Amazon Agent Policy readiness: audit and fix the tools that act on your seller account
To get ready for Amazon's Agent Policy, list every piece of software that acts on your seller account: repricers, bid tools, listing and review tools, scrapers, scripts, SP-API apps and AI assistants. For each, check that it names itself as an Agent in every request, never acts like a person or solves CAPTCHAs, respects Amazon's limits, can be stopped, needs no password and does not train models on Amazon data.
- Amazon's Agent Policy requires every Agent to put 'Agent/[agent name]' in the user agent string of all HTTP and HTTPS requests; the policy page was unchanged at revision 2 when we checked it on October 9, 2026.
- The Business Solutions Agreement lists Selling on Amazon, FBA, Amazon Advertising and the Selling Partner APIs among its Services, so ad tools and SP-API apps are in scope as well as browser bots.
- Section 4.2 of the agreement bars sellers, and any third party they allow, from using Amazon's materials or services to develop or improve large language, multimodal or machine learning models.
- Amazon's Acceptable Use Policy tells SP-API developers not to request or share Seller Central usernames or passwords, and to ask for secondary user permissions instead.
- Most SP-API rate limits apply per selling partner and application pair, and a throttled request returns HTTP 429, which Amazon calls retryable.
Our Agent Policy explainer covers what Amazon's rules say. This page is the practical part: an audit you run on your own tool stack this week. It goes tool type by tool type, says what Amazon's text asks of each one, how to check it without reading code, and what to do when a tool fails. We re-read the Agent Policy and the Business Solutions Agreement (BSA) on October 9, 2026; both were unchanged since our September check.
Which tools on your seller account does the Agent Policy cover?
The BSA defines an Agent as "any software or service that takes autonomous or semi-autonomous action on behalf of, or at the instruction of, any person or entity." Section 19 applies whenever you "deploy, enable, or authorize" one. The agreement's Services include Selling on Amazon, FBA, Amazon Advertising and the Selling Partner APIs (BSA, checked October 9, 2026). So the audit is not only about AI. A rule-based repricer, a bid tool on the Amazon Ads API and a script your analyst wrote all act on your account without a person approving each change.
Before you judge anything, build the list. Most brands find more software than they expected, because agencies, freelancers and past employees each added their own. Look in four places:
- Seller Central, Apps and Services, Manage Your Apps. Every SP-API app you have authorized, with its status.
- User permissions. Every person and service provider with a login, including any shared or generic users.
- The browsers your team uses for Seller Central. Extensions that read pages, fill forms or click buttons.
- Your own code and workflow tools. Scripts, spreadsheet add-ons and Zapier, Make or n8n scenarios that call Amazon's APIs.
What does each kind of tool need to pass?
This is the audit itself. Each row is one kind of software, the parts of Amazon's text that matter most for it, how to check it, and the usual fix. The Agent Policy's four rules apply to every row: put Agent/[agent name] in the user agent string of every request, never hide being an Agent by "mimicking the speed or pattern of human keystrokes" or by solving CAPTCHAs, answer truthfully when asked if it is a person, and never get around measures that "block, limit, modify, or control" Agents (Agent Policy, checked October 9, 2026). The table adds what each tool type tends to get wrong.
