AI still needs your discernment

To tell if AI marketing advice makes sense for your service business, check what it assumes, compare it with how your business actually operates, and test a manageable change before committing. AI can help you think through a decision. You still need to decide whether its answer holds up.

By Eric Yim, Protocol Text, a marketing services business based in Lincoln, Nebraska.

Why I think AI's value depends on your judgment

I think AI is super overhyped. I also think there's a real advantage available if you bring high critical thinking and a lot of discernment to the outputs. You can really amplify your skills when you can recognize what's useful and what needs work.

Discernment means judging an answer instead of accepting it because it sounds confident. Does it fit the work you want? Can you deliver what it suggests promising? What information would change the recommendation?

A polished answer can still depend on something that isn't true. A marketing plan that assumes you have someone answering calls all day may fall apart if you're doing the work yourself. Your experience matters because you know the practical details the answer may have skipped.

Give AI enough context to make a useful recommendation

I think people overcomplicate learning AI. You can start by chatting with it. But if you want advice you can use, give that conversation some boundaries.

Describe the services you actually offer, the area you serve, the work you want more of, and the time you can spend following up. Include constraints, such as a full schedule or a service you don't want to promote.

Then ask for the reasoning: “Based only on these details, what would you recommend first, and why? Separate facts I've given you from assumptions you're making. Tell me what you still need to know.”

That gives you something specific to review. If it recommends a campaign based on staff, availability, or services you never mentioned, you can correct the premise before discussing the campaign.

A worked example: catch the promise before you publish

Here's an illustrative example, not a client result. Suppose a cleaning business wants more recurring appointments. The owner works alone and schedules visits in advance. AI suggests a page advertising “same-day cleaning, guaranteed availability, and the lowest prices in town.”

The writing might sound ready to publish, but the recommendation has several problems. Same-day availability wasn't supplied. Guaranteed availability conflicts with a schedule that needs checking. The price claim has no support. The offer also points toward urgent bookings instead of the recurring work the owner wants.

I would correct it directly: “We offer scheduled cleaning and want more recurring appointments. We don't promise same-day service, guaranteed availability, or lowest prices. Remove those claims. Write a draft using only these facts, and flag anything else you need me to confirm.”

The revised draft still needs review. A more suitable next step might invite someone to ask about recurring appointments. The owner should confirm that the inquiry process actually supports that request before putting it on the page.

Use a review process before spending or publishing

My suggested process is to review the recommendation in separate passes. Trying to judge accuracy, writing, and business value all at once makes it easier to miss something.

1. Check the facts. Verify services, locations, availability, credentials, and promises. Remove claims you can't support. Don't let a rewrite quietly introduce new ones.
2. Check the fit. Ask whether the advice attracts the work you want and whether someone has time to carry it out. A plan can be reasonable and still be wrong for your current capacity.
3. Check the reasoning. Ask what assumptions drive the recommendation and what would make another option better. Compare those assumptions with your own records and experience.
4. Choose a limited test. Define what you'll change, who will handle it, and what you'll watch. Avoid changing everything at once if you want to understand what helped.

For a service page, finish by reading it as a prospective customer. What would someone reasonably expect after reading it? If the page creates an expectation you can't meet, better wording alone won't solve that.

Keep your experience inside the process

I think businesses will find it harder to compete if they don't incorporate AI into some of their processes. That doesn't mean the customer-facing experience has to be AI driven.

You could use AI to compare drafts, organize ideas, or challenge a plan while keeping the final decision with someone who understands the business. Start with work you can review competently. Be more cautious when the recommendation reaches outside your knowledge.

I also think the way a product feels and looks matters. Real experiences and thought should shape what you put in front of people. Generating more versions won't help much if nobody takes the time to decide which version deserves to be used.

Frequently asked questions

What if two AI tools give me conflicting marketing advice?

Compare their assumptions rather than taking a vote. One may be prioritizing more inquiries while another is prioritizing a particular kind of job. Ask each to explain its objective, required resources, and missing information. If the disagreement depends on facts you don't have, gather those facts before choosing.

Should I give AI customer messages or business records for context?

Start with an anonymized summary and only the details needed for the task. Remove names, contact information, addresses, and private job details. Review the tool's data settings and terms before sharing sensitive material. You can often explain a marketing problem without uploading the original customer record.

How should I judge whether an AI-assisted marketing change helped?

Choose a measure tied to the decision, such as inquiries for the work you want and how many become bookings. Record the starting point and what changed. Consider seasonality, capacity, and other marketing changes before crediting AI. Decide when you'll review the test rather than expecting immediate results.