How should product and service FAQs support both search and website AI?

Build one answer around one customer decision

Collect repeated questions from service, the shop and website. Group product care, dimensions and material separately from class fit, process, timing and delivery; keep purchasing conditions and exceptions visible. Use the customer's own question, such as 'Can this ceramic cup go in a dishwasher?' Answer first, then state the applicable model, limits and next step. A vague instruction to see the product page does not resolve a five-product difference.

Give each answer a maintenance card: exact question, answer, model or service scope, source page, owner, approval date and recheck trigger. If only cup A has verified machine-washing guidance, do not silently extend it to cup B. If size changes the price, explain the pricing rule while the current product option or owner supplies the final price. This makes the answer inspectable instead of inviting AI to reason from promotional copy.

Google's English SEO Starter Guide recommends readable sections and maintained useful content, and its people-first guide asks whether visitors can complete their goal. Put answers on accessible product or service pages, with an FAQ section where helpful, rather than hiding all information inside chat. Search display and citation by another AI are not guaranteed. Google stopped showing FAQ rich results on May 7, 2026, so markup must not be sold as a route to that feature.

Connect stable answers to live facts and human decisions

FAQ is suited to stable rules, not a substitute for live inventory or enrolment. Cup care can cite approved guidance; 'are two blue large cups available now?' requires the current published option and stock. Age rules can be explained from approved class policy, but a Saturday group seat needs school staff or a separately verified enrolment system. Native kakinang staff slots do not represent class capacity. Route each question to the right source instead of adding more generic FAQs.

kakinang can use approved FAQs, supported files, successfully synchronized and activated website knowledge, and published product data. The owner reviews candidate content, activates it and tests representative questions. A changed web FAQ is not instantly learned, and a social post is not automatically approved knowledge. Where sources are missing or conflicting, state uncertainty and route to a responsible person rather than select an old plausible sentence.

In the fictional shop, three cards test the boundary. Cup A has a model-specific care answer; two Saturday trial participants have a class rule but no confirmed seats; custom engraving tomorrow lacks a published lead time. Answer with the approved condition, request staff seat verification, or collect requirements for a human estimate respectively. Visitors should be able to read general information without giving a phone number. Seek consent and explain purpose only when follow-up is needed.

Maintain answers through change records and wrong-question tests

Before publishing, ask the same question from a mobile search landing page, product page and website chat. Include synonyms, old product names, discontinued options, price changes, out-of-scope requests and a human-contact request. Record expected source, actual answer and correction owner. If a FAQ says 'returns allowed' while policy excludes some custom goods, resolve applicability in the underlying pages rather than instruct AI to choose one.

When a product, class or policy changes, edit the authoritative page, update FAQs and effective dates, then complete knowledge synchronization, activation and representative retesting. Keep the old answer's expiry, approver and affected pages. If AI still cites an old price, disable or correct that source or route to a person until verified. This is ongoing content operation, not a one-time paste of ten questions.

Judge whether visitors find the correct model, understand cancellation, reach a person for unknowns and receive actual lead follow-up. Search impressions and chat volume are clues, not revenue. Repeated questions about whether engraving is included mean the product page needs a clearer option and fee; the FAQ then explains the boundary.

Write clear negative answers too. If there is no delivery, say pickup only and give opening or collection arrangements; do not merely say 'visit us'. Where only some products ship, name the eligible options and how shipping cost is obtained. If private weekend lessons are unavailable, give the days that are available and distinguish staff-managed group enrolment. Visitors can then judge fit before disclosing contact details.

Layer sources deliberately. Current catalog records govern specifications, an approved course page governs service rules, and an undecided custom request is only an enquiry. If staff verbally promise rush work while the site says no rush orders, an owner must settle the current rule and update page, FAQ and knowledge. Do not keep two contradictory 'official' pages. Have a person familiar with the item review safety or use restrictions; AI cannot derive load, heat or allergen limits from a photograph.

Brevity must preserve conditions. 'Yes' without 'only model A, hand wash, no microwave' becomes an unsafe general rule. Compare price, date and exception across translated pages so an old campaign does not survive in one language. Give images descriptive alternatives and keep critical specifications in text rather than inside an image. This helps assistive reading and provides a textual basis for AI.

Monthly, sample ten real questions and mark answered with a valid source, correct but stale source, wrong intent, or missing fact. Repair the source and retest the second group, improve wording for the third, and assign the fourth to the product or service owner. Never remove unknowns to flatter a metric, or force a visitor to submit a form to read public facts.