A catalog agent that keeps listings consistent on every marketplace
Updated · By Ecomsellertool editorial team
Multichannel catalog management automation keeps one master record per product and turns it into listings that meet each marketplace's rules, such as Amazon's 75-character titles, Walmart's 80-character key features and eBay's required item specifics. A catalog agent runs each marketplace's own validation, flags live listings that drift from the record, and sends new listings and claims to your team for approval. We build and deploy it on your own accounts.
- From July 27, 2026, Amazon titles in every category except media must be 75 characters or less, per its US Seller Forums announcement; a new Item Highlights field adds 125 characters for materials or use cases.
- Walmart allows titles of up to 150 characters and 3 to 10 key features of no more than 80 characters each, and holds sellers responsible for reviewing AI-generated content before it is published.
- TikTok Shop reviews every product after it is created or edited, and a product's GTIN, UPC or EAN can no longer be changed once it is submitted for review.
- eBay gives titles 80 characters and requires every Required item specific on any new or revised listing.
- On September 24, 2026, Amazon announced multichannel tools in Seller Central for eBay, Shopify, TikTok and Walmart, with a planned edit-once description update. Status: rolling out (US), at no additional cost.
- We build and deploy the agent on your own accounts. New listings, model-drafted copy and product claims come back to your team before they publish.
Listing problems found across a catalog, with the reason for each
Read the Grow8 case study ↗What the listing and catalog agent watches, does and sends back
Trigger: A new or changed product in your master record, a scheduled drift check of every live listing, and any listing issue, suppression or failed review a marketplace reports.
| What it watches | What it does within your rules | What comes back to your team |
|---|---|---|
|
|
|
Steps that follow fixed rules
- Schema checks against each channel's required attributes and allowed values
- Length and character limits per field and channel, such as 75 characters for Amazon titles and 80 for Walmart key features and eBay titles
- Banned wording filters: promotional claims, other retailers' names, refund promises and claims missing from the fact sheet
- Identifier checks: a valid GTIN per SKU, no duplicates, or an exemption on file
- Consistency checks: every published fact equals the master record
- Upload pacing and the approval thresholds your team sets
Steps that use an AI model
- Drafting each channel's title, bullets, key features and description from the master record within that channel's limits
- Mapping your attribute values to a channel's allowed values when the names differ
- Suggesting the product type or category for each channel from the product data
- Reading supplier spec sheets and manuals to propose missing attribute values, with the source quoted
- Spotting live listings that state a different fact in different words, which a string comparison misses
- Reading reviews and buyer questions to propose facts the listing is missing
A brand on Amazon, Walmart, TikTok Shop and eBay keeps the same product facts in four places, under four rule books that change on their own schedules. This page describes the listing and catalog agent we build: how one master record becomes a correct listing on each channel, what it checks before anything goes live, and what comes back to your team. Every marketplace rule below was checked on September 25, 2026.
What does a listing and catalog agent do?
It keeps one catalog true everywhere you sell. The agent holds a master record per product with every fact your team has approved: dimensions, materials, counts, compatibility, identifiers and images. From that record it drafts each channel's version to that channel's rules, runs the marketplace's own validation and publishes what your team approves. Then it keeps watching. When a live listing drifts from the record, because someone edited it by hand, a marketplace rewrote it or another seller's content won the page, the agent flags the field and proposes the fix.
That is a different job from making one Amazon listing convert better, which is what our AI listing optimization software does: audit and rewrite at scale on Amazon. The catalog agent's job is consistency across channels, the same product described with the same facts and accepted by four sets of rules. It is one of the AI agents we build on top of Growth OS, the base system we deploy on your own accounts. Its listings module covers AI listing audits and rewrites, attribute completeness for AI shopping assistants, and one catalog pushed to every marketplace with per-channel rules.
Most of the work is rules. Length limits, banned characters, required attributes and identifier checks are applied the same way every time. A model is used where rules break down: drafting channel copy inside each limit, mapping your attribute values to a channel's allowed values when the names differ, reading supplier spec sheets for missing facts, and spotting two descriptions that state different facts in different words. Where it acts on Amazon through the SP-API, it is designed to follow Amazon's Agent Policy. Stock and orders across the same channels belong to the multichannel operations agent; the two share one SKU map.
How does one catalog become correct listings on Amazon, Walmart, TikTok Shop and eBay?