| Tool type | What Amazon's text requires most here | How to check it | Usual fix |
|---|---|---|---|
| Repricers | Agent name in every request (rule i); no getting around throttling (rule iv, and the BSA's bar on use intended to avoid "exceeding usage limits or quotas"); stop when Amazon asks (Section 19.1) | Find it in Manage Your Apps. Ask the vendor for one logged request header showing Agent/ and the tool's name, and what it does on a 429 response | Ask the vendor in writing; if it runs as a browser extension, move to an SP-API repricer |
| Bid and PPC tools | Amazon Advertising is a BSA Service, so the same four rules apply; no use of Amazon's data to "develop or improve" models (Section 4.2) | Ask which API it uses and where you authorized it. Ask whether your ad data trains a model shared across the vendor's clients | Get a no-training clause in the contract; disable it if the vendor will not give one |
| Listing tools (bulk edits, AI copy) | No human mimicry (rule ii); Section 4.2 on training models; request only the data the app needs (Acceptable Use Policy 3.8, checked October 9, 2026) | Ask whether it edits through the SP-API or by filling Seller Central forms. Ask which SP-API roles it requests and why | Move to SP-API authorization; drop roles it does not need |
| Review and feedback request tools | No clicking through Seller Central at human speed (rule ii); no getting around limits (rule iv) | Ask whether it sends requests through the SP-API or by pressing buttons in a logged-in browser | Switch to an SP-API tool; remove any browser automation |
| Scrapers and browser bots | The BSA bars "data mining, robots, or similar data gathering and extraction tools" for collecting Amazon's materials (Section 4); no CAPTCHA solving (rule ii) | Ask your team what monitors competitors, prices or reviews, and how. Anything that needs a CAPTCHA service or rotating proxies fails | Disable and remove; rebuild the data you need on SP-API reports |
| In-house scripts and workflow tools | The SP-API already requires a user-agent header on every request; add Agent/ and a name. Respect per-account rate limits | Search the code for the user-agent header. Check logs for 429 responses and how retries are spaced | Add the agent name, back-off on 429s and a stop switch, or rebuild |
| SP-API apps from vendors (analytics, inventory, accounting) | No passwords requested (Acceptable Use Policy 3.3); only necessary data (3.8); be explicit about "the use of models such as artificial intelligence" (2.4) | Read each app's roles in Manage Your Apps. Ask which features use an AI model and which follow fixed rules | Disable authorization for anything unused or unexplained |
| AI assistants with Seller Central access | All four rules at once; and you are "solely responsible for any use of or action taken under your password" (BSA) | Ask who on the team uses an AI browser, an AI extension or an assistant connected to the account, and how it connects | Stop any assistant driving a logged-in browser; connect through an authorized SP-API app that names itself |
The last row is the one most teams miss. An AI browser or extension that operates your logged-in Seller Central session acts as you, under your password, at the pace of a person clicking. That is the pattern rule ii describes. An assistant that connects through an authorized app, for example through an SP-API MCP server, can name itself in every request and be switched off in Manage Your Apps.
How do you check a tool without reading its code?
You do not need to be a developer to run this audit. You need evidence, in writing, for five questions. A vendor who has done the work can answer all five in a day.
- Its name. One logged request header that shows
Agent/and the tool's name. Amazon's Connect to the SP-API guide already requires a user-agent header on every SP-API request, up to 500 characters, with at least the app name, version and language (checked October 9, 2026). Amazon's pages show no combined example, so ask for a real header, not a promise. - Its limits. What happens on a throttled request. Amazon's Usage Plans and Rate Limits page says most limits apply per selling partner and application pair and that a 429 "is a retry-able status code" (checked October 9, 2026). A good answer is back-off and an alert. A bad answer is extra apps, accounts or proxies.
- Its stop. How it is paused for your account alone, and who presses the button if Amazon asks it to stop.
- Its data. Whether anything from your account trains or improves the vendor's models, or its AI provider's, with the provider named.
- Its access. Whether it ever needs your login or verification codes. The Acceptable Use Policy says developers must not "request or share Amazon Portal usernames or passwords," and should ask for secondary user permissions when a person needs Seller Central (3.3 and 3.7).
What should you do with a tool that fails?
Not every failure means switching the tool off today. Match the fix to what failed, and keep a manual fallback for repricing, bids and reorders while you change anything, because Section 19.2 lets Amazon limit or restrict Agent access at its sole discretion.
| Verdict | When it applies | What to do |
|---|---|---|
| Keep | The vendor answered all five questions in writing with evidence | File the answers; ask again whenever Amazon changes the policy |
| Ask the vendor | The tool connects as an SP-API app, but the agent name, limits or data terms are unclear | Send the five questions; set a date for answers |
| Move to SP-API authorization | The tool uses a shared login, a browser or an extension | Replace it with an app you authorize in Manage Your Apps; give people their own users |
| Disable authorization | Nobody can say what the app does or who uses it | In Manage Your Apps choose Disable authorization; Amazon's Revoke Authorizations guide says it then stays listed as disabled (checked October 9, 2026) |
| Rebuild | It is your own script, or the vendor will not change and the job matters | Rebuild on the SP-API with an agent name, back-off on 429s, a stop switch and a change log |
Do not wait for an old app to drop off by itself. Amazon's Seller Authorizations FAQ says it deactivates authorizations that have not been renewed and have made no API call for 365 days (checked October 9, 2026). An app nobody remembers can stay connected for a year. For people and agencies rather than apps, our guide to giving an agency access to Seller Central safely covers which permission each one should get.
