AI Makes It Easy to Look Like an Expert. Then the Implementation Falls Apart.

A 50-slide strategy deck used to take weeks of work. Now it takes an afternoon. So does a website, a market analysis, a content calendar, and a confident-sounding recommendation about where to spend your budget. AI has made it easier than ever to look like an expert. The trouble starts when someone has to make the work actually function: the website has to track leads, the campaign has to count the right conversions, the plan has to survive contact with your team, your data and your budget. That is where the gap between looking like an expert and being one shows up, and by then you have already paid.

I use Claude and ChatGPT every day. They make me faster at analysis, reporting and working through large amounts of information. But I also spend real time checking and correcting work, from AI and from the people using it, that looks completely convincing on the surface.

1. What changed?

For decades, buyers judged expertise by its signals: polish, volume, speed, and confident language. A beautiful deck suggested a serious firm. A long report suggested deep work. Those signals are now cheap. Anyone can produce them, whether or not they understand the problem. I wrote about the same shift in advertising in The Polish Paradox. When polish costs nothing, it stops telling you anything.

2. What does fake expertise look like in marketing?

Here is what I run into most:

• The beautiful deck with no point. Fifty slides of frameworks, charts and trends, and nowhere does it state the business problem or what to do about it.

• The website that looks finished. Three of my clients recently went through website rebuilds with agencies that said they could handle both design and development. In each case, the finished site raised serious implementation concerns. AI-assisted development appeared to be part of the process. The technical expertise and quality control behind it were missing.

• The analysis that sounds right. A B2B SaaS company I worked with had paid another firm for a sophisticated marketing mix model and was about to make budget decisions with it. An independent review found structural problems that could have pointed them in the wrong direction. More on what to look for below.

• The dashboard nobody uses. Lots of charts, connected data, and no decision that changed because of it.

• The content that sounds like everyone else's. Technically clean, and interchangeable with every competitor.

3. Why does it fall apart in implementation?

Producing a deliverable and making it work in the real world take different skills. AI can draft the plan, the copy, the code and the slides. Implementation takes knowing how the platforms behave, where the data is wrong, what breaks on mobile, which settings quietly cost money, and what your business can realistically execute. That knowledge comes from having done the work and fixed it when it went wrong.

The early output is good enough to pass a quick look, and a quick look is how most of us review work. The deck reads well. The site loads. The model has a methodology slide. Then the forms don't send leads to your CRM, conversion tracking counts the wrong action, the site is slow and hard to update, or the budget moves in the wrong direction. You end up paying once for the work and again to fix it.

4. What does a weak AI-built marketing mix model look like?

Marketing mix modeling is a good example, because the output looks impressive either way: coefficients, charts, and a confident return for every channel. Free tools and AI assistants now let almost anyone produce one. When I review a model, this is what is often missing:

• Only one or two channels. If the model includes Google and Meta and nothing else, it will hand them credit for sales driven by everything else you do.

• No outside variables. Seasonality, pricing, promotions, competitor activity, the economy and changes in distribution all move sales. Leave them out and the model assigns their effect to marketing.

• No PR, influencer or creator data. Press coverage, creator posts and word of mouth drive demand. If the model has no measure of them, paid media takes the credit.

• No list of the statistical tests run on the data. A credible model shows how it was checked: how well it predicts data it was not built on, whether the variables are too closely related to separate, and whether the results hold when the model changes slightly. If none of that is listed, you have no way to know whether to trust it.

• Correlation treated as causation. Spend and sales rising together does not prove the spend caused the sales. Budgets often go up because sales are already going up.

• No split between local and national. National campaigns, regional efforts and local search behave differently. Lumping them together hides what is working where.

5. Seven questions that separate expertise from polish

Real experts answer these easily. People leaning on AI usually can't.

• "What would make you change this recommendation?" Experts know the conditions under which they are wrong.

• "What did you leave out, and why?" Judgment shows in what someone chose to cut.

• "Walk me through one number on this slide." Pick any figure. Where did it come from, and how much would you trust it?

• "What is the biggest risk in this plan?" A plan with no stated risks has not been thought through.

• "What have you seen go wrong when companies do this?" This asks for experience, which AI cannot supply.

• "How will we know in 90 days whether this is working?" Look for a measure tied to revenue, leads or margin.

• "Which parts did AI help with, and who checked them?" Using AI is fine. Unchecked AI is the problem.

6. Should you avoid consultants and agencies who use AI?

No. The good ones use it, and you benefit from the speed. The question is whether the person can explain the work, defend it, and fix it when it's wrong. Ask for that, and you will learn more in a 30-minute conversation than from any deck.

7. How should this change the way you hire marketing help?

Spend less time on the pitch materials and more on the conversation. Ask someone to talk through how they would approach your problem, with no slides. Start with a small paid project before a long contract. When you call references, ask what changed in the business, beyond whether they liked working with the person. For more on this, see How to Hire a Fractional CMO in 2026.

AI is making it easier than ever to look like an expert. Implementation is where you find out who actually is. If you want a second opinion on a plan, a deck or an agency's work, book a free 30-minute call.