Amazon Communication Guidelines: What You Can Send to Buyers, What Gets Flagged and How Your Team Stays Compliant

Amazon's Communication Guidelines set out what sellers may send buyers through Buyer-Seller Messaging, and when. Two things get sellers flagged most often: marketing content of any kind, and messages sent for a reason Amazon does not permit. This guide covers what is allowed, what is not, what a violation actually looks like, and the review and record-keeping process that keeps a support team inside the policy.

What the policy permits

Amazon's Communication Guidelines permit proactive messages for a specific list of reasons: completing an order (a question needed to ship it), scheduling delivery, handling a return, requesting an invoice, and requesting a product review through Amazon's own Request a Review feature rather than a custom message.

Responding to a message the buyer sent first is always permitted and is not restricted to this list, provided the reply stays inside the 24-hour response rule.

Our Amazon seller messaging guide covers where these messages live and how buyer opt-out works, and our guide to the Amazon 24-hour response rule covers the clock.

What the policy prohibits

Marketing or promotional content of any kind, including discounts, coupon codes, or invitations to follow a brand elsewhere.

Links to external websites, except a secure link needed to complete the order. Tracking pixels and images used for anything other than the order are treated the same way.

Email addresses, phone numbers, or other contact details meant to move the conversation off Amazon.

Requests for a positive review, for a specific star rating, or for a buyer to remove or change a review already left. Amazon scans messages for this automatically, so words like "review", "feedback", "five-star" and "discount" are the ones most likely to trip a filter even when the sentence around them is innocent.

Repeated messages to a buyer who has not responded to an earlier one.

Order confirmations and shipping notices Amazon already sends automatically. Sending your own is not a violation by itself, but it still has to follow every other rule on this list.

What a violation looks like and what it costs

A message that trips Amazon's filters, a phrase like "five stars", a phone number, an external link, typically does not send. The seller sees a notice in Seller Central that the message was not delivered rather than finding out from the buyer.

A pattern of violations moves from a warning to a restriction on messaging privileges, and repeated or serious violations affect account health more broadly.

The cost is rarely the warning itself. A blocked message is a buyer who did not get an answer, and a buyer with a damaged item or a late FBM shipment who does not get an answer files an A-to-z claim, which counts against Order Defect Rate whether or not Amazon rules in your favour. Compliance and speed are the same problem: the fastest way to protect account health is to fix the buyer's problem, inside policy, before they have reason to complain to Amazon.

Who on the team can say what

Not every reply needs the same level of approval. A simple authority model keeps things moving without letting anyone improvise into a policy violation, and without a routine refund waiting on a manager who is in a meeting while the 24-hour clock runs.

Start the threshold conservative and widen it as the team's record shows it can be trusted. The point of the model is that prevention and defence are two different jobs: an agent's job is to resolve the problem before a claim is filed, and defending a claim that has already been filed goes through the compliance contact.

Templates do the policy work here. Write and lock a template for each of the ten or so situations that come up most (shipping delay, damaged item, partial refund confirmation, return label, warranty approval, wrong item) so an agent under pressure picks approved wording rather than free-typing a reply that might carry a flagged phrase.

Internal notes and escalation

When a message needs a second pair of eyes, standardise what the note says before it moves. A useful note answers three things: what was checked (the order, the tracking, a photo the buyer sent), which policy line applies to the situation, and what the escalation needs to decide.

The same rule applies when an agent uses their own authority. A one-line note ("Issued partial refund of [amount] to resolve a damaged-item complaint; buyer supplied photo evidence") takes ten seconds and turns a decision into a record. Over time those notes are also where patterns surface, such as one SKU generating repeat damage complaints.

Keep the internal note separate from the buyer-facing thread. It should never be reachable from the message the buyer sees.

Once approved, the reply goes out from the same thread, referencing the approval rather than restating the whole internal discussion. The original 24-hour clock still applies through the entire handoff; escalating does not pause it, so the note exists to make the handoff faster, not to buy the team more time.

Keeping the record

A dispute with Amazon, whether it is a claim, a blocked message, or a review of your account, gets decided on the record you can produce. Keep, per message: the text as sent, the timestamp, who sent it (a named agent, not just "support"), and which template or internal note it was built from if one applies.

Seller Central holds its own copy of buyer-seller messages, but its search is limited and gives you no control over how long anything stays accessible, which is the practical reason most sellers keep an independent copy of anything that could matter later.

Review the record monthly: agent-issued refunds against their notes, how many messages were escalated and why, and which ticket types still need a manager. Each of those is either a template that needs writing or a threshold that can move.

Checking policy before the message goes out

Templates written inside the policy solve most of this before it becomes a problem: an agent picks a template built to pass Amazon's filters rather than free-typing a reply under pressure. A policy check on every outbound message, including one an AI drafts, catches the phrase that would trip a filter or read as review solicitation before it sends rather than after a buyer reports it.

Anything needing approval carries its internal note and the approval into the same ticket, so the record stays with the thread as it moves. Refunds and order edits happen from the same ticket through the Amazon connection, so an agent inside their threshold resolves a damaged-item message in one contact instead of switching to Seller Central and back. The message archive stays tied to the order, so a policy question or a claim months later starts with the full history already in one place.

AI-drafted replies run through the same policy check as anything a person writes; nothing skips review because a person did not type it first. Our Amazon helpdesk software page has the full setup across marketplaces.

Frequently asked questions

What are Amazon's Communication Guidelines?

Amazon's policy on what sellers may send buyers through Buyer-Seller Messaging and when. It sets out permitted reasons to message a buyer first, and prohibits marketing content, external links, and review solicitation.

Can I send marketing messages to Amazon buyers?

No. Promotional content, discounts, or invitations to engage with your brand off Amazon all violate the policy, whether the buyer asked for it or not.

Can I ask an Amazon buyer for a review by message?

Not directly. Amazon prohibits asking for a positive review or a specific rating in a message. Request a Review, Amazon's own feature, is the permitted route to ask for feedback.

What happens if I violate the Communication Guidelines?

A flagged message usually does not send, and you will see a notice in Seller Central. Repeated violations escalate from a warning to a restriction on messaging privileges and can affect account health.

Does a manager still need to approve an A-to-z claim response?

Yes, for a claim that has already been filed; that goes through your compliance contact. An agent's authority covers prevention: resolving the refund or replacement inside their threshold before the buyer files.

How do I make sure my support team stays inside Amazon's messaging policy?

Templates built inside the policy, a check on every outbound message, and a clear approval path for anything unusual. eDesk applies the same policy check to AI-drafted replies as it does to a person's.

Ready to keep every Amazon message inside policy?

Keep every Amazon message inside policy without slowing your team down. Start for free or book a demo.