By mapping every field once and applying each channel's rule to it. The table below is the mapping we start from. The requirement in each cell comes from the marketplace's own documentation, linked underneath; the rule after it is how the agent transforms the master record for that channel. Every rule is a setting your team can read and change, per channel and per product type.
| Field | Master record | Amazon | Walmart | TikTok Shop (US) | eBay |
|---|---|---|---|---|---|
| Title | Brand, product type, model and key attributes (size, count, material, color) stored as separate fields | 75 characters or less since July 27, 2026 (media excepted); no special characters such as ! $ ? unless in the brand; no word more than twice. Rule: brand, type and one attribute; extra detail moves to Item Highlights (125 characters) | 150 characters or less; no special characters such as ~ ! * $, promotional claims, other retailers' names or, in most cases, years; key values with units (12 oz, 4 count). Rule: brand, type and attributes with units | 1 to 255 characters. Rule: the Walmart-length title, run through the same promotional-wording filter | 80 characters; brand, model or type, color, style and model number; avoid special characters. Rule: fill the 80 with the attributes buyers search |
| Short facts | An approved fact sheet: one checkable fact per line, each with its source | Bullet points; since August 15, 2024, limits on special characters, emojis and phrases such as refund-related promises. Rule: one fact per bullet | 3 to 10 key features, 80 characters each at most, no HTML. Rule: the shortest form of the top facts | Selling points inside the description; TikTok recommends 3 to 5, each under 250 characters, with an image. Rule: top facts, one image each | Item specifics carry the facts buyers filter on. Rule: every fact that maps to a specific goes there, not just into the text |
| Description | Long-form copy, approved once per product | Product description. Rule: the same facts as the bullets and nothing the fact sheet does not support | One paragraph, at least 150 words (varies by category); no bullet lists, no "Amazon exclusive" or other retailer names. Rule: rewritten for Walmart, never pasted from Amazon | HTML, up to 10,000 characters; over 300 recommended. Rule: Amazon bullets become short HTML sections | A detailed description. Rule: the same facts, no claims beyond the fact sheet |
| Product ID | A GTIN per SKU, or an exemption on file | UPC, EAN or GTIN; without one, a GTIN exemption approved in Seller Central first | GTIN, UPC, EAN or ISBN, cross-checked against GS1, or an approved exemption; listings with invalid IDs are removed | GTIN, UPC or EAN where provided, unique per SKU; editable while the product is a draft, locked once submitted for review | Required in most categories; a catalog match pre-fills fields such as title and photos |
| Category and attributes | Your own attribute names and values, mapped once per channel | A product type; attributes and constraints as a JSON Schema from the Product Type Definitions API | A product type from Walmart's taxonomy; required attributes as a JSON schema from Get spec; values must fit each attribute's unit | A leaf category in the 7-level US category tree; required product and sales attributes; category rules can add certificates, size charts and dimensions | Item specifics per category, labeled Required, Required Soon or Recommended |
| Images | One approved set, main image first | The approved set, in your order | Square, seamless white background, 2200 by 2200 pixels (1500 by 1500 minimum for zoom); no retailer names or logos; four or more recommended | Up to 9 main images, 300 to 4,000 pixels; at least 5 and a white first image recommended | Up to 24 photos; 500 by 500 pixels minimum, 1600 by 1600 recommended |
Amazon. Amazon's announcement of the July 27 title update, posted in the US Seller Forums, says titles in all categories except media must be 75 characters or less, adds an Item Highlights field of 125 characters for materials or recommended use cases, and says titles still over the limit are updated gradually to Amazon's AI recommendation while the listings stay active (checked September 25, 2026). The title requirements that took effect on January 21, 2025 ban special characters such as !, $ and ? unless they are part of the brand, and repeating a word more than twice. Since August 15, 2024, Amazon's bullet point requirements restrict special characters, emojis and phrases such as refund-related promises.
Amazon's Listings APIs FAQ says a product without a UPC, GTIN or EAN needs a GTIN exemption, requested manually in Seller Central, before it can be listed through the API (checked September 25, 2026). The attribute rules for each product type come as a JSON Schema from the Product Type Definitions API, which the agent caches and re-reads when Amazon changes a schema.
