How to Hire a Generative AI Consultant: A Brand's Vetting Checklist

There is no license behind the title "generative AI consultant." That makes the buying process evidence-heavy by necessity.

The goal is not to find the person who can name the most models. It is to find someone who can understand your problem, expose the important tradeoffs, produce relevant evidence, and leave you with a clear result.

Use this checklist on any candidate—including me.

1. Write the buying problem before the job title

"We need AI" is not a brief. Write down:

• +The business or audience outcome.
• +The workflow or asset you want to improve.
• +What happens today and why it is insufficient.
• +The intended channel, users, or operating team.
• +Deadline, budget range, and internal owner.
• +Brand, legal, factual, security, or technical constraints.
• +What evidence would justify expanding the work.

This prevents candidates from reshaping the problem around whatever they already sell.

2. Match the candidate to the actual category of work

A strategy advisor, machine-learning engineer, automation builder, and AI video director are not interchangeable.

For creative work, look for the same type of finished output you need. A strong micro-drama portfolio does not automatically prove product-ad craft; a polished avatar demo does not prove cinematic direction. The case-study archive and AI video watch pages show how I separate formats and evidence on this site.

For systems work, ask to see the workflow boundaries: inputs, decisions, human reviews, outputs, and failure handling. A screenshot of a tool is not evidence that a repeatable operating process exists.

3. Ask the candidate to explain one project

A useful walkthrough should answer:

1. 01What was the original brief and constraint?
2. 02What did the candidate personally own?
3. 03Which approach failed or was discarded?
4. 04Where did human judgment change the output?
5. 05What was delivered?
6. 06Which results were measured?
7. 07Which results were not available?

The last question matters. Production evidence can prove that work was completed to a stated scope. It cannot prove revenue, ROAS, retention, or lift unless those metrics were actually observed and attributable.

4. Test whether the recommendation follows the problem

Ask: "Where should we not use generative AI in this project?"

A credible answer may identify tasks where conventional production, deterministic software, or a human specialist is more reliable. It may also identify a narrower use of AI—concept exploration or selected shots rather than the whole workflow.

Be cautious when every problem receives the same model, the same automation pattern, or the same content package.

5. Define a production-representative pilot

A pilot is useful only when it tests the real risks. For an AI video engagement, that could include product fidelity, character consistency, voice, runtime, aspect ratios, edit quality, brand review, and delivery format. A five-second spectacle may not test any of those.

Write down before starting:

• +The hypothesis.
• +The smallest realistic deliverable.
• +Acceptance criteria.
• +Reviewers and review rounds.
• +Inputs the client must provide.
• +What will be documented at the end.
• +The decision the pilot will unlock.

The pilot can fail and still be valuable if it resolves an important uncertainty honestly.

6. Compare scopes, not headline rates

Two quotes are not comparable until they describe the same output and responsibilities. Confirm:

• +Deliverables and variants.
• +Concept, script, generation, and post-production ownership.
• +Number and type of revisions.
• +Usage and licensing responsibilities.
• +Source files, prompts, references, and workflow documentation.
• +Meeting and handoff expectations.
• +Dependencies, exclusions, and change control.
• +Timeline and approval responsibilities.

The consulting budget guide provides a scope-first way to evaluate cost without pretending every engagement has a universal market rate.

7. Agree on ownership and operating access

"You own everything" can still be vague. Specify which assets and accounts are included:

• +Finished outputs and project files.
• +Prompts, reference images, and configuration.
• +Code, automations, and documentation.
• +Third-party model accounts and paid licenses.
• +Reusable templates versus the consultant's pre-existing tools.
• +Rights to voices, likenesses, music, stock, and brand material.

The right arrangement depends on the project. What matters is that it is explicit before work begins.

Red flags

• +A portfolio made mostly of tool interfaces or isolated generations.
• +Business-outcome claims without a measurement method or access to the data.
• +Guaranteed virality, rankings, or performance.
• +No answer for where the proposed workflow is likely to fail.
• +A pilot that avoids the hardest production constraint.
• +Unclear rights, handoff, or third-party licensing.
• +A quote that omits review rounds, dependencies, or exclusions.
• +Pressure to scale before the first real output is evaluated.

Green flags

• +Relevant finished work and a clear explanation of personal contribution.
• +Candid separation of known results from unavailable data.
• +A plan that starts with the business constraint, not a model name.
• +Specific human review and quality-control points.
• +A small, realistic first test with an explicit expansion decision.
• +Documentation and ownership terms suited to how your team will operate later.

A simple final scorecard

Score each candidate from 0 to 3 on relevance, evidence quality, problem diagnosis, pilot design, review discipline, handoff clarity, and working fit. Reject anyone who fails a critical rights, safety, or truthfulness requirement regardless of total score.

If you want to evaluate my work, start with the generative AI consulting approach, then inspect the linked proof. If the scope is already a finished video, use the AI video creator brief instead.