Hands-On: Build Your AI Workflow

In organizations that have adopted AI tools professionally, documenting how AI was used is increasingly expected — not because AI use is suspicious, but because it creates accountability and enables quality review. A documented AI workflow shows the human decisions behind the AI output:

• The task — what office task was being completed and why. This provides context for every other decision in the workflow.
• The tool — which AI tool was selected and why it was appropriate for this task rather than a different one.
• The prompt — the exact text entered into the AI tool. Documenting the prompt enables another person to reproduce, evaluate, and improve the output.
• The raw output — what the AI produced before human editing. Preserving the raw output makes the editing work visible and demonstrates the gap between AI output and professional output.
• The edited final version — the polished, reviewed version ready for professional use. The difference between the raw output and the final version is where your professional judgment shows.

Reaching for the same AI tool for every task is the equivalent of using a screwdriver for every job in a toolbox. The right tool matches the nature of the task. Demonstrating that you made a deliberate tool selection — rather than defaulting to whatever you opened first — is a sign of genuine AI fluency:

• For first-draft generation (emails, reports, summaries, agendas) — ChatGPT or the integrated AI in your document or email app (Copilot, Gemini).
• For grammar, tone, and clarity review — Grammarly or a specific tone-check prompt in ChatGPT ('Does this read as professional and respectful?').
• For meeting transcription and summarization — Otter.ai or Fireflies.ai, or the built-in transcription feature in Microsoft Teams or Google Meet.
• For research overviews on a topic or company — ChatGPT for initial orientation, followed by independent web searches to verify and deepen.
• For generating or explaining spreadsheet formulas — Microsoft Copilot (within Excel), Google Gemini (within Sheets), or ChatGPT with the formula pasted in for explanation.

The difference between a prompt that produces useful output and one that produces generic output is specificity. This lesson's challenge requires you to document your prompts — which means taking the time to write prompts that are specific enough to produce genuinely useful results. Here is what that looks like for common office tasks:

• Email draft — weak prompt: 'Write a follow-up email.' Strong prompt: 'Write a 3-paragraph follow-up email to a corporate client who attended our product demo yesterday. Recap the demo highlights, address their question about pricing tiers, and propose a 30-minute call next week. Use a professional but warm tone.'
• Document summary — weak prompt: 'Summarize this.' Strong prompt: 'Summarize the attached 800-word project proposal in 5 bullet points for a non-technical manager. Focus on budget implications, timeline, and the decision that needs to be made by the end of the week.'
• Research overview — weak prompt: 'Tell me about [Company X].' Strong prompt: 'Summarize what is publicly known about [Company X]'s core services, target clients, and competitive positioning in 4 bullet points. Format for a professional business audience. Flag any claims that may need independent verification.'
• Meeting agenda — weak prompt: 'Write a meeting agenda.' Strong prompt: 'Write a 45-minute meeting agenda for a quarterly client review with 3 agenda items: project status update (15 min), open issues and risks (15 min), and next quarter planning (15 min). Include a 5-minute buffer and a time-keeper note.'
• The rule of specificity — whenever you catch yourself about to type a vague, one-line prompt, stop and add: the audience, the format, the length, the tone, and the most important constraint. This 30-second investment consistently produces dramatically better output.

Reviewing AI output is not the same as reading it. Reading registers the words. Reviewing evaluates them: Are they accurate? Is the tone appropriate? Does the structure serve the purpose? Is anything missing? Does anything need to be removed? A professional review of AI output is the step that separates useful AI assistance from risky AI delegation:

• Check for factual accuracy — any specific claim (a date, a statistic, a person's title, a product feature) must be independently verified before it stays in the document.
• Check for tone appropriateness — AI tends toward formal or neutral. Sometimes that is right; sometimes a warmer, more personal, or more direct tone is needed. Adjust the register to match your relationship with the recipient and the context of the communication.
• Check for completeness — AI does not always include everything. Compare the output to your original brief: is everything you intended to communicate present? Are all required elements included?
• Remove generic filler — AI output often opens with a line like 'I hope this email finds you well.' These filler phrases weaken professional writing. Remove them or replace them with something specific and genuine.
• Personalize before sending — replace every placeholder (the client's name, the specific date, the particular detail that makes this communication about this relationship) with real, accurate information. Generic AI output that goes out as-is is immediately recognizable as AI output — and it communicates that you did not care enough to personalize it.