Walmart. Walmart's product details policy caps titles at 150 characters and bans special characters, promotional claims, other retailers' names and, in most cases, years in them. It asks for 3 to 10 key features of no more than 80 characters each and a one-paragraph description of at least 150 words, with no claims of competitor exclusivity such as "Amazon exclusive" (checked September 25, 2026). It also says sellers are responsible for reviewing and substantiating AI-generated content before publication, which is one reason every model-drafted Walmart item comes to your team first.
Walmart's Get spec API returns the required attributes for up to 20 product types per request as a JSON schema. Its product identifier policy says IDs are cross-checked against the GS1 database and listings with invalid IDs are removed, which can lead to suspension. Its image guidelines ask for square images on a seamless white background, with no retailer logos or names (checked September 25, 2026). More in our Walmart guide.
TikTok Shop. TikTok's Create Product API allows titles of 1 to 255 characters for US shops, up to nine main images and an HTML description of up to 10,000 characters (checked September 25, 2026). A product's GTIN, UPC or EAN can be changed while the product is a draft; once it is submitted for review, changes are not allowed. US shops list in a leaf category of TikTok's 7-level tree, and Get Category Rules returns what that category needs beyond its attributes, such as certificates, size charts and dimensions.
TikTok's Products API overview says TikTok reviews every product after it is created or edited, and that a live product whose edit fails review keeps its previous version. It also requires brand authorization for certain brands and limits new shops to 100 product uploads a day, rising to 1,000 after probation (checked September 25, 2026). See our TikTok Shop guide for fees and fulfillment.
eBay. eBay's guide to creating listings gives titles 80 characters and asks for the brand, model or type, color, style and model number in them (checked September 25, 2026). Its item specifics page says every Required item specific must be filled in for any new or revised listing, and flags the ones that will be Required Soon. Product identifiers are required in most categories, and a match in eBay's catalog pre-fills fields such as title and photos, so the agent accepts a match only for the exact product. eBay's photo tips allow up to 24 photos. More in our eBay guide.
This is why copying an Amazon listing to Walmart field for field rarely passes. Amazon bullets are not Walmart key features, a 75-character Amazon title leaves out unit values that Walmart's longer title has room for, and a pasted description can carry wording Walmart bans. The agent drafts every channel from the master record, never from another channel's listing, so a mistake on one channel does not spread to the others.
What does it check before anything is published?
Every listing passes two layers of checks: the marketplace's own validation, called before submission wherever the marketplace offers one, and the agent's rules on top. Then it waits at the approval step your team sets. Nothing reaches a marketplace until it passes both layers, and nothing a model drafted reaches a marketplace for the first time without a person approving it. The last column is what happens after the marketplace accepts the change.
| Channel | The marketplace's own check | Checks the agent adds | Approval step (your team) | After publishing |
|---|---|---|---|---|
| Amazon | Payload checked against the product type's JSON Schema, then a validation preview that runs the production checks without saving anything | Title length and characters, banned phrases, every fact equal to the master record, identifier or exemption present | New ASINs, title or claim changes, and edits to detail pages where another contributor's content shows | Listing issues through Amazon's issue notifications; Amazon's own title rewrites flagged as drift |
| Walmart | Required attributes from the product type's JSON schema; product IDs cross-checked against GS1 | Key feature count and length, 150-word description, retailer names, values that match each attribute's unit, image size and background | Every model-drafted title, key feature and description, and every new item | Item and feed status; content Walmart shows from another seller instead of yours |
| TikTok Shop | Check Product Listing, which returns the issues to fix before the product is created; the category's rules | Leaf category, required attributes, brand authorization, package weight, daily upload limit | Category and identifier choices, which lock once the product is submitted for review, and certificates | TikTok's review result, reported through its product webhook; a failed edit leaves the previous version live |
| eBay | Required item specifics for any new or revised listing; catalog match on the product identifier | Title length, exact identifier match, photo size and count | Catalog matches and every new listing | Required Soon specifics tracked before they become required |
On Amazon, the Listings Items API's validation preview mode runs the same checks as a production submission but saves nothing to your catalog, according to the Listings APIs FAQ. Amazon's listings workflow guide adds that some issues appear only after a submission is accepted, and recommends subscribing to the LISTINGS_ITEM_ISSUES_CHANGE notification to hear about them in near real time (checked September 25, 2026). So the agent treats "accepted" as a step, not the finish: it confirms each change on the live listing and reopens the task if an issue appears.
