Small businesses are adopting AI in the places where customer promises are made: sales emails, proposals, website copy, chat support, appointment booking, delivery estimates, and refund conversations. That is where the technology can save the most time. It is also where a plausible sentence can quietly become a costly commitment.
The danger is not limited to a chatbot inventing a fact. A larger risk is promise drift. One employee asks an AI tool to make a proposal more persuasive. Another uses it to draft a product page. A third turns the same language into an email campaign. By Friday, an unsupported claim has spread across four customer touchpoints, and nobody can say who approved it.
Small Business Currents recently documented how AI is already changing the balance of power in customer disputes, making it easier for people to produce polished complaints and legal-sounding demands. That should sharpen an owner’s attention to the evidence behind every promise the business makes. When a customer challenges a claim, the relevant question is not whether the copy sounded reasonable. It is whether the business can show what the promise meant, what supported it, and who had authority to make it.
A recent Federal Trade Commission settlement offers a blunt reminder. The agency said Air AI and its owners misled entrepreneurs and small businesses with claims about business growth, earnings potential and refund guarantees. The lesson is broader than a single company or a single sales pitch: AI does not lower the standard for substantiation. It makes disciplined proof more important because persuasive language is cheaper to produce and easier to repeat.
The practical answer is not a thick compliance manual. Most owners do not have the staff or time for one. They need a lightweight customer-promise register.
Start With the Promises That Can Hurt
A customer-promise register is a shared record of claims that could materially affect a customer’s buying decision or expectations. It does not need to capture every adjective in every social post. Start with statements about price, savings, delivery, availability, qualifications, performance, refunds, guarantees, privacy, security, and the limits of a service.
For each important promise, record six things:
- The exact claim customers will see.
- The evidence that supports it.
- Every channel where it appears.
- The person authorized to approve or change it.
- The correction path if the claim proves inaccurate.
- The date and result of the next review.
Consider a local service business that uses AI to draft estimates. The tool may turn “most jobs are completed within a week” into “your project will be finished in seven days.” Those sentences feel similar, but they create different expectations. The register forces the owner to decide which sentence the business can actually support, whether exceptions need to be stated, and who can authorize a tighter commitment.
The same discipline applies to a retailer using AI-generated product descriptions. “Designed to reduce energy use” is not the same as “cuts your energy bill by 30%.” A professional firm describing its expertise should distinguish between experience, certification, and guaranteed outcomes. A subscription business should ensure the refund language in a chatbot matches the actual policy customers will encounter when they try to cancel.
Make the Register Part of the Work
The register will fail if it lives in a folder that no one opens. It should sit inside the process where claims are created and revised. A spreadsheet can work. So can a table in the customer-relationship system or project platform. The important part is that the record is easy to find before language goes live.
Give employees a simple rule: if AI creates or materially strengthens a customer-facing claim, check the register. If the claim is not there, add it or ask the owner responsible for that category. This is not a ban on fast drafting. It is a pause at the point where draft language becomes a business commitment.
A weekly 15-minute review is often enough for a small company. Look at newly added claims, customer complaints, refund requests, and corrections made during the week. Ask whether one inaccurate statement has migrated into other channels. If a website claim was corrected, was the sales deck corrected too? Did the chatbot continue using an old answer? Did a reseller or contractor copy the language?
This is where the register becomes more than a defensive tool. It reveals recurring gaps in the business. If employees repeatedly overstate delivery speed, the underlying problem may be weak scheduling data. If refund language keeps drifting, the policy may be too complicated. If nobody can support a performance claim, the company may need better measurement before it needs better marketing.
Reward the Person Who Finds the Problem
Owners often say they want employees to speak up, then react badly when someone catches a risky claim before launch. That teaches the team to stay quiet. The better signal is to praise the correction, especially when the language came from a tool that appeared confident.
The National Institute of Standards and Technology’s AI Risk Management Framework emphasizes ongoing governance, measurement and management rather than one-time approval. Small businesses can apply that principle without creating a new department. Every correction should improve the register, the prompt, the source material or the approval rule so the same error becomes less likely.
Customers rarely expect perfection. They do expect a business to know what it promised and to correct mistakes without making them fight through a maze. A register gives employees the evidence and authority to do that quickly.
AI can help a small company sound larger, faster, and more polished. That advantage becomes a liability when polished language outruns operational reality. The businesses that benefit most will not be the ones that generate the most copy. They will be the ones that can trace every important promise from evidence to approval to customer outcome.
Gleb Tsipursky, PhD, is a behavioral scientist, the CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026).
Photo courtesy A Chosen Soul for Unsplash+
