Consulting is reading, structuring, and writing under time pressure, and AI is good at all three. Interview synthesis that took a weekend takes an afternoon. An issue tree, a storyline with action titles, a first-draft SOW, a red-team review of your own recommendation: each is minutes of prompting plus your editing. The client-specific judgment and the accountability for the answer do not move.
The risks are specific. Models produce plausible market sizes, benchmarks, and case studies that do not exist, and a fabricated statistic on slide four ends an engagement. They default to average frameworks; the value you sell is the opposite of average. And they will summarize a transcript by inventing a quote that sounds like what the interviewee meant.
Confidentiality is the hard constraint. Most master services agreements restrict third-party processing of client data, some now address AI specifically, and public-company clients bring material nonpublic information with them. Check the contract, use tools your firm and the client have approved, and de-identify before pasting.
Quick wins this week
- Upload three anonymized interview transcripts to NotebookLM or Claude and ask for the five themes, two supporting quotes each, and the interviewee code for every quote. Check each quote against the transcript.
- Paste your draft recommendation and ask for the ten hardest questions the client CFO will ask, ranked by how badly each would hurt without an answer.
- Describe the client's problem in three sentences and ask for a MECE issue tree with the data needed to test each branch.
- Paste the deck's slide titles in order and ask whether they read as a complete argument on their own. Rewrite the ones that are topics instead of claims.
- Give the scope notes from discovery and ask for an SOW draft with deliverables, assumptions, exclusions, and a change-control clause.
What AI can do for consultants, task by task
Interview and research synthesis
Anonymize transcripts (names become codes), upload them, and ask for themes, supporting quotes with interviewee codes, contradictions between interviewees, and gaps in what you asked. Verify every quote against the source; models paraphrase and present the result as verbatim. Keep the synthesis in the firm's approved workspace, never a personal account.
Structuring the problem
Give the client situation, the question they actually asked, and the constraints, and ask for an issue tree or hypothesis tree with the analyses and data needed to test each branch. Then cut it to what matters for this client; the model's tree is textbook-complete and engagement-generic until you prune it.
Storyline and slide writing
Paste your findings and ask for a governing thought, three supports, and an action title per slide, then check that the titles alone tell the story. Draft in Gamma or your firm's template. Every number on a slide traces to a source you can name; AI supplies structure, not evidence.
Red-teaming your recommendation
Give the recommendation and the evidence, and ask the model to argue the opposite case, list what would change your mind, and name the assumption the whole argument rests on. Use the output to strengthen the analysis or narrow the claim before the steering committee finds the hole.
Proposals, SOWs, and status reports
Paste discovery notes and your firm's SOW template and ask for scope, deliverables, timeline, assumptions, exclusions, and a change-control clause. For weekly status, paste the tracker and ask for a one-page update in the client's format. Read the assumptions section twice; it is where scope creep starts.
Market and competitor scans
Use a citation-based tool for the first pass and open every source; a figure without a link you have checked is not a fact. Ask what the sources disagree on, which is usually the interesting part. Never put a market size or growth rate on a slide you cannot trace to a named, dated source.
Prompts for consultants
Replace the bracketed placeholders, paste into any chat assistant, and iterate on the result.
Synthesize interview transcripts
You are a consulting analyst synthesizing stakeholder interviews for [CLIENT DESCRIPTION, ANONYMIZED] about [ENGAGEMENT QUESTION]. I have uploaded [NUMBER] transcripts labeled by interviewee code. Produce: (1) the five to seven themes that recur across interviews, (2) for each theme, two verbatim quotes with interviewee code and line reference, (3) points where interviewees disagree, (4) questions we did not ask that the transcripts suggest we should have. Quote exactly; never paraphrase inside quotation marks. Do not attribute a view to more interviewees than actually expressed it.
Tip: Spot-check five quotes against the transcripts before the synthesis leaves your laptop.
Build an issue tree
You are a strategy consultant. Client situation: [THREE SENTENCES]. The question the client asked: [QUESTION]. Constraints: [BUDGET, TIMELINE, POLITICAL CONSTRAINTS]. Build a MECE issue tree three levels deep. For each leaf, name the analysis that would resolve it, the data required, and where that data usually lives in a company like this. Mark the three branches most likely to hold the answer and explain why. Do not recommend anything yet.
Tip: Prune to the branches that matter for this client; completeness is the model's default, not the deliverable.
Red-team my recommendation
You are a skeptical [CLIENT ROLE, E.G. CFO] reviewing a consultant's recommendation. Recommendation: [ONE PARAGRAPH]. Evidence: [BULLETS]. Argue the strongest case against it, list the three assumptions the argument depends on and how each could fail, identify what evidence would change your mind, and write the single hardest question you would ask in the steering committee. Be specific to this client, not generic. End with a one-line verdict: approve, approve with conditions, or reject, with the condition stated.
Tip: Run it the night before, while there is still time to fix the analysis.
Storyline and action titles for a deck
You are a consulting engagement manager. Findings: [PASTE FINDINGS OR NOTES]. Audience: [WHO THEY ARE AND THE DECISION THEY FACE]. Write a governing thought in one sentence, three supporting arguments, and an action title for each of [NUMBER] slides, each title a claim of no more than 12 words that the slide's evidence must prove. Under each title, list the one chart or table that proves it. Flag any title the findings given cannot yet support.