On TikTok Shop, the Check Product Listing API takes the full product information before it is created and returns a list of issues to resolve (checked September 25, 2026). Volume is a rule too. The same Amazon FAQ puts the JSON listings feed at up to 25,000 SKUs per request and five requests every five minutes, and TikTok limits new shops to 100 product uploads a day. The agent queues a large catalog push inside those limits instead of failing halfway through a launch.
Where the approval step sits is your decision. We start with every new listing, every model-drafted text, every product claim and every category or identifier choice going to a person, because those changes are the hardest to undo: TikTok locks an identifier once a product is submitted for review, and Walmart holds sellers responsible for AI-generated content. Once drafts for a product type stop needing edits, your team can let re-publishing of already approved text run on its own.
How does it make listings readable by AI shopping assistants?
By filling every attribute and keeping the facts identical everywhere. Amazon said in 2024 that Rufus answers product questions from product listing details, customer reviews and community Q&As, and in May 2026 it brought Rufus and Alexa+ together as Alexa for Shopping for US customers (checked September 25, 2026). A bowl listed as 5 quarts on Amazon, 4 quarts on Walmart and with no capacity on TikTok Shop gives any assistant a reason to hedge or recommend something else. The agent supplies what that depends on: complete attributes and the same numbers on every channel.
What to write for assistants is covered in our guide to AI shopping agent-ready listings and our Amazon Rufus explainer. The agent turns that guidance into checks: a completeness score per product and channel against the channel's schema, a list of attributes still blank, and a flag on any fact that appears in an image but in no text field. The same master record can also fill Google Merchant Center's product data specification and OpenAI's product feeds for ChatGPT (checked September 25, 2026). For the wider shift, see agentic commerce.
Amazon's multichannel tools, a platform or a catalog agent: which fits?
On September 24, 2026, Amazon announced multichannel selling tools in Seller Central. Amazon says the experience starts with connecting eBay, Shopify, TikTok and Walmart accounts, importing and linking those listings, viewing orders and fulfilling across channels, and that sellers will be able to enhance a description once on the Manage Products page and have it update everywhere it is live, formatted to each channel's requirements (checked September 25, 2026). Status: rolling out (US), gradually over the coming months, at no additional cost. Marketplace Pulse reported the news the same day. If your catalog already lives in Seller Central, try it first: Amazon charges nothing extra for it.
Platforms are the second option. Rithum says brands can "manage product listings centrally across 600+ marketplaces in the Rithum network" (rithum.com, checked September 25, 2026). Salsify, a product experience management platform, says it helps brands "syndicate product content to every consumer touch point" (salsify.com, checked September 25, 2026). If standard workflows fit your catalog and you want connectors that already exist, a platform is the better choice. Our Rithum and ChannelAdvisor alternatives page compares the options.
| Your situation | The better choice |
|---|---|
| Your catalog lives in Seller Central, and you sell on Amazon plus channels Amazon connects | Amazon's multichannel tools, once they reach your account |
| Many channels, standard content workflows and a team that wants ready-made software | A platform such as Rithum, or a PIM such as Salsify |
| A master record outside Amazon, your own per-channel rules, claims that need approval, and systems no one connects for you | A catalog agent built and deployed on your own accounts |
The difference is where the rules and the record live. Amazon's announcement describes editing in Seller Central and reformatting for each channel. It does not describe approval steps, claim rules of your own, or a master record that sits outside one marketplace's account. An agent on your accounts can also sit on top of a PIM you already run: the PIM stays the record, and the agent does the per-channel mapping, validation and drift checks.
What stays with your team?
The decisions and the record. Your master product data, your channel rules and your approval limits are settings your team can read and change, on your own accounts. The agent drafts, checks and publishes within those rules; your team approves what is new, what makes a claim and what cannot be undone. You keep your accounts, your data and the custom code we build; the Growth OS base is licensed to you. We never buy your inventory or sell under our own accounts, so you stay the seller on every channel.
Brand ownership matters on shared catalogs. Amazon's Listings APIs FAQ says brand owners have the highest precedence to update detail page data. Walmart's product details policy says Walmart picks the best available content for an item unless a brand owner or authorized reseller provides it. If you resell another brand, the agent treats detail page content you cannot control as something to watch, not to publish, and tells your team when that page changes.
What does deployment involve?
