6 min read
35% of American consumers have already used a personal AI agent to handle a customer service request, and among Gen Z and Millennials the number is 49%. That figure comes from The Agentic Consumer Report, a survey of 1,000 U.S. consumers released by the support automation company Capacity on October 27, 2026. A second number in the same report matters even more for a one-person business: 44% said an AI tool had contradicted a company’s representative or its policies.
Put those together and you get a shift most solo owners have not planned for. Your next refund request, booking change, or “do you deliver to my area?” question may not be typed by a person at all. It may be drafted, researched, or sent by the customer’s own AI assistant, which has already read your website and formed an opinion about what your policy says. If your policy is vague, the agent fills the gap. If the agent fills it wrong, you are the one arguing with a confident customer.
This is a strategy piece about getting ahead of that, without buying anything new.
What the survey actually says about customers and their agents
The headline statistic is easy to overread, so here is the fuller picture from Capacity’s release:
- 55% have used some AI tool, such as a chatbot or agent, to help with a customer service request.
- 36% would use AI to research a return, refund, or cancellation policy before contacting a business.
- 34% say AI makes them ask more specific questions, and 23% would ask a representative to explain or cite the company’s policy.
- 20% would escalate faster when they have AI help.
- When an AI tool and a company representative disagree, consumers were roughly equally likely to trust either one.
Two caveats are worth keeping in mind. Capacity sells support automation, so it has a reason to make this trend look big. And “used an AI agent at least once” is not the same as “uses one for every purchase.” Still, the direction is clear, and the trust finding is the one to sit with. A customer whose assistant says “their site implies a full refund within 30 days” will not automatically take your word over the machine’s.
Why a solo business is more exposed than a big one
A large company has a policy team, a legal page that has been through review, and a support desk with scripts. When a customer’s agent shows up with a wrong reading, a trained rep can point to the exact clause.
A solo business usually has a single FAQ written in an afternoon, a few promises scattered across Instagram captions, and the owner’s memory of “how we usually handle that.” That is fine when every customer talks to you directly. It is a problem when a tireless reader assembles your policy from fragments. As our beginner’s guide to AI hallucinations explains, AI tools fill gaps with plausible guesses. A gap in your published policy is an invitation for a guess.
The three priorities, in order
1. Write one canonical policy page and make it boringly clear
Pick a single URL, something like yoursite.com/policies, and put every rule a customer could argue about on it: refunds, cancellations, rescheduling, deposits, delivery area, turnaround times, and what happens when something goes wrong. Use short sentences with numbers in them. “Cancel at least 48 hours before your appointment for a full refund of the deposit. Inside 48 hours, the deposit is kept.” An agent cannot misread that.
Then remove or link the contradictions. If an old blog post says “satisfaction guaranteed” and your policy says “no refunds on custom work,” a reading agent will find both and pick the one its user prefers. One source of truth beats five friendly fragments.
2. Make the page easy for machines to read, not just people
You do not need to become a developer. Three simple moves help:
- Use real headings (Refunds, Cancellations, Delivery) rather than one long paragraph. Agents, like search engines, lean on headings to find the right section.
- Put the policy in text, not an image. A screenshot of your terms is invisible to many tools.
- Add FAQ-style markup if your site builder offers it. Structured data is a small block of code that labels content so machines know what it is; Google’s introduction to structured data explains the idea in plain terms. Many WordPress, Squarespace, and Shopify themes offer an FAQ block that adds it for you.
3. Decide how you will answer an AI-assisted message
Some messages will now arrive unusually polished, oddly formal, or stuffed with quotes from your own website. Treat them the same as any customer, but with one habit: always reply by citing the canonical page. “Thanks for reaching out. Our cancellation policy is here: [link]. Since the booking is 30 hours away, the deposit is non-refundable, but I am happy to move you to another date at no charge.” That reply works whether a human or an agent reads it, and it trains the customer’s tool toward your real rule.
If a message keeps looping, asking the same thing in new words, step out of the thread and offer a phone call. Capacity’s data shows 21% of consumers would trust their agent to hand them off to a human. Make that handoff easy, because an AI talking to your AI assistant can go in circles politely forever.
Where your own AI tools fit
None of this requires you to put a chatbot on your site. If you already use AI to draft replies, give it the same canonical policy page as its reference, and nothing else. Our AI autonomy ladder is a good framework here: drafting policy replies is a sensible rung for AI, but sending refunds or exceptions should stay with you.
If you are considering an AI receptionist for calls, the same rule applies. Our look at AI phone answering at $4.80 an hour found the tools work best when they read from a short, unambiguous script. Your policy page is that script.
What to watch over the next year
Expect three developments. First, more customer agents will act, not just research: booking, cancelling, and requesting refunds directly through forms. Second, platforms like Shopify, Square, and booking tools will likely add settings that tell agents what they may and may not do on your behalf. Third, disputes will increasingly hinge on what your website said on a given day, so keep a dated copy of your policy page each time you change it. A simple PDF export into a folder is enough.
The broader lesson is that your website is turning into an instruction manual read by software. The businesses that write clear manuals will spend less time arguing.
Your one-hour version of this plan
- Spend 20 minutes gathering every promise you have made publicly about refunds, cancellations, and delivery.
- Spend 20 minutes rewriting them into one page with headings and numbers.
- Spend 10 minutes deleting or linking the contradicting fragments.
- Spend 10 minutes saving a dated PDF and a reply template that cites the page.
Then ask your favorite AI assistant, “What is this business’s refund policy?” and paste in your homepage URL. If its answer surprises you, a customer’s agent will be surprised too. Have you received a message that looked AI-written yet? Tell us how you handled it in the comments.
