Editorial path — 3 · Understand AI without hype · Part 4/9
Prompt quality depends on business context, not model magic.
"Write an email to the client." — five words, zero context. The AI will invent a tone, a problem, and maybe a promise you never made.
Context in a prompt is not padding. It is what turns a plausible answer into one that is usable for your business, your client, this decision. This article focuses on what you write before you press Enter — not technical context windows or knowledge bases (covered in articles 6 and 8 of this series).
At a glance
- Without business context, AI fills gaps with generics — often credible, sometimes wrong
- Useful context: who, what, for whom, constraints, tone, what to avoid
- Two or three sentences of context beat a vague paragraph of instructions
- Always review — context reduces error; it does not remove it (human-in-the-loop)
Why context changes everything
A model does not know:
- You have worked with this client for three years
- You already promised delivery on Friday
- Your team uses first names internally but formal address with clients
- You cannot name a supplier in writing
Without that, it guesses. And it guesses well — that is the trap. A fluent, professional answer can contain a promise you never authorized, a tone mismatched to a long relationship, or invented facts to fill gaps.
Pitfall 2 from AI adoption failures — vague prompts — gets fixed here, at the level of each request. The next article in this series shows how to standardize these practices across your team.
Example 1 — Email to an unhappy client (late delivery)
Weak prompt
Write an email to a client apologizing for a delay.
Problem: the AI does not know which delay, which industry, which tone, or what you are allowed to offer. Typical output: generic apologies, vague promise ("we are doing our best"), sometimes a refund you never authorized.
Prompt with context
You write for Alain, operations director at a 25-person services SMB in Quebec.
Context:
- Client: accounting firm, 4-year relationship, professional but warm tone (formal "you").
- Situation: analysis report promised Monday, now due Thursday (3-day slip) because data access arrived late.
- Already done: call Tuesday to warn; team prioritized the work.
- Constraints: do not blame the client; do not promise a discount; confirm delivery Thursday 4 p.m.; offer a 15-minute call Thursday morning if helpful.
- Length: 120–150 words, Canadian English.
Draft the email (subject + body).
Difference: the draft matches facts, tone, and limits. You still review — especially date and time — before sending.
Example 2 — Internal announcement (new tool)
Weak prompt
Announce to the team that we are adopting a new project management tool.
Problem: unknown tone (cheerful? top-down?), no "why," no timeline, no answer to usual fears. The team gets generic corporate copy — and ignores half of it.
Prompt with context
You draft an internal note for a 12-person team (consultants + coordinators) at a Quebec SMB.
Context:
- Tool: Asana (replaces shared Excel lists and "who does what" email threads).
- Why: wasted time tracking work, missed client follow-ups, no visibility for leadership.
- Timeline: 1-hour training next Tuesday; mandatory from July 1; Excel read-only for 2 more weeks.
- Known concerns: "another tool," data entry burden, fear of surveillance.
- Tone: direct, respectful, no IT jargon; acknowledge this is an adjustment.
- Format: subject + 3 short paragraphs + "Next steps" bullets (max 4).
Write in Canadian English.
Difference: actionable message, clear dates, objections addressed — ready for a colleague review before send.
Example 3 — Meeting summary for executives
Weak prompt
Summarize this meeting.
(with 8 pages of notes pasted with no structure)
Problem: summary too long or too vague; decisions buried; actions without owners; invented numbers if notes were fuzzy.
Prompt with context
You prepare a summary for leadership (3 people who were not in the meeting).
Meeting context:
- Topic: choosing a vendor to automate time reporting (3 options evaluated).
- Duration: 90 min; attendees: ops, finance, one consultant.
- Decision: option B (integration with existing accounting tool), cap $18,000 first year.
- Not decided: exact go-live date (depends on accounting API).
- Actions: Marie — final quote by Friday; Jean — validate data security with IT; Alain — present to committee on the 15th.
Raw notes below (do not invent beyond this text):
[… paste notes …]
Output format:
- Max 5 bullets for decision and rationale
- Table: Action | Owner | Due date
- "Risks / open points" section (max 3 bullets)
- Max 350 words, Canadian English
Difference: leadership gets what it needs — decision, numbers, next steps — without reading 8 pages.
Context mini-checklist (copy-paste)
Before an important prompt, check:
| Question | Example answer |
|---|---|
| Who am I / who is speaking? | Ops director, 25-person SMB, Quebec |
| Who is the audience? | Long-term client, formal tone |
| What is the core fact? | 3-day delay, delivery Thursday 4 p.m. |
| What must we avoid? | No discount, no blame |
| Format / length? | 120 words, subject + body |
| Which language? | Canadian English |
Often five lines of context cut revision time in half.
What context does not replace
- Sensitive data in unapproved tools — see data safety with AI
- Judgment on contracts, HR, or regulatory advice
- Fact-checking — dates, amounts, proper names
- Structured knowledge bases — when answers must draw on internal policies, not just what you paste into chat
Context brings AI closer to your reality. It does not make it accountable for you.
Where you are
You've identified the five adoption pitfalls; this article shows how business context in each prompt turns generic answers into usable drafts. Next: Prompting tips for leaders and teams — repeatable practices at team scale.
Want a working session to adapt templates to your workflows? Book a conversation.