Your First AI Prompts as a Consultant

A consultant's prompt is not a Google search. It is a brief — the same way you would brief a junior analyst before sending them off to do work. Once you make that mental shift, your AI output quality jumps dramatically.

This lesson walks you through a battle-tested prompt structure that works for any consulting task, then gives you copy-paste templates for the situations you will hit in your first week.

What You'll Learn

• The CRAFT framework: a five-part prompt structure for consulting work
• Five high-leverage starter prompts you can use today
• How to push back on weak AI output and force it to improve
• When to use ChatGPT vs Claude vs Gemini for first-draft work

The CRAFT Prompt Structure

Most consultants get mediocre AI output because they type one-liners like "summarize this" or "make a SWOT for Acme Corp." That treats AI like Google. Instead, structure every prompt around five elements:

• C — Context: Who is the client, what is the engagement, what is the specific situation?
• R — Role: Who should the AI act as (a McKinsey EM, a CFO, a market analyst)?
• A — Action: The exact task — synthesize, draft, critique, structure, list.
• F — Format: How should the output be shaped — table, slide outline, exec summary, bullets?
• T — Tone & Constraints: Audience seniority, length, things to avoid, what "good" looks like.

Here is a weak prompt vs a CRAFT prompt for the same task.

Weak: "Make a SWOT for a regional bank."

CRAFT:

You are a senior strategy consultant at a top-3 firm advising a regional retail bank with $12B in assets that is losing 18-to-30-year-old customers to neobanks. The CFO has asked us to support a board discussion next month.

Act as a financial-services strategy partner. Build a SWOT analysis focused specifically on the threat from neobanks and the bank's ability to respond. Each quadrant should have 4–5 items, each backed by a specific reason or example. Output as a clean markdown table. Use board-level language — concise, senior, no buzzwords. Do not list generic items like "good brand recognition" without explaining why it matters in this competitive context.

The second prompt produces work you could put in a pre-read. The first produces a generic Wikipedia-style answer.

Five Starter Prompts for Your First Week

1. Synthesize a Discovery Call

You are an experienced consultant. Below is a raw transcript of a 60-minute client discovery call with the COO of [client]. Produce: (1) a 5-bullet executive summary, (2) the top 3 problems the COO mentioned in their own words, (3) any contradictions between what the COO said and what was said earlier in the call, (4) 5 follow-up questions we should ask in the next session. Output as markdown with clear headers.

Transcript: [paste transcript]

2. Draft a Hypothesis Tree

You are a strategy consultant building an issue tree for a client engagement. Client: [describe]. Core question: [insert question, e.g., "Should ClientCo enter the German market?"]. Build a MECE issue tree with 3 main branches and 3–4 sub-branches each. For each leaf, suggest the analysis or data we would need to test it. Output as a nested bullet list.

3. Critique Your Own Slide

Act as a McKinsey engagement manager reviewing a junior consultant's slide. Below is the slide title and three bullet points. Tell me: (1) is the title an action title or a topic title? (2) do the bullets support the title? (3) what is the so-what? (4) rewrite the slide to be sharper.

Title: [...] Bullets: [...]

4. Draft a Client Status Email

Draft a weekly status email to the client sponsor [name, role]. Tone: confident, brief, no fluff. Sections: progress this week (3 bullets), what's next (2 bullets), one open question for them, one ask. Use these inputs: [paste your week-in-review notes].

5. Pressure-Test a Recommendation

You are a skeptical client CFO who has seen many consulting recommendations fall apart in execution. Read the recommendation below and respond with the 5 toughest questions you would ask in a steerco. For each, predict how a weak consultant would answer and what a strong consultant should answer.

Recommendation: [paste your recommendation]

Pushing Back on Weak Output

If the first answer is generic, do not abandon the prompt — push back. Real exchanges from working consultants:

• "This reads like a Wikipedia article. Rewrite it as if it were the executive summary slide of a board pre-read."
• "You used three buzzwords: 'synergy', 'leverage', and 'unlock'. Rewrite without any of them."
• "The so-what is missing. After each bullet, add the implication for the client in italics."
• "Be more contrarian. Argue the opposite case in 4 bullets so I can stress-test my recommendation."

Each of these turns a weak draft into something usable in 10 seconds.

ChatGPT vs Claude vs Gemini for Consulting

A quick guide based on real-world consultant usage in 2026:

• ChatGPT (GPT-5): best general workhorse, strongest with structured outputs (tables, JSON, frameworks), excellent file uploads, the largest Custom GPT ecosystem.
• Claude (4.6 / Opus): best for long-form synthesis (50+ page documents), nuanced executive writing, and stress-testing arguments. Claude Projects is excellent for ongoing engagements.
• Gemini (2.5 Pro / 3): best when your client lives in Google Workspace, when you need very long context windows (1M tokens), or when you need to ground answers in Google Search results.

Most independent consultants pay for two — typically ChatGPT Plus and Claude Pro — and use each for what it does best.

Key Takeaways

• Use the CRAFT framework: Context, Role, Action, Format, Tone & Constraints. Stop typing one-liners.
• Keep five reusable prompts ready: discovery synthesis, issue tree, slide critique, status email, recommendation pressure-test.
• If the output is weak, push back with specific edits before starting over — three rounds of refinement beats one perfect prompt.
• Pick tools by task: ChatGPT for structure, Claude for long synthesis, Gemini for Workspace integration.