A note before the split. This is general guidance rather than professional advice. Your firm's policy, your engagement terms and your client's data requirements override everything below. Tools change constantly too, so the specifics are a snapshot and the judgement is the durable part. It applies whether your firm runs on ChatGPT, Claude or a sanctioned internal tool.
Consulting has adopted AI faster than almost any profession, which makes sense for a job built out of documents. It has also produced the sharpest version of the problem. A consultant's deliverable was never really the deck or the report, it was the thinking inside them, and the thinking is exactly what these tools counterfeit most fluently. A framework-shaped slide with generic content underneath looks finished at ten paces and reads hollow at one, and clients have been standing at one pace for a while now.
So the useful question is not whether to use ChatGPT on an engagement, because nearly everyone around the table already does. The question is where in the workflow it belongs, and the answer follows one distinction all the way down. The tool builds scaffolding. You supply what makes the scaffolding worth the fee.
The short answer. Trust ChatGPT with structure and volume: storylines, workplans, first drafts of frameworks and summaries, synthesis of your own anonymised notes, industry orientation before a kickoff, and the slide grind. Never put client-identifiable material into a consumer account, never use a statistic it volunteered without finding the real source, and never ship its generic output as insight. Clients can increasingly tell, and the moment they can tell, the fee stops making sense.
Where it earns its place
These are the jobs where the tool converts blank-page time into sharpening time. The pattern across all five is the same: it produces the structure fast, and the client-specific substance goes in during your edit.
| The job | What that looks like |
|---|---|
| Storylines and workplans | A deck storyline or phase plan to react to in minutes, instead of a blank page to fear |
| First drafts of frameworks and summaries | Executive summaries, section intros and framework skeletons that take your edit rather than your evening |
| Synthesising your own notes | Interview and workshop notes clustered into themes on request, anonymised before they go in |
| Industry orientation | Conversational fluency in an unfamiliar sector the week before kickoff, possibly its highest-value hour |
| The slide grind | Ten wordings of the same framework, or one message reshaped for the steering committee and the working team |
Where it will hurt you
The dangers cluster where the tool's confidence outruns its grounding. Each of these has ended careers or engagements somewhere, which is why they get a table rather than a caveat.
| Keep it away from | Because |
|---|---|
| Client-identifiable material on consumer accounts | Engagements run under NDAs, diligence work touches price-sensitive information, and consumer tiers can train on what you paste. That turns a shortcut into a breach |
| Statistics it volunteers | It will offer "73 per cent of mid-market firms" with total confidence and no source behind it. Every number gets a real citation or gets cut |
| The analysis itself, unchecked | Its reasoning is plausible rather than grounded in your client's situation. Its conclusions are hypotheses to test, never findings to ship |
| Generic frameworks shipped as insight | Boilerplate two-by-twos and committee-voiced prose have a recognisable flavour, and a client who detects it starts questioning the fee |
| Current market and company facts | Its knowledge has a cutoff, so anything about the market as it stands today needs a live source before it reaches a deliverable |
The client-data rule
Consulting's confidentiality obligation is contractual, which makes this rule easier to state than most professions get. Nothing that could identify the client, the deal or the counterparties goes into a consumer AI account. The NDA does not contain a productivity exception.
The workable version is aggressive anonymisation. The client becomes "a mid-market logistics business", the numbers get stripped or rounded beyond recognition, and the situation becomes a fact pattern. At that altitude the tool can still help with structure and language while the engagement stays inside its walls.
Where your firm provides a sanctioned enterprise tool, the sensitive synthesis belongs there instead. And where the client's own data requirements are stricter still, the client's rules win, which is worth knowing before the data room opens rather than after.
The line that holds it together
These tools have surfaced an uncomfortable truth, which is that some consulting output was always scaffolding wearing a fee. Clients now have a machine that produces scaffolding for roughly nothing, so the part they are actually paying for has become more visible, not less. That is good news for anyone who intends to supply it.
Let the tool build the structure, grind the slides and orient you fast. Then spend the reclaimed hours where the fee lives: the client-specific insight, the judgement call the data does not make by itself, the recommendation you would defend in the steering committee. The consultants worried about being replaced by ChatGPT are mostly the ones producing what it produces, and the fix for that was never going to be a better prompt.
Not sure where to start?
Scaffolding from the tool. Insight from you.
The free finder tells you which AI to open for the task in front of you, in about 60 seconds, and the full "AI, sorted." reference lands in your inbox.
Try the free finder →One question in, one answer out.
Clair helps non-technical professionals know when to trust their AI, when to check it, and when to skip it.