How to spot fake AI consultants: a ten-minute check before the first call.

Before you take a call with an AI consultant, do three checks. Make sure that the person is real. Make sure that the person built something. Make sure that you know who pays the person. These checks take about ten minutes. They remove most bad offers.

A fractional CFO said this clearly in a short video in 2026. He gets many messages on email and LinkedIn from people who offer AI support. Some of these “people” are AI bots. They collect a “yes,” then they sell the lead to somebody else. His advice: work with people who you trust, learn their method, check references, and make sure that the reference is a live person.

This page turns that advice into a checklist.

Why did this problem grow in 2026?

Three changes made fake AI consultants more common.

The demand grew quickly. Many companies want help with AI, and few people have production experience. When demand is high and supply is low, many new sellers enter the market.

The cost of outreach fell to almost zero. An AI tool can write thousands of personal messages each day. It can also make profiles, photos and follow-ups. One person can now send the volume of messages that a sales team sent before.

The title has no standard. Anybody can call themselves an AI consultant or a fractional Chief AI Officer. No certificate proves that a person can put an AI system into production.

The result: you get more messages, and each message tells you less.

What types of fake AI consultant are there?

Not all of these are dishonest. A lead reseller can find you a good agency. A generalist can be excellent at change management. But you must know which type you have before you pay.

Check 1: Is the person real? (3 minutes)

Do these checks on the profile and the message.

• Search the name with the company name. A real consultant has a company website, a history of posts or talks, or public code.
• Look at the age of the profile. A profile that started this year with many connections and few posts is a warning sign.
• Look at the photo. Generated photos often have strange backgrounds, ears, glasses or text. Search the photo with a reverse image search.
• Read the message. Bot messages are general. They use your name and your company name, but nothing that shows that they read your website. They ask a yes or no question to start a sequence.
• Reply with a specific question. Ask: “Which AI system did you build that is like our problem?” A bot replies with a general text or a booking link. A real person replies with a real answer.

Check 2: Did the person build something? (4 minutes)

Look for public evidence of real work.

• Code. A GitHub or GitLab profile with real projects. Look at the dates and the commits. Projects copied from tutorials are not strong evidence.
• Write-ups. Articles that describe a real system with details: the data, the model, the cost, the failures. General articles about “the future of AI” are not evidence.
• Talks. Conference talks or podcasts where the person explains a system that they built.
• Products. A product or tool that the person built and that people use.

If you find no evidence, ask for it. The hiring guide has the exact message to send.

Check 3: Who pays the person? (3 minutes)

Look at the website and the profile for these signs:

• Partner badges. A partner badge from a platform vendor means that the person gets benefits from that vendor. This is not bad. But the advice can lean toward that vendor.
• One product in all case studies. If all examples use the same platform, the person probably resells it.
• Free audits. A free “AI readiness audit” with a high stated value is usually a sales tool. The audit ends with a proposal.
• No price information and no scope. The person will price after the call, based on what you can pay.

Then ask directly: “Do you get referral fees, commissions or discounts from any vendor that you recommend?” A real consultant answers this question without problems.

What does a bot outreach sequence look like?

Many automated sequences follow the same pattern. Learn it, and you can delete most messages quickly.

1. Connection request with a short note: “I help companies like yours with AI.”
2. First message after you accept: a question about your AI plans, with your company name.
3. Follow-up after two or three days: a statistic about AI, for example “most AI pilots fail.”
4. Second follow-up: an offer of a free audit or a free call.
5. Last message: “I will stop here. Reply if you change your mind.”

A real consultant can also use templates. But a real consultant replies to a specific question with a specific answer.

What questions find a reseller?

Ask these questions on the first call:

• “Will you personally do the work?”
• “Who else will work on this? Can I meet them now?”
• “Where are the people who do the work? Are they employees or subcontractors?”
• “Will the code be in our repository from the first day?”

A reseller often avoids these questions. The reseller says “our team” or “our partners” and cannot name the people.

What questions find a rebranded generalist?

Ask these questions:

• “What did you work on before AI?”
• “Which AI system did you put into production? What was its cost per task?”
• “How did you measure the quality of the output?”
• “What is an eval? How do you build one?”

A generalist can be honest and useful. Some generalists are good at training and change management. But do not hire a generalist for architecture, model choice or security.

What questions find a platform seller?

Ask these questions:

• “If we already use another platform, can you work with it?”
• “Can you recommend a solution without a platform, for example a script and an API?”
• “What does the platform cost for three years, including all users?”
• “What happens to our data and prompts if we stop the platform?”

A platform seller always comes back to the same platform. A neutral advisor can recommend “no platform” when that is the correct answer.

How do I check references?

References are the strongest check. Do them correctly.

1. Ask for two references from clients with a system in production.
2. Find each reference yourself on the company website or on LinkedIn. Do not use only the contact details that the consultant gives you.
3. Call the reference. Do not only send an email. A bot can reply to an email.
4. Ask: “What did the consultant deliver? Is it still in use? Who maintains it?”
5. Ask: “Would you hire this person again for the same work?”

If the consultant cannot give a reference that you can verify, stop.

What are the signs of a real expert?

These signs together show a high probability of a real expert:

• Public work with details: code, write-ups or talks about real systems.
• Specific answers about failures, cost and quality.
• Questions about your data and your process before any recommendation.
• A recommendation to start small, or to not build something.
• A clear statement of conflicts of interest.
• A contract where you own all code, prompts and evals.
• References that you can call and verify.

Which claims need a second question?

Consultants use some claims often. Each claim needs one more question.

A real expert answers each question with details. A fake consultant moves to the next topic.

The ten-minute checklist

Copy this list. Use it for each consultant before the first call.

Two or more fails: do not take the call.

What if I already work with a consultant who fails these checks?

Do not panic. Check the work, not only the person.

1. Ask for all code, prompts and documents to be in your accounts now.
2. Ask for a list of all vendors and costs.
3. Ask for a short written report on what was delivered and what is in production.
4. Get a second opinion on the most important system. An independent review takes one to two weeks.

The AI agency handover checklist lists what you must own.

Next step

Use the checklist on the next three messages in your inbox. For a full selection process, read how to hire a fractional Chief AI Officer. To compare an independent consultant with an agency, read independent AI consultant or agency?