not Remember You The Way You Think It Does
Half of U.S. adults now use an AI chatbot, up from a third in 2024, and one in four uses one daily, according to the Pew Research Center’s Americans and AI 2026 report, based on a survey of 5,119 adults conducted in February 2026. Yet in the same survey only 18 percent of adults describe themselves as extremely or very confident in their ability to use these tools. The gap between use and confidence has a cause, and much of it comes down to a single misunderstanding: people assume the AI knows them. It does not, and the research on why is unusually clear.
Three findings matter.
One: AI Gets Lost In Conversation
Researchers at Microsoft Research and Salesforce Research tested 15 leading language models across more than 200,000 simulated conversations and found an average performance drop of 39 percent when a task was revealed gradually across several turns rather than specified up front. The drop appeared in conversations as short as two turns, and the authors describe the failure plainly: when a model takes a wrong turn, it gets lost and does not recover (Laban et al., 2025). The practical lesson is the opposite of how most people use AI. Giving the tool everything it needs at the start of a conversation outperforms building up to it.
Two: More Context Is Not The Same As Usable Context
A 2025 study from Chroma Research, Context Rot, evaluated 18 frontier models including GPT 4.1, Claude 4, and Gemini 2.5, and found that every one of them degraded as input length increased, well before the context window was full. This confirmed and extended the Stanford finding published in Transactions of the Association for Computational Linguistics, known as Lost in the Middle, that models use information at the beginning and end of a long input far more reliably than information buried in the middle. A long, rambling chat history is therefore the worst possible place to keep the facts about your business. A short, structured document placed at the top of a conversation is the best.
Three: Model Updates Can Reset The Relationship
Every major AI vendor now ships new models several times a year. Behaviour shifts with each release, memory features change, and the assistant that seemed to understand your business last quarter can answer like a stranger after an update. Nothing in a chat history survives that reliably. A file does.
Give the AI a complete, well structured briefing at the start of every conversation, from a file you control. That is a context document.
What A Context Document Contains
Think of it as the briefing a brilliant new hire would receive on the first morning. The core version covers:
- Who you are and what you do
- Who you serve, specifically: industry, size, role, situation
- The problems those people bring you most often, in their words
- What you offer, with pricing and what is included
- What makes your approach different, in plain language
- How you sound: three adjectives and two or three sentences you have actually said or written
- Words and phrases you never use
- Your goals for the next ninety days and for the year
- Boundaries: what the AI must never claim, promise, or invent
- Current projects the AI should know about
The comprehensive version adds the material an assistant needs in order to act for you rather than draft for you:
- Your strengths, with an example of each
- Your weaknesses, with an example of each
- How you make decisions, and where that goes wrong
- What you tend to say yes to that you should decline
- The tests an opportunity must pass before you agree to it
- What drains you and what energises you
- The boundaries an assistant should hold on your behalf
The Method: Let The AI Interview You, One Question At A Time
Writing a context document from a blank page produces the website version of a business. Interviewing produces the true one. Spoken answers carry sentiment and intent that typed answers tend to polish away: the real frustration with a certain kind of client, the actual reason behind a revenue target, the phrase a founder uses with every customer and never writes down.
One rule governs the interview. The AI must ask one question at a time and wait for the answer. Left to itself, it will ask six questions in a row, two will get answered, and the document will have holes. One question, one answer, one follow up.
This is the method used in training sessions with executive teams, association members, and, most recently, a room of some of the top professional speakers in the United States. It is consistently the piece people say they wished they had learned first.
Version One: The 20 Minute Context Document
This is the right starting point for almost everyone. Open ChatGPT, switch to voice mode, and read this prompt aloud or paste it:
Copy and paste prompt
Walk while you talk if it helps. When the interview ends, read every line of the write up and change anything that sounds like marketing copy rather than you.
Version Two: The 90 Minute Comprehensive Context Document
This version is for anyone who wants AI to do more than draft: to triage, to schedule, to protect time, to say no on their behalf. It goes where the short version does not, into strengths, weaknesses, and the decisions you tend to get wrong. Use this prompt:
Copy and paste prompt
Set aside the time. One of the comprehensive interviews behind this article ran past 90 minutes, and it is the reason the next section exists.
Case Study: The Document That Became James
The first context document in this business took twenty minutes and improved every tool overnight. The one that changed how the business operates took ninety, and most of that time went to the uncomfortable questions: what are your strengths, what are your weaknesses, and what do you keep doing that costs you?
The honest answer to the third question was saying yes to too many things. It has been the most expensive habit in this business for years, and it went into the document in plain language, together with the four tests every opportunity now has to pass: does it bring joy, impact, profit, and peace?
That document became James, the AI Chief of Staff who handles email, runs the calendar, and, most importantly, knows that the person he works for will overcommit if nobody is watching. When a request arrives, James holds it against the four tests before it ever reaches a yes. He can do that only because the context document told him the truth about his boss. A document written from the website version of the business could never have built him.
A context document is workflow redesign at the scale of one person. It changes what the AI is given before it is asked to do anything.
The broader point is supported by enterprise data. McKinsey’s The State Of AI In 2026 survey, published in August 2026, found that nearly nine in ten organisations now use AI in at least one business function, but only about six percent qualify as high performers seeing significant financial impact, and the practice most strongly associated with that impact is redesigning the workflow rather than adding the tool.
