Questions to ask an AI consultant before you hire one
Short answer: Ask eleven things, grouped by risk: what you will own, who will see your data, what will actually be running when they leave, and what happens after. An AI consultant is someone paid to find where AI fits your business and, often, to set it up. A good one answers each question in plain words with a specific example. A bad one answers with a tool name, a buzzword or a slide. Score the answers on one page and hire the person who is clear, not the person who is impressive.
You are about to sign with an AI consultant and want to know what to ask so you do not pay for slides and get nothing.
I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.
Full disclosure, I am an AI consultant too, so if you want to see how I would answer these, the AI Strategy Sprint is a paid 3-day first step.
Anyone can call themselves an AI consultant. There is no license, no exam and no standard job description. So the quality of your hire depends mostly on the questions you ask before you sign. This page gives you eleven, with what a good answer sounds like and what a bad one sounds like. It is written by someone who sells this kind of work, so I also say where I would be the weak answer.
If you want the wider picture first (what consultants do and how they charge), read hiring an AI consultant: a plain guide for business owners. This page is the checklist you carry into the call.
Why the questions matter
The US Federal Trade Commission (the FTC, the government agency that polices unfair and deceptive business practices) ran a crackdown in September 2024 called Operation AI Comply. It announced five cases against businesses that overhyped AI or sold AI as a way to get rich. One of them marketed a chatbot as "the world's first robot lawyer" while, the FTC says, never testing it against a human lawyer's work. The FTC chair said there is "no AI exemption from the laws on the books." Those cases are mostly about consumers, not consultants, but the pattern is the same one you will meet in sales calls: big claims, no testing.
The other reason to ask is that AI projects fail in boring ways. A RAND Corporation study (RAND is a US research nonprofit) interviewed 65 experienced AI engineers and found that the most common cause was leaders and technical staff misunderstanding what problem the AI was supposed to solve. A consultant who cannot restate your problem plainly is a risk before a line of work is done.
Group 1: What you own
1. When this ends, what do I own, in writing?
Good answer: "You own the code, the prompts, the workflows and the documents. Here is the clause." Or, if they use a subscription tool, "You rent this part, here is its monthly cost, and here is what you keep if you cancel."
Bad answer: "Don't worry about that." Or an answer that mentions a "proprietary platform" you cannot leave. Ownership is where lock-in hides. Ask for the exact contract wording, and ask a lawyer to read it. Who owns AI-generated output is also a legal question in its own right: the US Copyright Office has published a report on it (Part 2, January 2025, on whether AI-generated output can be copyrighted). I am not a lawyer, so treat that as a reason to ask one.
2. Whose accounts will everything run on?
Good answer: "Your accounts. Your login with OpenAI or Anthropic (the companies behind ChatGPT and Claude), your login for any workflow tool (such as Zapier, which connects apps so one triggers another), your cloud account. I get temporary access and you can remove me."
Bad answer: "It runs on mine, it is easier." That means your process stops the day the relationship does. Make sure you are the owner of every account and that the keys are in your name.
3. What is the full cost, including tools that bill monthly?
Good answer: A number for the work, plus a list of every subscription and usage fee the system will run on, with who pays each. The usage part matters: AI services bill by how much you use them, so ask for an estimate at your expected volume.
Bad answer: A project fee and silence about running costs. Check your current subscriptions first using how to audit your AI tool subscriptions, because you may already pay for half of what they propose.
Group 2: Who has your data
4. Which AI services will see my customer or business data, and are they set to not train on it?
Good answer: A named list. For example, OpenAI says it does not train its models on data from its business products (ChatGPT Business, ChatGPT Enterprise and its API) by default, and Anthropic says the same for its commercial products (such as Claude for Work and the API), with exceptions if you submit feedback or opt in. A good consultant knows which plan you are on, because the free and personal plans follow different rules.
Bad answer: "It is secure." Ask which plan, which setting and where the policy page is. Our guide on keeping client work out of AI training has the exact questions to put to a vendor, and is it safe to upload client documents to an AI tool covers private options.
5. What data do you need from me, and what will you never touch?
Good answer: "I need sample documents with names removed, and I do not need your bank logins or your customers' payment details." They should start with the least data that works.
Bad answer: "Send me an export of everything." If the work involves health, tax or financial records, you may have legal duties about who handles them. Ask a lawyer before anything leaves your systems.
6. Will you use my business, or what you learn from it, as an example for others?
Good answer: "Not without your written permission." Put it in the contract. This also covers whether they reuse your documents to build something for a competitor.
Group 3: What is actually live
7. What will be running on the last day, and how will I see it working?
Good answer: A description of the finished thing: "On the last day, when a customer emails an invoice, this reads it, fills these three fields and flags anything it is unsure about. You will watch it run on your real emails before I leave." A demo on your own data beats a demo on theirs.
Bad answer: A deliverable called a "roadmap," a "strategy deck" or a "proof of concept" with no date for anything live. Some projects should end with a plan. That is fine, as long as you know that is what you are buying and the price matches. A plan sold as a working system is the problem.
8. Where does a person check the AI's work, and what does it get wrong?
Good answer: "It will be wrong sometimes. Here is the step where a person reviews it, here is the kind of mistake I expect, and here is how we will count the mistakes in the first month." AI output is probabilistic, meaning it varies and can be confidently wrong. The RAND study found that leaders who expected certainty lost faith when a model was merely good.
Bad answer: "It is accurate" with no number and no review step. If a consultant promises no errors, walk away.
