Most AI copy sounds polished but gets your business wrong
A paragraph can sound confident and still describe a service you do not offer. That is the risk with using AI to write about your business from a thin brief.
The US National Institute of Standards and Technology’s Generative AI Profile describes “confabulation” as generated content that is presented confidently but is erroneous or false. NIST notes that the risk can matter in open-ended and longer-form tasks. That does not mean every AI draft is wrong. It means smooth writing is not proof that the facts are right.
How an inaccurate detail sneaks in
Say a brief says, “We are a plumbing company in Johannesburg.” It does not say which suburbs the team serves, whether it takes emergency calls, what jobs it handles or when it is available. A model may still produce copy that sounds complete, such as “24/7 emergency plumbing across Johannesburg”. Unless those details came from the business, that line is an invention.
The same problem can affect pricing, qualifications, turnaround times, guarantees, customer results and the steps in a sales process. Some claims are easy to miss because they sound like ordinary marketing language. “Same-day service” is still a factual promise. “Trusted by leading companies” still needs evidence.
Give the model a source of truth
Before drafting, collect the facts a customer would use to decide whether to contact you. Include the exact services, service area, operating hours, price or quote process, who the service is for, what is excluded and what happens after an enquiry. Add qualifications and proof only when you can verify them. A short approved fact sheet is more useful than a folder of vague old copy.
Give the writing tool a clear instruction: use only the supplied facts; do not invent missing details; flag any question the brief cannot answer. OpenAI’s prompt-engineering guidance describes how instructions and context shape generated responses. Context helps the model work from the material you provide, but it does not independently verify that material.
Review the claims, not just the grammar
Read the draft with the business owner or someone who knows the work. For each sentence, ask:
- Can we prove this detail from an approved source?
- Does the wording make a promise about price, timing, availability or results?
- Does it describe the right customer, service and area?
- Has a qualification or limitation disappeared during rewriting?
If a claim cannot be checked, remove it or rewrite it more carefully. Do not leave an invented statistic in place because it makes the page sound stronger. NIST recommends reviewing and verifying generated sources and citations as part of risk management; the same habit is useful when reviewing business copy.NIST Generative AI Profile
Use AI for the work it can support
AI can help organise a rough outline, turn approved notes into a first draft, suggest clearer headings or adapt a paragraph for different audiences. A person still needs to own the facts and decide what the business is willing to promise.
This matters most for regulated brands and services where a wrong claim can mislead a customer or create compliance risk. Use approved language for qualifications, regulated claims and customer-facing commitments. A human review is part of the writing process, not a final spelling check.
A sensible next step
Before your next page goes live, compare it with a one-page fact sheet for your business. Check the service names, areas, prices, process and proof. If those facts are already scattered across your site, share the business name and website with me; I can help identify where a customer might get the wrong impression.
FAQs
Can I trust AI to write accurate website copy?
Use it to draft, but verify every business-specific claim. A fluent sentence does not show where its facts came from.
What should I give an AI writing tool?
Provide approved details about the offer, audience, service area, process, limits and proof. Tell it to flag gaps instead of guessing.
Does AI copy always need a human review?
If it describes your business, offer or customers, a person who can verify those facts should review it before publication. The higher the cost of a wrong claim, the more careful that review should be.
Sources: NIST AI 600-1, Generative AI Profile, OpenAI prompt-engineering guidance.