ChatGPT Prompts for Consulting: 7 Steps for Client Work

Useful ChatGPT prompts for consulting specify the decision, source material and required output, then say how to handle missing evidence. Give the model a bounded piece of work you can review before using its answer in the next step.

The seven AI prompts for consultants below form a workflow: prepare context, research, qualify, plan a conversation, interpret the notes, draft a client memo and review it. Adapt them to the inputs and capabilities of your chosen assistant.

One practitioner in our consulting conversations described separate stages for qualification, research and outreach. These prompts apply that idea; they aren't a tested performance claim. OpenAI recommends clear context and iterative refinement in its ChatGPT prompting guidance.

Start with the example and source pack, then copy the context prompt or go to the final review.

ChatGPT prompts for consulting: set up the source pack

Fictional example: An operations consultant is considering a workflow review for a growing services firm. A public announcement says the firm is adding a delivery team. An existing contact has agreed to discuss handoffs between sales and delivery; no consulting project or budget is confirmed.

Label sources, such as S1: expansion announcement or S2: meeting notes. Include dates and the text needed to check claims. A link the model cannot open isn't source material it has read.

Use material permitted in your workspace, remove unrelated personal details and review summaries before saving them. The AI tools for consultants guide explains how to compare tools using a known task.

Replace the bracketed fields in each prompt. Each output needs your review before it becomes input for the next stage.

1. Build an editable context brief

Start with a brief you can update when your service or the assignment changes.

Create a context brief for this consulting assignment.

Inputs:
Our service and delivery limits: [paste]
Intended reader: [paste]
Decision we need to make: [paste]
Approved background and writing example: [paste]

Use only these inputs. Do not infer credentials, past results or
client needs. Omit unrelated personal information.

Return a brief under 250 words with: our relevant work, the
reader's role, the current decision, writing preferences and
facts that need confirmation. Separate explicit instructions
from your interpretation of the writing example.

Finish with any conflict in the inputs that would change the
assignment. Ask a specific question rather than silently choosing.

Review: Check the service boundaries and credentials first. In the fictional case, “we review operating handoffs” must not become “we implement enterprise software” unless you supplied that capability.

2. Build a research brief from evidence

This prompt uses supplied material. For fresh research, ChatGPT's deep research supports files, web sources and a reviewable plan, with availability depending on your plan.

Prepare a research brief for this decision: [paste decision].

Our relevant service: [paste checked context]
Source pack, with source labels and dates: [paste or attach]

Use only the supplied sources. If you cannot access a source,
identify it and do not summarize what you think it contains.

Return:
- Confirmed facts, each with its source label and supporting passage.
- Possible implications for our service, labeled as hypotheses.
- Questions the sources cannot answer.
- One next action and the fact it would establish.

Separate the date an event happened from the publication date.
Do not turn growth, a new hire or an announcement into proof of
budget or demand for consulting. Keep the brief under 400 words.

Review: Open the passage behind each material claim. An expansion announcement can establish growth plans; it doesn't establish a broken handoff process. Correct that distinction before moving to qualification.

A filled fictional input and acceptable answer

Here is a small source pack you can paste into prompt 2. Every detail is invented for this exercise.

Decision: Should we propose a discovery conversation?
Service: Review operating handoffs; no software implementation.
S1, company announcement, October 1: We plan to add a second
delivery team next quarter. Our sales handoff process is unchanged.
S2, contact email, October 2: Happy to explain how awarded work
moves into delivery. We haven't decided if we need outside help.

A hand-written example of an acceptable answer:

Confirmed facts: S1 states, “We plan to add a second delivery team next quarter.” This establishes a plan, not a completed expansion. S2 says, “Happy to explain how awarded work moves into delivery,” and “We haven't decided if we need outside help.” The contact has offered a conversation; demand for consulting remains unconfirmed.

Hypothesis: The additional team could make the handoff process worth reviewing. Neither source establishes a process problem.

Unanswered questions: How will the additional team affect handoffs? Does the firm need outside help? Who would decide, and is there a budget?

Next action: Accept the offered conversation and ask how the current process will serve the additional team. This would establish the contact's assessment of the process and any support they want.

Catch a response that recommends “a new CRM to fix handoff delays.” Neither source establishes delays or a software need, and implementation falls outside the stated service. This example defines the review standard; it isn't a reported result from testing a model.

3. Qualify the opportunity without filling the gaps

Identify the buying condition that still needs checking, even when a company fits your target market.

Assess this possible consulting opportunity.

Service, ideal engagement and exclusions: [paste]
Checked research brief: [paste]
Contact history or buyer statements, labeled and dated: [paste]

Use only these inputs. Distinguish a reported statement from a
verified purchasing fact. Keep unknowns explicit.