Ecomsellertool builds this agent. We deploy Growth OS on your own accounts, pull your catalog from every channel into one master record, write each channel's rules into it, and add model steps where rules break down: drafting channel copy, mapping attribute values and spotting drift. New listings and claims go to your team for approval. Scope, price and go-live date are fixed in an architecture doc first.
- Connect. Each marketplace account is authorized through that marketplace's own app authorization, so nobody shares a password.
- Build the master record. Live listings from every channel are matched by identifier and SKU, and every fact that disagrees goes on one list. That conflict list is worth having before anything is published.
- Write the rules. Field mapping per channel, length limits, banned wording and approval thresholds, set with your catalog team.
- Run in draft mode. The agent drafts and validates but publishes nothing, and your team compares its drafts with what they would have written.
- Go live channel by channel, starting where drift costs you most.
We have built on Amazon's seller APIs since 2017, and the pieces come from systems we have shipped. For Ergode we built Amzster, a product-data analysis system: rule sets score title, bullet point and description fields against defined criteria and recommend improvements, and a review module filters top reviews for listing insight. It works from Amazon's merchant listings report, Brand Analytics search terms and catalog reports. The screen near the top of this page comes from Grow8, the AI listing software we built, and shows the kind of issue the agent looks for, including a description that names a different brand from the title.
For Ecom Global, a commerce platform built for 100M+ products, 20+ marketplaces and 13+ carriers, we built the marketplace abstraction layer that makes adding a channel a configuration and adapter job. We also built per-marketplace content rules for tone of voice, keywords to include or avoid, title length and SKU format, with operators able to override any field per marketplace. The catalog agent brings the same rules to your accounts, with your team approving what goes live.
Where should you start?
Pick your 20 best sellers and pull their titles, key facts and identifiers from every channel into one sheet. Count the facts that disagree. If you find more than a handful, or nobody can say which version is right, your catalog has no master record. 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. To map Walmart, TikTok Shop and eBay as well, schedule a call from the card below.
Frequently asked questions
Can it list my Amazon products on Walmart automatically?
It drafts the Walmart version from your master record, validates it against Walmart's item spec and content policy, and publishes it once your team approves the first version. Copying the Amazon listing field for field rarely passes: Walmart wants 3 to 10 key features of up to 80 characters, a one-paragraph description of at least 150 words and no references to other retailers. Walmart also holds sellers responsible for reviewing AI-generated content before it is published.
Will Amazon's new multichannel tools in Seller Central do this for free?
Partly. Amazon announced on September 24, 2026 that sellers will be able to enhance a description once on the Manage Products page and have it update everywhere it is live, formatted for each channel. Amazon says the experience starts with connecting eBay, Shopify, TikTok and Walmart accounts, importing and linking listings, and viewing orders; it rolls out gradually to US sellers at no additional cost. If your catalog lives in Seller Central, try it first. An agent fits when the master record lives outside Amazon, or when you need your own per-channel rules, claim checks and approvals.
Do we need a PIM before we start?
No. The master record can live in a PIM such as Salsify, in an ERP, in a spreadsheet or in the Growth OS catalog. If you already run a PIM, it stays the record: the agent reads from it and handles the per-channel mapping, validation and drift checks.
What happens when Amazon rewrites one of our titles?
Amazon said that after July 27, 2026, titles still over 75 characters are updated gradually to its AI recommendation. The agent's scheduled drift check spots the change, compares it with your master record and proposes one of two things: accept Amazon's wording into the record, or submit your own 75-character title with the extra detail in Item Highlights. Your team picks. Amazon says listings stay active while titles are updated.
Which marketplaces can it publish to?
We have shipped integrations for 20+ marketplaces, including Amazon, Walmart, Shopify, eBay and TikTok Shop. During scoping we confirm which channels the agent will publish to on your accounts and the access each one needs, and we build any missing connector on top of Growth OS.
Does the AI write our listing copy?
A model drafts each channel's version from facts your team has approved, and fixed rules check every draft for length, banned wording and missing attributes. A person approves the first version of every listing and every new claim. After that, re-publishing text that was already approved, to fix drift, can run on its own if your rules allow it.
How long does it take to deploy?
It depends on how many channels, product types and systems you connect. We scope the agent and any custom connectors in an architecture doc with a fixed price and go-live date before work starts, so you know both before you commit.
