Why AI helps some founders—and leaves others cleaning up generic copy

The difference between useful AI and “AI slop” often starts before the model answers. Two competing founders can use the same subscription and get very different results: one saves hours and builds a successful marketing campaign; the other gets forgettable copy that takes extra work to fix. The more telling difference is their prompts and what they do with the replies. Better tools alone won’t close that gap.

IBM’s guidance on prompting points to a practical lesson: a model needs a clear task, relevant context, examples and a defined audience. That’s what makes a request answerable. A vague instruction to “improve marketing” leaves the model guessing what success means.

Give the task a measurable target

Start with an outcome you can measure. A B2B software company selling to mid-market manufacturers, for example, could ask for changes aimed at lifting its email open rate from 18% to 25% over 90 days, along with metrics to track progress. That request gives the model a market, a baseline and a deadline.

Compare that with asking an AI to help a company scale or write better copy. Those are ambitions, not instructions. A prompt should identify the work to be done and the result that would count as progress. Prompting guidance can help make the request more specific, but the business owner still has to decide what the goal should be.

Supply the context that changes the answer

Useful context includes a company’s stage, customer profile, existing process, constraints and preferred format. IBM describes this as context engineering. Since a language model doesn’t retain a lasting understanding of your business, each request needs enough information to make its answer relevant.

That doesn’t mean pasting in every company document. Choose information that could change the recommendation. For a campaign, that might mean details about target customers. For a sales email, explain what those customers are trying to solve and what they’re likely to respond to. More information isn’t automatically better; the right information is.

Refine the answer, then check it

A first response may be generic. Treat that as a starting point: add examples, tighten the instructions or set clearer limits, then ask again. IBM’s description of an iterative approach is straightforward: improve the prompt and response over successive rounds until the result is clearer and more useful.

That same discipline should apply to how a company relies on AI. The NIST AI Risk Management Framework calls for managing risks and considering trustworthiness in how AI systems are designed, used and evaluated. The practical point is simple: review the output, especially when mistakes could affect customers or compliance.

AI can speed up research, analysis and routine work. It still needs human judgment when the stakes reach customers or regulators, and marketing teams shouldn’t mistake machine generated material for a substitute for their own decisions. Salesforce reported that 72% of customers surveyed considered it important to know when they were communicating with an AI agent; 71% said human validation of AI output mattered to building trust. Those figures don’t settle every question about customer attitudes, but they’re a clear reason to keep people accountable for what a business sends out.

Test one recurring task

You don’t need to overhaul your AI strategy to find out whether this process helps. Pick a repeated task—customer support drafts, sales research or meeting summaries—and use the same method for two weeks: define the goal, provide relevant context, refine the output and check the result. Then measure what changed.

If the process saves time in your workflow and delivers the outcome you set, you have a useful basis for expanding it. If not, the results may show where the task, context or review process needs work. Either way, the expensive subscription isn’t the main story. What matters is whether people know what they’re asking the system to do—and whether they verify the answer.

AI can be powerful, but people remain responsible for how it’s used.

Opinions expressed by Entrepreneur contributors are their own.

By Richard Rothschild, edited by Chelsea Brown. Oct. 8, 2026.

Richard Rothschild is Founder & CEO of Richard Roths Media in Beverly Hills, California, and an Entrepreneur Leadership Network® contributor. Read more.

X · LinkedIn · Instagram · Richard Roths Media