Marketplace Buyer Questions Automation: Triage and First Replies

Most pre-sale questions on a marketplace are repetitive. Buyers ask whether a size fits, whether an item is in stock, how long delivery takes, or whether a colour matches a photo. Each question takes a few minutes to answer by hand, and the same answer is written again and again. When the seller is away, the questions wait, and buyers often move to another seller while they wait.

This page explains how AiStaffo would set up an AI worker for that inbox, what it would do on its own, what stays with a person, and where the approach does not fit. It is written for owners and operations heads who manage one or more marketplace shops, whether the business has five staff or several thousand.

What the service covers

The service covers the questions a buyer sends before purchase. It does not cover the full order lifecycle. The AI worker reads each new message in the marketplace messaging inbox the seller already uses. It identifies the topic, finds the matching listing, checks the stock sheet for availability and reads the size chart where the question is about fit. If the answer is already written in the seller's own data, the worker drafts or sends a standard reply that uses the seller's approved wording.

If the answer is not in the data, or if the message touches refunds, damaged goods, product safety, pricing or a complaint, the worker does not reply. It places the message in a review queue and names the person who handles that category. The person decides what to send.

What is included

• A review of the marketplace inboxes in use, with a count of the recurring pre-sale question types
• A map of which listing, stock and size chart fields answer each question type, and which fields are missing or out of date
• A set of standard reply templates drafted from the seller's own listing text, to be approved by the seller before use
• Routing rules that send refund, damage, safety and complaint messages to a named reviewer
• Connections to the stock sheet and size chart so that replies reflect the current data
• A daily log of every question received, the reply sent or the reason it was routed to a person
• A written list of any new policy wording the worker proposes, held until the seller signs it off
• Support after launch, including changes to templates and routing when the seller adds a product line

How the service runs

The work follows five stages. Each stage ends with a decision by the seller, so nothing goes live without sign-off.

1. Audit. The free automation audit looks at the inboxes, the recent question history and the state of the listing and stock data. The output is a short list of question types worth automating and the ones that should stay manual.
2. Process map. Each question type is written as a flow: the trigger, the data the answer depends on, the reply or the route, and the person who approves it. The seller reviews the map line by line.
3. Build. The AI worker is connected to the marketplace messaging inbox, the stock sheet and the size chart. Templates are drafted and tested against past questions from the log.
4. Run. The worker runs in a supervised period. During this time every automatic reply is checked by a person, and the mismatches are corrected in the templates or the data.
5. Support. After the supervised period, the worker runs on the agreed rules. The seller receives a regular log and can change templates or routing at any time.

Which messages go where

The split between automatic and person-led handling is the part that matters most. The table below shows how the service would treat the common question types. The categories are a starting point; the process map sets the final list for each shop.

Routing by type protects the account more than any template does. A reply that promises a refund, or that describes a product as safe for a use the listing does not cover, creates a problem that the marketplace's policies and the buyer both react to. Keeping those messages with a person is the reason the worker is built with routing first.

Marketplace rules that shape the design

Each marketplace has its own messaging rules, and the design has to fit them. On Amazon, Buyer-Seller Messaging is the approved channel for buyer contact, and the platform expects a reply to every buyer message within 24 hours, weekends included, according to eDesk's 2026 guide to Amazon messaging. The same guide notes that messages must not include marketing, outside links or requests for reviews. Amazon's own help pages are the reference for the current rules, and the worker is configured to stay inside them.

On eBay, the seller can turn on eBay.ai, which suggests replies to questions about item details or postage, using information from the seller's listings and order details. According to eBay's help page, it avoids questions about pricing and offers, and the seller can review or change each suggestion. eBay also offers a basic automatic response for times away, which does not pause listings. On Etsy, the Shop Manager offers temporary auto-replies of up to five days and saved replies for common questions, according to Etsy's seller help. The service uses these built-in features where they exist, rather than replacing them.

Two points apply to every marketplace. First, a canned line that does not answer the question does not help the buyer, even if it meets a response-time rule. Second, third-party tools that send unrequested messages or push buyers off the platform can put an account at risk. The worker sends only the standard replies the seller has approved.

What the client provides

The service depends on the seller's own information. The client provides access to each marketplace inbox through the permissions the platform allows, a current stock sheet with one row per saleable variant, the size charts used on each listing, and the shipping terms the shop actually follows. The client also names one person for each review category, such as refunds, damage and safety, and that person must be reachable within the response window the shop has chosen.

The seller also decides what the worker may say about each product. Where the listing does not state a fact, such as whether a material is suitable for a particular use, the worker does not supply one.

Typical timeline

For a single shop with one or two marketplace inboxes and a clean stock sheet, the work usually runs in about six to eight weeks. The audit takes one week, the process map one week, and the build two to three weeks. The supervised run takes about two weeks. Shops with several marketplaces, many variants or size charts kept in different formats take longer, because the data has to be cleaned before the worker can match against it. The audit sets the actual timeline for each shop.

What this service does not cover

The worker does not negotiate prices, offer discounts or approve custom work. It does not decide refunds, judge whether a product is damaged or safe, or answer legal, tax or allergen questions. It does not handle returns, disputes or complaints. It cannot answer a question that the listing, the stock sheet and the size chart do not cover, and it does not know about a shipping delay until that delay is written into the data.

It also cannot make a listing accurate. If the size chart is wrong, the worker will repeat the error faster. The audit is the point where those gaps show, and fixing them remains the seller's task.

Automation is not worth the effort for a shop that receives only a few pre-sale questions a month, or for a catalogue where every product is custom and every question is unique. In those cases a saved-reply list in the marketplace's own tools is usually enough.