One Document, Every Tool
Nobody works in a single AI tool anymore. On the personal side of this business, ChatGPT, Claude, and Muse are the current go to tools, and all three hold the identical context document. OpenAI’s prompting guidance names context, meaning information the model could not know on its own, as a core component of an effective prompt; Anthropic’s guidance makes the same recommendation. When that context lives in one file, a keynote outline in Claude, a LinkedIn post in ChatGPT, and a quick draft in Muse all arrive in the same voice, because the source of truth is the document, not the memory inside any one app.
How Can A Context Document Get Me More Business?
A context document is usually introduced as a time saver. It is also a revenue tool, for three reasons.
1. It Connects You With The Right Clients
When the AI knows exactly who you serve, what they struggle with, and the words they use, everything it drafts speaks to that buyer: the proposal, the follow up email, the post, the pitch. Content written for everyone attracts no one. Content written for a defined client attracts that client.
2. It Improves Your AI Search Ranking
This matters most for members of the Get FOUND community. AI search engines surface businesses whose content answers a specific question for a specific audience, consistently, across every channel. Every piece of content created in ChatGPT, Claude, or Muse with a context document attached carries that consistency automatically: the same positioning, the same language, the same proof. Over time that is what ranks in AI search and what makes a business the answer when a buyer asks ChatGPT who to hire. Members can build theirs step by step inside the Get FOUND community.
3. It Gets The Work Right The First Time
Generic prompts produce generic drafts and several rounds of correction. A context document means the AI understands who you are before the first word is written, so the first draft is usually the usable one. That saves time, it saves money that would otherwise go to rework or outside help, and it frees the hours that actually generate revenue: conversations with clients, delivery, and selling.
Save time. Save money. Make more money. That is the business case for a context document in one line.
How To Save, Use, And Maintain The Document
Save it as a file, not as a chat. A chat can be lost; a file can be re uploaded. Keep a Word document as the master, export a PDF each time it changes, and put the date in the file name, for example “Context Document, November 2026”, so you always know which version you are handing over. Store it where every device can reach it.
At the start of a new conversation, upload the file or paste its contents and add one line: “Use this as background for everything we do in this chat.” In tools with a project or memory feature, store it there once so every conversation in that project starts fully briefed.
Update it every three months, or sooner whenever business objectives change: a new offer, a new price, a new goal, a new season. Re run the full interview if a great deal has shifted; otherwise ask the AI to interview you only about what has changed since the last version. A stale context document produces confident, outdated answers, which is worse than no document at all.
Frequently Asked Questions
What is a context document for AI?
A one to two page file describing who you are, who you serve, what you offer, how you sound, how you decide, and what you are working toward, supplied at the start of an AI conversation so the output is specific rather than generic.
Why does AI forget what I told it?
Research from Microsoft and Salesforce shows models lose an average of 39 percent of their performance when information arrives gradually across a conversation, and Chroma’s Context Rot study shows accuracy declines as inputs grow. Model updates can also reset memory. A saved context document is re uploaded in one message and restores the background.
How long does it take to create a context document?
About 20 minutes for a working version. A comprehensive version that covers strengths, weaknesses, and decision patterns takes over an hour, often 90 minutes.
Why should the AI ask one question at a time?
Several questions at once produce shallow answers and gaps. One question, one answer, then a follow up is how an interview surfaces what is actually true.
Why use voice mode instead of typing?
Spoken answers capture sentiment and intent that typed answers polish away. The interview surfaces the real customer, the real goal, and the phrases you actually use.
Should I save it as a PDF or a Word document?
Keep the Word document as the editable master and export a PDF for uploading. A PDF uploads cleanly to every major tool and is harder to change by accident.
Which AI tools can use a context document?
Any of them. ChatGPT, Claude, Gemini, Muse, Copilot, Perplexity, and most specialised tools accept a pasted or uploaded file at the start of a conversation.
How often should I update it?
Every three months, or whenever an offer, price, goal, or project changes.
Can a context document be used to build an AI assistant or agent?
Yes. The comprehensive version is the foundation for an assistant that acts on your behalf, because it tells the assistant not only what you do but how you decide and where you need protecting.
What Speakers Say
“In one hour she gave me more insight into the wise use of AI than mini courses had ever done.”
Jim Cathcart, CSP, CPAE, Speaker Hall of Fame
“Susan Sly is a gift! She blew my mind, taught me tons, and helped me change my business for the better.”
Brad Montgomery, CPAE, Speaker Hall of Fame
Once AI Knows Your Business, Does AI Search?
The Get FOUND Method teaches the five steps to becoming the answer when customers ask ChatGPT, Google, and Perplexity who to hire.
Bringing a team, an association, or a room of professionals up to speed on AI? Susan Sly is a female AI keynote speaker and two time award winning AI founder who trains organisations and professional speakers on practical, human centred AI. Check Susan’s availability.
References
- Pew Research Center. Americans and AI 2026: Chatbots, Smart Devices and Views on Impact. June 17, 2026. Survey of 5,119 U.S. adults, February 17 to 23, 2026.
- Laban, P., Hayashi, H., Zhou, Y., and Neville, J. LLMs Get Lost In Multi-Turn Conversation. Microsoft Research and Salesforce Research, 2025.
- Hong, K., Troynikov, A., and Huber, J. Context Rot: How Increasing Input Tokens Impacts LLM Performance. Chroma Research, 2025.
- Liu, N. F., et al. Lost in the Middle: How Language Models Use Long Contexts. Transactions of the Association for Computational Linguistics, Vol. 12, 2024.
- McKinsey and Company. The State of AI in 2026: On the Road to ROI. August 25, 2026.
- OpenAI. Text Generation and Prompting Guide.
