AI Tools for Consultants: What to Use and How to Test Them

Choose AI tools for consultants by the work you need to finish. ChatGPT can support research across sources, Claude Projects can hold reference material and instructions, and Microsoft Copilot works within familiar office applications. For account research and relationship work, consider a more focused product such as alkemy.

The selection question is what happens after the first answer. Can you check the evidence, change the brief and pass usable work to a colleague without rebuilding the context?

This is a task-based shortlist, not a ranking from a comparative product test. The recommendations below combine provider documentation checked on October 2, 2026 with our proposed evaluation method. Check the features available in your own plan and workspace before choosing.

Jump to the tools, research evidence or selection test.

Which AI tools for consultants fit each task?

ChatGPT: research with an explicit question and source boundary

OpenAI's deep research documentation describes work with uploaded files, the public web and specified sites. You can review the research plan before it runs, and the resulting report includes citations or source links. Usage and availability depend on the plan and location. See OpenAI's deep research guide.

For a consultant, a useful assignment might be: compare the published service models of a defined group of firms and identify questions for interviews. Supply the group, the comparison criteria and the period you care about. Ask the tool to distinguish a company's own claims from independent evidence.

Read the plan before approving the research. If it proposes collecting broad industry trends when you need evidence about delivery models, correct the task at that point.

A cited report still needs review. Open the sources behind claims that would change your recommendation. A source might establish that a company announced a program without establishing that the program achieved its objectives. Requesting citations helps locate evidence; it doesn't settle what the evidence proves.

Claude Projects: a working reference set for repeated tasks

Claude Projects lets you add documents and other reference material to project knowledge, then set project instructions. That gives repeated conversations a shared starting point. These are documented capabilities, rather than a claim that every answer will use every supplied fact correctly. See Anthropic's explanation of Projects.

Consider a project containing an approved service description, a writing sample and a small set of client documents you have permission to use. You could ask for a discovery agenda, revise it after a meeting, then prepare a memo using the same scope and terminology.

The setup deserves editorial work. Label current documents, remove superseded instructions and state which source wins when two documents disagree. Otherwise, a larger collection gives the model more conflicting material to interpret.

In one of our consulting working sessions, a practitioner inspected an AI-generated context summary and removed personal material before transferring it. The lesson we take from that action is practical: make reusable context visible and editable. Copying a whole conversation history is rarely the clearest way to tell a tool what this assignment requires.

Microsoft Copilot: work that already lives in Office

Microsoft documents Copilot support for creating and editing Word documents, working with Excel data and preparing PowerPoint presentations. It also describes meeting summaries and action items in Teams. See Microsoft's Copilot overview.

That makes it a candidate when the bottleneck sits inside an existing document workflow. For example, test whether you can turn an approved narrative into a draft presentation that the team can edit in its normal format.

Evaluate the resulting artifact. Check whether slide titles match the evidence, charts retain their units and any proposed action has an actual owner. A polished presentation can still contain an unsupported recommendation.

For spreadsheet work, independently reproduce the calculations that drive the decision. Specify the population, period and exclusions before asking for an interpretation. If the source data mixes completed and open projects, a neat average may answer the wrong question.

Confirm access in the applications your team actually uses. A colleague's experience in a different account or configuration won't establish what is available in yours.

alkemy: account research and relationship work

We build alkemy, so this entry describes where our product fits and what to examine. It isn't an independent endorsement.

Luna works with account research and the context you supply to prepare briefs and draft messages. Key Accounts brings company research, people and developments together. Warm Intros shows possible connection paths and shared professional history, which still needs confirmation from the person who knows the contact.

The task to test is continuity: can you start with a company question, investigate a relevant person and prepare for a conversation while retaining the evidence and relationship context?

Start with an account you understand. Look for incorrect identities, outdated roles or a supposed introduction route that a colleague wouldn't recognize. Finding a possible connection doesn't establish willingness to make an introduction.

Our guide to key account management for consultants describes the surrounding work. You can inspect a prepared example in the sample workspace, then define what a test with your own approved material would need to establish.

