Electronic communication is no longer written exclusively for human readers. Emails, support tickets, instant messages and project updates are increasingly being summarised, organised and acted upon by artificial intelligence.
A recipient may ask an AI assistant to summarise a long email thread, prepare a response, create a task or identify the next steps. In other cases, the message may be copied directly into a chatbot with a simple instruction such as, “What should I do about this?”. The problem is that AI often receives only the visible message. It may not know what was discussed during a phone call, decided in a meeting or agreed upon in another platform. When important background information is missing, the AI has to make assumptions – and those assumptions may be wrong.
As AI becomes more involved in everyday business communication, companies should begin writing messages that are understandable not only to people, but also to the digital assistants helping them.
The Invisible Context Problem
Human communication relies heavily on shared context. A colleague may understand a message such as: “Please proceed with the second option we discussed and have it ready by Friday.”
The colleague attended the meeting, knows which project is being discussed and remembers what ‘the second option’ means. An AI assistant receiving only that message does not.
- Which project the instruction relates to
- What the available options were
- Who approved the decision
- What must be delivered
- Which Friday is intended
- Whether ‘ready’ means a draft, a completed document or something sent to a customer
Even when an AI tool is connected to company systems, it may not have permission to access every meeting, private message, document or project. The information may exist somewhere, but not within the AI agent’s available context.
Restate Important Off-Thread Conversations
When a decision was made outside the current email or message thread, briefly explain it. Instead of writing: “As discussed, let’s go with the managed option.” Write: “Following our Teams call on 2 September, we agreed to use the fully managed server option because it includes monitoring, backups and after-hours support. That discussion is not included in this email thread.”
This may seem repetitive to the people who attended the call, but it gives future readers, and AI assistants, the information needed to understand the decision. It also creates a better written record. Months later, employees will not have to rely on memory to understand why a particular option was selected.
Replace Vague References with Specific Information
Words such as ‘it,’ ‘that,’ ‘the document,’ ‘the usual one’ and ‘the other option’ are often clear to the sender but ambiguous to everyone else. Use the actual name of the project, customer, document or system wherever practical. For example, “Please update the proposal” becomes: “Please update the Mokoena Retail Cybersecurity Audit Proposal, version 4, using the pricing approved during the finance meeting.” Specific names make communication easier to search, summarise and convert into tasks. They also reduce the risk of an AI agent confusing one client, project or document with another.
Explain Exactly What Must Be Done
An AI-friendly instruction should clearly identify the action, owner, output and deadline.
- What needs to happen?
- Who is responsible?
- What should the completed result look like?
- When is it due?
- What must not happen without further approval?
Consider the difference between “Please sort out the report before Friday” and: “Thabo must update the August network-security report, correct the missing firewall results and export the final version as a PDF. Send the draft to me for approval by 15:00 SAST on Friday, 11 September 2026. Do not send it to the client until I approve it.” The second message is easier for a person to follow, but it is also much safer for an AI agent to convert into a task or draft response.
Use Exact Dates, Times and Time Zones
Relative dates can easily cause mistakes, especially when messages cross time zones or are processed later. Instead of saying “Please send it tomorrow morning,” use: “Please send it by 09:00 SAST on 14 September 2026.” An AI agent reading the message several days later may not know which date “tomorrow” originally referred to. Exact dates also make automated reminders and calendar entries more reliable.
Make the Status of Decisions Clear
AI assistants can struggle to distinguish between an idea, recommendation, request and approved instruction.
- Proposal: This is still being considered.
- Recommendation: This is the preferred option, but it has not been approved.
- Decision: This has been approved and may proceed.
- Action required: The recipient must complete a specific task.
- For information only: No action or response is required.
For example: “Decision: The company will migrate the application to Azure. The migration date still requires approval.”
This prevents an AI assistant from treating every sentence in a brainstorming conversation as a confirmed decision.
Introduce an ‘AI Context’ Section
One practical solution is to include a dedicated AI Context section at the end of important electronic communication.
The human recipient does not necessarily need to read it. Its purpose is to provide structured background for AI assistants, automation tools and future employees who may not have access to the original discussion.
| AI CONTEXT – Optional for the human reader
Communication purpose: Background not included in this thread: People and roles: Decision already made: Action required: Required output: Deadline: Constraints: Relevant references: Do not assume: |
Let AI Generate Context for Other AI
In some organisations, the sender’s AI assistant may have access to more information than the recipient’s assistant.
- Previous email threads
- Calendar meetings
- Project-management tickets
- Customer relationship management records
- Approved proposals
- Internal documents
- Earlier decisions and action items
The sender could ask that AI to generate an AI Context section before the message is sent. A label should make this transparent: “AI-generated context summary – reviewed by the sender.”
The sender must still review the section before sending it. AI can confuse similar projects, include an outdated decision or present an assumption as a fact. A generated context section should therefore identify its source material wherever possible. For example: “Sources used: Teams meeting notes from 2 September, ticket INFRA-221 and Azure quotation Q-1048.”
Separate Facts, Assumptions and Unknowns
AI systems can respond confidently even when information is incomplete. Good communication should therefore distinguish confirmed facts from assumptions.
- Confirmed: The customer approved the project budget.
- Assumption: The customer will accept a Saturday migration.
- Unknown: The final downtime window has not been confirmed.
This structure helps prevent an AI agent from presenting the Saturday migration as an approved arrangement.
Name Attachments and Sources Clearly
Avoid referring only to “the attached file.” Include the full filename, version and purpose. Instead of “Please review the attachment,” write: “Please review Mokoena-Cybersecurity-Proposal-v4.pdf, particularly the pricing table on page 6. This version replaces v3.” Clear references allow AI assistants to connect instructions to the correct material.
Do Not Put Sensitive Information into AI Context Sections
An AI Context section should improve clarity without becoming a container for unnecessary personal or confidential information. Do not include passwords, private keys, banking credentials, authentication codes or personal data that the recipient does not require. Organisations should also understand which AI tools their employees are permitted to use before encouraging messages to be copied into external chatbots. AI-friendly communication should still follow the principles of minimum access and minimum necessary disclosure.
Keep the Human Message Clear
Writing for AI does not mean turning every email into a technical specification.
The main message should remain concise and readable. Additional machine-useful information can appear beneath it in a separate context section.
- Subject: Clear project or action name
- Main message: What happened, what is needed and by when
- Action required: The specific next step
- AI Context: Optional structured background and constraints
The goal is not to make communication longer for the sake of it. The goal is to remove ambiguity where ambiguity could lead to the wrong response or action.
Why AI-Friendly Communication Matters
Better contextual communication can reduce unnecessary clarification, improve handovers and create more reliable written records. It can also help AI assistants produce more accurate summaries, draft more appropriate responses and create better tasks or reminders. Most importantly, it makes automation safer.
An AI agent should not be expected to infer whether a request has been approved, which customer it concerns or whether it is allowed to send a document externally. Those details should be stated clearly. As AI assistants become routine participants in business communication, writing with context will become as important as writing with correct spelling and grammar.
Final Thoughts
The future of electronic communication is not simply human-to-human. It is increasingly human-to-AI-to-human, and sometimes AI-to-AI. Messages must therefore be able to stand on their own. They should explain relevant off-thread conversations, identify the exact action required and clearly state deadlines, approvals, constraints and unresolved questions.
An AI Context section offers a practical bridge between natural human communication and the structured information AI agents need. It gives machines additional background without forcing the human recipient to read unnecessary detail.
The more responsibility we give AI assistants, the less we should expect them to guess.
This post was written by AI and reviewed for quality and accuracy by a real human.