9. Is this the simplest thing that works?
Good answer: They can explain why a plain automation (a rule such as "when this arrives, do that") will not do, if they propose an AI agent. An agent is software that decides its own next steps instead of following fixed ones; see what is an AI agent. Anthropic's own engineering guide recommends finding "the simplest solution possible" and adding complexity only when needed. Gartner, a research firm, predicted in June 2025 that over 40% of agentic AI projects will be canceled by the end of 2027, and said many use cases sold as agents do not need one.
Bad answer: Every problem gets the same fancy answer. See hiring an AI agent development company for how to judge that kind of pitch.
Group 4: After the project
10. Who fixes it when it breaks, and what does that cost?
Good answer: A named person, a response time you can write down and a price for support (or a plain "nobody, here is the documentation so another developer can"). Models and tools change, so things drift. Ask what happens when the AI provider changes a model or a price.
Bad answer: "It should not break." Ask for the handover documents too. A good handover lets someone else take over.
11. Who else could take this over, and can you show me a previous result?
Good answer: They hand you references you can phone, or they say plainly that they are new and offer a smaller first job. Do not accept a logo slide. If they are one person, ask what happens if they get sick. This is where I would be the weak answer: I am a one-person studio, so my protection is documentation, your accounts and a short first step, not a team behind me.
Bad answer: Case studies with no names, no figures you can check, or a claim like "we saved clients 10x." Vendor-written numbers are marketing until a client confirms them.
The one-page scorecard
Copy this into a note. Score each question 0 (vague or bad), 1 (partial) or 2 (clear, specific, in writing).
| Group | Questions | Max score | What a low score means |
|---|---|---|---|
| What you own | 1 to 3 | 6 | Lock-in or surprise bills |
| Who has your data | 4 to 6 | 6 | Privacy or legal exposure |
| What is live | 7 to 9 | 6 | Slides instead of a working system |
| After the project | 10 and 11 | 4 | Nobody to call when it stops |
My rule: any group scoring under half is a no until it is fixed in writing. A total under 14 of 22 means keep looking. Compare two or three candidates on the same sheet, because the sheet is what makes the difference visible.
Red flags that need no scorecard
- They promise a specific result (hours saved, revenue gained) before they have seen your process.
- They will not name the tools they plan to use.
- The first deliverable is a large document and the first working thing is months away.
- They want admin access to everything on day one.
- They criticize every other option but never say what would make them the wrong fit.
- They push you to sign this week.
When you may not need a consultant at all
If your goal is "use ChatGPT or Claude better for writing and summaries," a few hours of your own time and a team training session may be enough. Hire when you need a system wired into your real tools, when the data is sensitive, or when your team has tried and stalled. If you are weighing a consultant against hiring a developer, how to hire AI developers covers that route, and how to find the right AI agency covers bigger firms. Once you have chosen someone, the AI implementation plan shows what a sensible project looks like step by step.
How I'd approach this
I would take three candidates, put the same eleven questions to each by email so I have the answers in writing, and score them on the sheet above. I would trust the one who tells me what they will not do. I would insist on a small, fixed-scope first job, a few days to a few weeks, that ends with something I can see working, before agreeing to anything larger. A paid first step is how you learn whether the person is good at the work without betting the year on it.
If you are starting from "where would AI even help us," the AI Strategy Sprint is the first step I offer: a paid 3-day engagement where we find where AI could transform your business and generate new value, and you get the exact steps written down. It starts with a free 15-minute AI brainstorm call. Use the same eleven questions on me.
Questions people ask
How much does it cost to hire an AI consultant?
It depends on whether you hire by the hour, a fixed project or a monthly retainer, and prices vary widely by consultant and country. Ask for a fixed price for a small first project and a separate estimate for monthly running costs. Our AI consultant guide explains the three pricing models.
How do I know if an AI consultant is any good?
Check that they can restate your problem in plain words, show something working on your own data, name the tools and accounts, and describe where a person checks the AI. Phone a reference. The scorecard above turns that into a number you can compare.
Should I let a consultant put my customer data into ChatGPT?
Only on a plan whose data terms you have read. OpenAI and Anthropic both say they do not train on business-plan data by default, but personal plans follow different rules. See keeping client work out of AI training, and ask a lawyer if the data is regulated.
Do I need an AI consultant or an AI developer?
A consultant helps you decide what to build and where AI pays off. A developer builds a specific thing. Many small firms need both at small scale, often from one person. If you already know exactly what you want built, start with hiring an AI developer.
What is a fair first project to give a new AI consultant?
One process, a fixed price, a short timeline and a finished result you can see working, such as a tool that reads one kind of incoming document. Avoid starting with a company-wide transformation. A small first job tells you how they work.
Who owns what an AI consultant builds for me?
That is set by your contract, not by default, so write it down: code, prompts, workflows, documents and accounts. Ask a lawyer to read it. Ownership of AI-generated output is its own legal question, so do not assume.
More in this topic: Problem solving: AI strategy and adoption (24 guides) →
Sources
- FTC: Operation AI Comply, crackdown on deceptive AI claims (September 2024)
- RAND: The Root Causes of Failure for Artificial Intelligence Projects (2024)
- Gartner: Over 40% of agentic AI projects will be canceled by end of 2027
- Anthropic Engineering: Building effective agents
- OpenAI: Enterprise privacy at OpenAI
- Anthropic: Is my data used for model training? (commercial products)
- US Copyright Office: Copyright and Artificial Intelligence