Who can audit and rebuild the tools that fail?
Ecomsellertool, a tech agency building on Amazon's seller APIs since 2017, rebuilds failing tools as SP-API apps on your own accounts, with an agent name, rate-limit back-off, a stop switch and a change log. Send one tool in a project brief: what it does today and how it connects. A person replies within 2 business days with a written plan, timeline and fixed price.
The work behind that is not new to us. For SellerDrive we built analytics on SP-API reports and the Amazon Ads API, combining order, finance, catalog and ad data with automated data retrieval. AccountDr is rule-based automation we built that alerts sellers to listing faults, stranded inventory, hijackers and ASIN changes. Repriser is a rule-based repricer we built for a client. New agents we build run on Ecomsellertool Growth OS and are designed to follow Amazon's Agent Policy; each AI agent page says which steps follow rules and which use a model. You keep your accounts, your data and the custom code we build; the Growth OS base is licensed to you.
If you need a custom SP-API integration rather than one replaced tool, see our Amazon SP-API development service. If you are not sure the tool is worth keeping at all, our list of what to automate first helps you decide before you rebuild anything.
Where should you start this week?
Run the audit once now, and again whenever you add a tool, an agency or a freelancer.
- Export the list from Manage Your Apps and your user permissions, and note who added each item.
- Ask your team which browser extensions, AI browsers and scripts touch Seller Central or the Ads console.
- Put each tool through the table above and record a verdict: keep, ask, move, disable or rebuild.
- Send the five evidence questions to every vendor in the "ask" column, with a date.
- Disable authorization for every app nobody can explain.
- Keep a manual fallback for repricing, bids and reorders while you change tools.
If the audit turns up account problems, such as suppressed listings or stranded stock, start with the free 24-hour diagnostic: connect Amazon with Login with Amazon, no password shared, and get a report within 24 hours of connecting, on business days. We only read data; we never change listings, prices, stock or ads. If a tool has to be replaced, send us a project brief for that one tool first.
Frequently asked questions
Which tools on my Amazon account count as Agents?
Read literally, any software or service that takes autonomous or semi-autonomous action on your account at someone's instruction. That covers AI assistants, but also rule-based repricers, bid tools, review-request tools, scripts your team wrote and browser extensions that click through Seller Central. A report you download and read yourself is not an Agent; a tool that changes prices, bids or listings on its own is. This is our reading of Amazon's text, not legal advice.
How do I check if my repricer follows Amazon's Agent Policy?
Ask the vendor for one logged request header that shows Agent/ and the tool's name in the user agent string, ask what the tool does when Amazon returns a 429 throttling response, and ask how it is stopped on your account alone. Then confirm in Seller Central under Apps and Services, Manage Your Apps, that it connects as an authorized app rather than through your login.
Are browser extensions that automate Seller Central allowed?
The Agent Policy bars Agents from hiding that they are automated, for example by mimicking the speed or pattern of human keystrokes or page navigation, or by solving CAPTCHAs. An extension that clicks through Seller Central in your logged-in session is the hardest kind of tool to square with those rules, and it acts under your password, which the agreement makes your responsibility. Most such jobs can be moved to the SP-API.
Can an AI assistant log in to Seller Central for me?
We would not let one. The Business Solutions Agreement makes you solely responsible for any use of or action taken under your password, and an assistant driving your logged-in browser acts as you. Use an assistant that connects through an authorized SP-API app and names itself in its requests, and keep money decisions, such as budget changes and purchase orders, with a person.
What should I do with a tool that fails the audit?
Match the fix to the failure. If the vendor can answer in writing, ask and keep the answer. If the tool uses your login or a browser, move the job to an SP-API app you authorize. If nobody uses it, choose Disable authorization in Manage Your Apps. If it is your own script or the vendor will not change, rebuild it on the SP-API with an agent name, back-off and a stop switch.
Does Amazon approve tools for the Agent Policy?
We found no approved-tool list or sign-off process in the Agent Policy, Section 19 of the Business Solutions Agreement or the Acceptable Use Policy as of October 9, 2026. Treat any vendor's claim that its tool is approved by Amazon for the Agent Policy as a question to put in writing, and ask for the evidence in this checklist instead.
How we research, fact-check and compare: our editorial standards. Spot an error? Email hello@ecomsellertool.com and we will correct it.
