AI marketing consultant: what to ask before you hire one

What an AI marketing consultant should actually do

A good AI marketing consultant picks one repeatable marketing task, changes how it is done, and hands you something you can run without them. That is the whole job.

GrowthRocks, a growth agency, makes a useful point in its framework for evaluating AI marketing agencies. It splits "AI-powered" into four tiers. Drafting with ChatGPT or Claude is table stakes. Custom agents and AI search optimization are real differentiators. GrowthRocks also says that "An agency operating honestly at tier two beats an agency claiming tier four vaguely, every time."

I agree, and I would add a rule. Match the tier to your bottleneck. If your problem is slow content production, you need tier one or two done well. If AI answers are replacing your clicks, you need someone who measures AI search visibility. Paying for tier four when you have a tier one problem is a waste, and it is easy to do when the pitch sounds impressive.

Four questions to ask an AI marketing consultant

Four questions do most of the work. Who does the work, how will you measure it, what do I keep, and what will you not promise.

Archer Web Strategies lists twelve questions to ask any AI consultant. It says "No credible provider guarantees specific rankings or AI citations, because engines change and off-site signals depend on third parties." That is the cleanest test of honesty in this market. A consultant who guarantees citations is guessing or selling.

Dan Cumberland Labs, in its guide to questions before you sign, flags a common trick. Senior people sell the engagement and junior people deliver it. Ask for the name of the person who will do the work, and ask what that person did last month.

Both guides stop at the questions. Here is what they skip, a way to score the answers.

• Name: did they give a person, not "our team of experts"?
• Metric: did they pick a number tied to your business before work starts?
• Exit: did they say what you own on the last day?
• Limit: did they say what they will not do?

Four yes answers is a hire. Two or fewer is a pass. Score each consultant on a single sheet so you compare answers, not impressions.

Red flags in an AI marketing consultant pitch

The strongest red flags are guaranteed placement, "proprietary AI" with no detail, and reporting with no method. Each one tells you the consultant is selling a story.

GrowthRocks adds two more worth keeping. One is no disclosure policy on AI-generated work, which leaves you holding the reputational risk. The other is headcount reduction as the headline benefit. Its view is that this kind of pitch sells cost, not growth.

I would add one the guides miss. Watch for a consultant who needs nothing from you. Real marketing work needs your facts, your approvals and your judgment about voice. Archer makes the same point, noting that "Nothing" is the wrong answer when you ask what they need from your team. If you are one person approving the work, say so early and see whether they plan around it.

What an AI marketing consultant costs

Prices vary so widely that a quote means little until you know the tier and the scope. Compare scope first, then price.

Dan Cumberland Labs states that AI consulting rates range from $100-$150/hour for junior consultants to $300-$500+/hour for senior specialists. GrowthRocks publishes a table in which a GEO programme runs 4,000 to 15,000 USD monthly. Those figures come from other firms and from enterprise-leaning work, so treat them as context, not a quote you should expect.

The better question is what sits outside the fee. Ask about API costs, automation platform licenses and integration work. A consultant who says there will be no extra costs is rarely right.

What you should own when the engagement ends

You should own the deliverables and the documentation, and any ongoing support should be a separate decision with its own scope. If the consultant cannot say that in writing, you are renting.

GrowthRocks puts the risk plainly. If a consultant owns every prompt, agent and fine-tuned artifact, you are renting a capability rather than building one. That can be fine if the price reflects it. It is a problem when nobody mentioned it.

This is how we handle it at NorthSignal. Jake Bauman, founder, runs every engagement himself, with no account team between you and the work. Inputs, output, timing and price are agreed in writing before paid work starts. You keep the agreed deliverables and documentation, and ongoing delivery is its own scope. You can read the full approach on the AI marketing consultant page.

How to test an AI marketing consultant before you commit

Ask for a small, bounded sample on your own material before any larger engagement. A sample shows voice, relevance and quality faster than any reference call.

References still matter, and both Archer and Dan Cumberland Labs recommend asking for a client of similar size and industry. A sample adds something references cannot, which is evidence on your offer. Our process starts with a free thirty-minute call where you bring one bottleneck and we choose one repeatable task. After the call I prepare a bounded sample from public material so you can judge the work before discussing more.

If your bottleneck is search or AI visibility specifically, the SEO and AI search consulting service covers search access, indexing signals and priority pages. It does not guarantee rankings or AI citations, and no honest provider does.

A decision rule for choosing an AI marketing consultant

Hire the consultant who gives a named person, a metric, an exit plan and a stated limit, and whose sample you would publish with light edits. Run it in four steps.

• Pick one marketing task you want changed.
• Ask each candidate who does the work, how success is measured, what you keep and what they will not promise.
• Mark each answer yes or no for name, metric, exit and limit. Four yeses is a hire, and two or fewer is a pass.
• Request a bounded sample on that task from your top two candidates, then choose the one you would publish with light edits.

How this post was made: drafted with AI assistance (Claude Sonnet 5.5) from the sources linked above, then checked by Claude Opus 5.5 and an automated quality gate that matched every figure and quotation to its source on 2026-11-02.