Return a short record covering service fit, the work under
discussion, problem owner, purchasing authority, current approach,
timing and any agreed next step. Cite the input supporting each.

Choose: investigate, pursue a defined conversation, revisit at a
known trigger, or decline a confirmed mismatch. Explain the
choice in two sentences and name the next question that could
change it. Do not infer authority from a senior title or invent
a project because the company matches our target profile.

Review: In our fictional example, an agreed conversation supports discovery. It doesn't support writing a proposal yet. Our guide to how consulting firms sell puts qualification in the wider relationship and buying process.

4. Prepare a discovery conversation

Use the unresolved qualification questions to shape the meeting.

Prepare a discovery agenda for [meeting length] with [role].

Purpose agreed with the contact: [paste]
Checked context and qualification record: [paste]
Questions already answered, with sources: [paste]

Draft an opening that states our understanding without asserting
an unconfirmed problem. Then suggest up to five questions, in
an order that follows how the work happens.

For each question, explain what decision its answer informs.
Include a question that could show our service is unnecessary.
Do not repeat answered questions unless the sources conflict.

Finish with a way to agree a next step if useful. Leave room
for the contact to say that the issue is already covered.

Review: “Where do handoffs become difficult?” assumes difficulty. In the example, begin with “How does work move from sales to delivery today?” Follow the answer before narrowing the question.

5. Interpret meeting notes without manufacturing agreement

Suppose the fictional meeting produces these notes: sales describes late scope changes, delivery says it sometimes lacks confirmed requirements, and the contact asks for an outline of a diagnostic. Nobody supplies delay measurements or approves an engagement.

Use actual notes in your own work. Never add the fictional details to a real account.

Turn these meeting notes into an evidence record.

Meeting date, participants' roles and notes: [paste]
Prior checked brief: [paste]

Use only the notes and brief. Do not recreate missing questions
or invent quotations. Paraphrase unless the notes explicitly
preserve the speaker's exact words.

Separate reported observations, disagreements, possible
explanations and explicit decisions. Attach a note reference
or short supporting passage to each material point.

List agreed actions with their stated owners and timing. Mark
unstated owners or dates as unknown. Flag any claim in the prior
brief that the meeting contradicts. End with the missing evidence
needed before recommending work; do not treat interest in an
outline as approval to start an engagement.

Review: Preserve the distinct observations from sales and delivery. These notes don't establish a disagreement or a shared diagnosis. Don't let the summary invent either, or quantify an impact nobody measured.

6. Draft a client memo with clear acceptance criteria

In the example, the contact requested a diagnostic outline. Keep the memo within that request.

Draft a client memo using the material below.

Reader and requested purpose: [paste]
Checked evidence record: [paste]
Approved scope and service limits: [paste]
Permitted external statements and writing sample: [paste]

Write no more than 400 words. Explain what we heard, what remains
unclear and a proposed way to investigate it. Label proposals
as proposals. Do not add fees, deadlines, results or commitments
unless the inputs explicitly authorize them.

Use ordinary language and the reader's terms. Avoid inflated
claims and generic praise. End with one decision the reader
can respond to.

Below the memo, provide internal review notes mapping factual
claims to source passages. Keep those notes separate from the
client draft and flag any sentence that lacks support.

Review: Check that every proposed activity fits the service you actually deliver. Remove claims that depend on an untested diagnosis. You can use the same discipline in client nurturing, where a useful note may be enough without a sales request.

7. Review the draft against the original evidence

Ask for a targeted review, then make the judgment yourself. Having an AI review its own draft doesn't independently establish accuracy.

Review the draft below before a consultant sends it.

Draft: [paste]
Original source pack and checked meeting notes: [paste]
Approved purpose, scope and acceptance criteria: [paste]

Identify only material problems: unsupported factual claims,
omitted contradictions, invented commitments, wording that
overstates the relationship, or scope beyond our service.

For each problem, show the sentence, the relevant source passage
or missing evidence, and the smallest proposed correction.
Do not rewrite the whole document for style. If you cannot
inspect a source, mark the affected claim unverified.

Finish with remaining checks for the human reviewer. Do not
declare the document fact-checked because it contains citations.

Review: Verify the flagged points and independently inspect the claims that matter to the decision. A second draft can sound more cautious while keeping the same incorrect assumption.

Make the workflow reusable

Save the checked context brief, useful prompts and dated sources with a named reviewer. Update the context when the assignment changes.

A failed output doesn't always call for a longer prompt. If the missing input is purchasing authority or a colleague's willingness to introduce you, ask the person who can establish it. Wording can't supply that evidence.

For repeated account work, Luna provides a place to work with account research and supplied context. Whichever tool you choose, keep the useful discipline: define the next piece of work, review what it produced and carry forward only what the evidence supports.