Put the Sales FAQ on the Website Before the Chatbot Answers It

A buyer asks about delivery, refunds, and whether a plan includes onboarding. The chatbot answers from memory. The website FAQ still says something else. The sales team spends the next call undoing a promise nobody approved.

Publish the durable answers on the website first. Let the chatbot repeat only what that page already states. Keep pricing exceptions, contracts, and one-off commitments on a human route.

Decide where each question lives

Copy this record into a spreadsheet. One row is one buyer question. Fill every field before the chatbot is allowed to answer that question.

Question class	Put on the website FAQ	Chatbot may answer	Must go to a person
Hours, channels, and how to reach the team	Yes, with staffed hours	Yes, quoting the page	No, unless the buyer asks for a person
Standard delivery, return, or cancellation steps	Yes, with the dated source page	Yes, linking to that page	Yes, if the buyer disputes a specific order
List price or published plan comparison	Yes, if the number is public	Yes, only the published number	Yes, for discounts, custom quotes, or contract terms
Whether a feature exists in a named plan	Yes, from the current plan table	Yes, if the table is current	Yes, if the table is missing, stale, or conflicting
Refund, exception, legal, safety, or privacy decision	Summary of the public policy only	No decision; route to the policy page or a person	Yes

A public FAQ is the source the buyer can re-read after the chat closes. The chatbot is a retrieval path, not a second policy author.

How the split works

1. Collect the questions sales and service actually receive for one product line.
2. Mark each row with the table above. If two owners disagree, the stricter route wins until a named owner updates the website page.
3. Publish or refresh the website FAQ before enabling the chatbot answer.
4. Give the chatbot the page URL and the exact sentences it may paraphrase. If the page is missing, the bot must not invent the gap.
5. Review mismatches weekly: chat answers that are not on the page, and page answers the bot never uses.

NIST's AI Risk Management Framework treats context, measurement, and human oversight as ongoing work, not a one-time launch checklist. Use that as a reason to keep the website page dated and owned. Read the AI RMF overview.

Anthropic's engineering guidance prefers the simplest controlled workflow that still does the job. Repeating a published FAQ is simpler than letting a model compose policy. Read the agents guidance.

Worked example: a delivery question on a standard plan

This is a fictional test case, not a customer result.

1. Question received: “Do you deliver to District 7 in two days on the Standard plan?”
2. Website FAQ row: Standard plan delivery is 3–5 business days in Ho Chi Minh City; District 7 is in scope; two-day delivery is not a published promise.
3. Chatbot may say: delivery is 3–5 business days for Standard in Ho Chi Minh City, and link the FAQ. It must not promise two days.
4. If the buyer says an order already paid for two-day delivery, the case leaves the FAQ path and goes to a person with the order ID.
5. The record is accepted only when the website sentence, the chat wording, and the handoff rule share one owner and one last-reviewed date.

Digital.gov separates acknowledgement, routing, and completion. An automated FAQ reply acknowledges and informs; it does not close a disputed order. Review the contact-center guidance.

Risks and common mistakes

Risk or mistake	Consequence	Control
Chatbot answers from model memory	Buyers receive unpublished promises	Allow only sentences tied to a current FAQ URL
Website FAQ is undated	Nobody can tell which answer is current	Show last reviewed date and a named owner
Every exception is stuffed into the FAQ	The page becomes unusable and still misses the real case	Keep exceptions on a human route
Sales and service publish different answers	The buyer gets two policies	One owner per question class; stricter route until updated

How to measure the result

• Coverage: buyer questions in the sample that have a filled row divided by questions in the sample.
• Drift: chat answers a reviewer cannot find on the linked FAQ page, counted each week.
• Handoff correctness: disputed or exception cases that used the human route divided by cases that should have.
• Review cadence: days since the named owner last checked the public FAQ for that product line.

These measures describe operating hygiene. They do not prove conversion lift.

Evidence and limitations

This guide is an editorial method for keeping published answers and chatbot answers aligned. It does not describe an Easy AI product capability, prescribe staffing, or claim that a public FAQ increases sales. Validate the question list against your own transcripts and approved policy pages.

What to do next

Fill the four-column record for one product line, then refresh that website FAQ before changing chatbot copy. For the conversation that should collect context and hand off, use the AI sales chatbot guidance flow.