Your business hours changed. A service is no longer offered. An old document still says customers receive a reply within two hours. If those details are scattered across notes, saved prompts and website pages, an AI draft may repeat the wrong version.
For a small service business in Greenville, Spartanburg or the surrounding Upstate, a useful starting point is one short, owner-approved reference sheet. It tells the person preparing a draft which facts apply, where they came from and what still needs a decision. It reduces guesswork; it does not guarantee an accurate answer.
Choose who approves the facts
Start with one workflow, such as drafting answers to routine service inquiries. Name the person who can approve business details for that workflow. A newer file is not automatically the right one: it could be an unfinished draft, an old document saved again or a policy that does not take effect until next month.
Keep the approval date and effective date separate. If two sources disagree, the owner or responsible manager must decide which applies. Do not ask AI to settle the conflict by choosing the newest-looking document.
Build a small reference sheet
Use a shared document or spreadsheet your team already knows. Add these fields for each detail:
- Topic: the question this fact answers, such as opening hours or whether a service is offered.
- Approved wording: the exact fact a draft may use, including limits and exceptions.
- Scope: the service, location or type of request it applies to.
- Source: a link or record reference, with the approving person's role.
- Dates: when it was approved, when it takes effect and when it should be reviewed or expires, if applicable.
- Status: approved, awaiting a decision or superseded. Keep unknowns visible.
Include only the details needed for this workflow. The FTC's business data-security guide recommends retaining only information needed for business purposes and limiting access. A reference sheet for routine service answers does not need customer addresses, payment details or account passwords.
Work through one policy change
Hypothetical example, not a Zasper customer case: a small repair business has an old handout that says, “We call back within two hours.” The owner approves new wording for routine inquiries: “Our office reviews routine inquiries on business days. Callback timing depends on the request.” It takes effect Monday. No exact callback deadline is approved.
The reference sheet should contain the new wording, its effective date, its routine-inquiry scope and the owner's approval. The old two-hour claim is marked superseded from that date. The business also identifies where the old claim appears so the responsible person can update those materials.
After the change takes effect, a useful draft might say: “Our office reviews routine inquiries on business days. What service are you asking about?” It should not promise a call today, offer an appointment or change the wording to “within one business day.” Those commitments were never approved.
If the owner has not confirmed whether the policy applies to warranty requests, mark that scope as awaiting a decision. A draft about a warranty request should go to the responsible person rather than borrowing the routine-inquiry rule.
Give the drafter a clear instruction
Try this with fictional information first. Supply the relevant approved rows and the date the draft is being prepared:
Prepare a short customer-response draft using only the approved business facts supplied below. Check that each fact applies to this request and is effective on the supplied date. Do not treat a newer date as proof of approval. If sources conflict, a needed fact is unknown or the request falls outside the stated scope, list the issue for human review. Do not invent hours, prices, service availability, deadlines or promises. After the draft, list the reference rows used and any unresolved points. Do not send the message.
Compare the result with the sheet and the original approval. A reference listed by AI is a claim to check, not proof that the answer is correct. NIST's Generative AI Profile, section 2.2, describes the risk of confident but false output. Keep human review in the workflow, especially where the wording commits the business to something.
Update the place the workflow actually reads
Changing the master document is only part of the job. Find out what the drafting process actually uses: a pasted excerpt, an uploaded file, a saved prompt or a connected source. Check your tool's current documentation and setup rather than assuming the change has reached it.
Keep superseded material available for your business's appropriate recordkeeping, but separate it from the active reference used for new drafts. Do not mix old and new policy text without clear status labels. If an approved change has not reached the drafting workflow, use a manually checked response until it has.
Then repeat the specific question affected by the change. Check the resulting wording, the source it used and any promise it added. This is an ongoing maintenance check, not evidence that every future answer will be right.
Use this checklist whenever a detail changes
- Confirm the approved wording, scope and effective date with the responsible person.
- Update the current reference and mark the previous wording superseded.
- Identify website pages, saved replies and workflow inputs that still carry the old detail.
- Have the responsible person update those copies and verify what the drafting workflow now reads.
- Check a draft for the changed topic before relying on the updated workflow.
Start with one fact that recently changed in your business. If you cannot identify who approved it or where the active copy lives, resolve that first. For help organizing a practical starting point, contact Russ at Zasper AI with a fictional or redacted example.