What research on AI in consulting does and doesn't establish

A field experiment involving 758 Boston Consulting Group consultants tested GPT-4 on defined consulting tasks. For tasks within the tested model's capabilities, participants using AI completed work 25.1% faster on average. On a separate task selected to sit outside those capabilities, the combined AI groups' correctness rate was 19 percentage points lower than the control group's. See Dell'Acqua and colleagues' paper, published in 2026.

Those results concern the experiment's tasks and GPT-4. They don't establish a productivity gain for a whole engagement, today's models or your firm's work. They support testing different tasks separately, including whether the answer is correct after the writing becomes more persuasive.

Our own research conversations raise a related workflow question. One consultant described separating qualification, deeper research and outreach into distinct stages with defined criteria. That is an observed working method, not evidence that the method improves revenue. It gives you something specific to test: whether an error at the research stage survives into the message.

Run a small test before rolling out a tool

Use a completed assignment with an answer you can check. Remove material the chosen workspace isn't approved to receive. Give each candidate the same permitted source pack and requested deliverable; record any extra setup it needs.

For a fictional test, ask each candidate for a short findings memo using this input:

Question: What should we investigate about sales-to-delivery handoffs?
S1, sales interview: Every handoff includes confirmed requirements.
S2, delivery interview: Some handoffs lack confirmed requirements.
S3, process document: Sales must confirm requirements before handoff.
No timing data or customer impact measurements are available.

An acceptable answer could say: “S1 and S2 disagree about whether every handoff includes confirmed requirements. S3 establishes the required process, not whether people follow it. Check a sample of handoffs before attributing a cause.” This is a hand-written review example, not output from a product test.

Reject a conclusion such as “Late sales changes cause most delivery delays.” The supplied evidence establishes neither the cause nor the share of delays.

Define what makes the result usable

Write the acceptance criteria before reading the output. For this example, a usable memo must distinguish what interviewees reported from what the data established. It must preserve the disagreement, cite the relevant source passages and avoid claiming a measured delay where nobody measured one.

If you only judge whether the memo sounds plausible, you miss the central risk. Include at least one incomplete source and one conflicting observation so the test checks how the tool handles uncertainty.

Count the work left for the consultant

Record the time you spend preparing inputs and reviewing the result, including manual corrections. Generation time is only part of the job.

Use this review record when comparing AI consulting tools:

Task and source pack:
Tool, configuration and date:
Required output and acceptance criteria:
Preparation time:
Time until first output:
Review and correction time:
Material factual errors:
Unsupported conclusions:
Useful work retained:
Decision: adopt for this task / adjust and retest / reject

A tool that writes quickly but leaves a senior consultant reconstructing the evidence may not improve this workflow. Count useful work retained, not pages produced.

Change the brief and test the continuation

Tell the tool that one source is outdated or a proposed interview is no longer possible. Ask it to revise the memo and identify what changed.

Then ask for the next artifact, such as an agenda. Check whether it carries forward the corrected facts and unresolved questions. This tests the handoff between tasks, where a mistaken assumption can become harder to notice.

Our AI prompts for consultants provide a staged workflow you can adapt for this exercise. Keep the prompts stable during the initial comparison so you know whether a changed result came from the tool or a different instruction.

Questions to settle before choosing

What are the best AI tools for consultants?

The answer depends on the task and your working environment. Shortlist tools that can use the required sources and produce the format your team needs, then compare the reviewed result. This guide doesn't establish a universal winner. A tool can fit research well and still be a poor choice for your team's document handoff.

Can one tool handle the whole consulting workflow?

Test that assumption at each transition. Research, analysis and client communication have different acceptance criteria, even when one product can attempt all of them. Keep using a tool where it produces checkable work; require a fresh test before relying on it for a task with different evidence or consequences.

Where should a small firm start with AI for consultants?

Choose one recurring task that already has a reviewer and a clear finish, such as preparing a meeting brief. Use approved material, record the remaining review effort and decide whether to keep the workflow. Add another task only when you can explain what the first test established.