Tip: The flagged titles are your remaining analysis plan.
Draft a statement of work
You are a consulting engagement lead drafting a statement of work. Client: [CLIENT TYPE]. Discovery notes: [PASTE NOTES]. Firm template sections: [LIST OR PASTE TEMPLATE]. Draft the SOW with objectives, scope, deliverables with acceptance criteria, timeline with milestones, team and client responsibilities, assumptions, explicit exclusions, a fee section using [FEE STRUCTURE], and a change-control clause. Where the notes are silent, write TO CONFIRM rather than inventing terms. Keep it under two pages.
Tip: Every TO CONFIRM is a question for the client before signature, not after.
Want a prompt for something else? Use the Prompt Builder or browse brainstorming and strategy prompts, education and training prompts, leadership and communication prompts, meetings and summaries prompts, planning and productivity prompts, presentations and reports prompts, research and learning prompts, sales and outreach prompts, writing and email prompts.
Skills to build
Anonymizing before you synthesize
Why: Transcripts and data files carry names and details that identify people and the client. Removing them is what makes AI synthesis contractually possible.
How: Build a find-and-replace routine (interviewee codes, client codename, product names) and run it before any upload. Keep the key offline.
Sourcing every number
Why: Consulting deliverables live or die on credibility. One unsourced figure from a chat window can undo months of trust.
How: Nothing goes on a slide without a footnote you have opened yourself. AI can find candidates; you confirm them.
Prompting with the client's specifics
Why: The model's default output is the average of every case study it has read. Your client is paying for the parts that are not average.
How: Start every prompt with three sentences on what makes this client different: constraints, politics, history. Ask the model to reference them explicitly.
Arguing against yourself
Why: Steering committees find the hole in your logic in the meeting. AI lets you find it the night before, when you can still fix it.
How: Run the red-team prompt on every major recommendation before it goes to the client. Keep a list of the objections you had missed.
Reading contracts for AI terms
Why: MSAs increasingly say which tools may touch client data. Knowing the clause is the difference between efficiency and a breach.
How: Before each engagement, find the confidentiality and subprocessor clauses, ask your risk team what is approved, and write the answer at the top of your project notes.
Tools worth knowing
Claude
A careful writing and analysis assistant that shines on long documents.
ChatGPT
The general-purpose AI assistant most of your coworkers already use.
NotebookLM
A research notebook that only answers from the sources you give it, with citations.
Perplexity
An answer engine that cites its sources, built for research rather than chat.
Gamma
Generate presentations, documents, and web pages from a prompt or an outline in minutes
Microsoft Copilot
AI inside Word, Excel, Outlook, and Teams, with your company's data protections.
Cautions for consultants
Most master services agreements restrict sharing client data with third parties, and many now name AI tools specifically. Pasting client material into a consumer tool can be a breach even if nothing leaks. Read the contract, use only tools your firm and the client have approved, anonymize before uploading, and never use a personal account for client work.
Public-company clients hand you MNPI: unreleased results, deals, restructurings. Insider trading rules and your firm's policies cover what you do with it, including where you paste it. Treat any tool that retains inputs as a disclosure risk.
Models invent market sizes, growth rates, survey findings, and named case studies with complete fluency. A fabricated benchmark discovered by a client is a professional liability issue and a reputational one. Every figure and source in a deliverable is unverified until you have opened it.
Some clients require disclosure of AI use in RFPs or contracts; ask rather than assume. Recording interviews for transcription requires consent, and several states require all parties to agree. Transcripts contain personal information; anonymize them and delete recordings on the schedule the engagement agreed.
Your 30-day plan
- Week 1: Read the confidentiality and subprocessor clauses of your current MSA and confirm with your firm's risk team which tools are approved for client data.
- Week 1: Set up your anonymization routine and test it on one transcript. Then run the synthesis prompt and verify every quote.
- Week 2: Build an issue tree for a live engagement question and compare it with the one you would have drawn.
- Week 2: Run the red-team prompt on your current recommendation and fix the two weakest points before the next client meeting.
- Week 3: Draft one deck's action titles with the storyline prompt and test whether the titles alone tell the story.
- Week 4: Save your prompts, anonymization routine, and client-specifics template in a Project or custom GPT, and brief your team on what was verified and how.
Frequently asked questions
Will AI replace consultants?
Can I put client data into ChatGPT or Claude?
Do I have to tell clients I use AI?
Can AI write a consulting deck?
Which AI tool is best for consultants?
Terms used on this page
Related roles
- AI for Project ManagersAI cannot run your project, but it can draft the status report, turn a messy meeting into an action list, and pressure-test your risk register in minutes. Here is how project managers use it without losing the plot.
- AI for Financial AnalystsAI writes the first pass of your variance commentary, audits your model for hard-codes and sign errors, and turns a 10-K into a table with page references. You decide what the numbers mean and what to tell the CFO.
- AI for Data AnalystsAI writes the SQL, the pandas, the DAX, and the sentences around the chart. It cannot know your data, so your job shifts to the right question, the checked join, and keeping regulated data out of the wrong tool.
- AI for Product ManagersAI turns forty interview transcripts into themes, a rough idea into a PRD draft, and a napkin sketch into a clickable prototype, and it will also confidently invent a customer need that does not exist. Here is how product managers get the first without the second.