Catalog Management Best Practices: 12 Rules for Clean Product Data
On July 27, 2026, Amazon cut its product title limit from 200 characters to 75 for every category except media — and started using AI to auto-rewrite any seller's title that doesn't comply. It's a useful illustration of a bigger truth: catalog rules aren't static, they're set by the channels you sell on, they change without much warning, and a catalog built on assumptions from two years ago is already breaking somewhere today.
These 12 rules are the ones we enforce across catalog engagements, ordered from foundational (get this wrong and everything downstream inherits the problem) to operational (get this wrong and you lose efficiency, but not integrity). Each stands alone as a citable rule; together they're a governance framework.
Channel-specific requirements, compared
Before the rules, the concrete numbers they're built around — because "optimize your titles" is useless advice without knowing what each channel actually enforces:
| Requirement | Amazon | Google Merchant Center | Shopify |
|---|---|---|---|
| Title length | 75 characters max (all categories except media, enforced from July 27, 2026); AI auto-rewrites non-compliant titles | 150 characters recommended max; ~70 characters visible on mobile | No hard platform limit; ~55-60 characters recommended for search snippet display |
| Unique identifier | UPC/EAN/ASIN required for most categories | GTIN required if manufacturer-assigned; disapproval risk if incorrect | Optional (barcode field), but required for most sales channel integrations |
| Description limit | 2,000 characters | 5,000 characters | No hard limit |
| AI-generated content disclosure | Amazon's own AI can rewrite titles; no separate seller-side disclosure required for AI-assisted copy at this time | Required: AI-generated titles must use the structured_title attribute with digital_source_type set to trained_algorithmic_media | Governed by Merchant Center rules if syncing to Google Shopping; no separate Shopify-native requirement |
Sources: Amazon Seller Central announcement, "Updates to improve your product titles," effective July 27, 2026; Google Merchant Center Help, product data specification and structured title documentation (support.google.com/merchants, accessed 2026).
Shopify is the outlier in that table for a reason worth naming: it has no independent shopping-ads channel of its own with hard character limits the way Amazon and Google do. Its own storefront and search snippet display are comparatively forgiving. But that flexibility disappears the moment a Shopify store syncs to Google Shopping, Meta catalogs, or a marketplace app — at that point, the strictest connected channel's rules become the effective limit for the whole catalog, whether or not Shopify itself enforces them. Treating Shopify's flexibility as "no rules apply" is how stores end up with beautiful on-site copy that gets rejected the moment it syndicates anywhere else.
The 12 rules
1. One SKU, one source of truth
Every product should have exactly one governed record — in a PIM or equivalent system — that every channel feed reads from. Editing a title directly in Seller Central without updating the source record guarantees drift within a few update cycles. This sounds obvious until you audit a real catalog: it's common to find the Shopify title, the Amazon listing, and the Google Shopping feed for the same product diverging within months of launch, each edited independently by whoever needed a quick fix that week, with no record of which version is actually current.
2. Fix a SKU naming convention before you scale, and never reuse an ID
A SKU should encode enough structure to be human-parseable (category, variant, sequence) without becoming so long it's error-prone to type. Once retired, a SKU ID should never be reassigned to a new product — Google explicitly warns against this because it corrupts historical performance data tied to that ID.
3. Match title length to each channel's actual current limit, not last year's
Write a canonical, information-dense title in your source of truth, then truncate or restructure per channel rather than writing separate titles from scratch. Amazon's front-loaded 75-character limit and Google's 70-character mobile visibility window reward the same discipline: put the decision-making details first. The practical failure mode here is treating "optimize the title" as a one-time project — Amazon's July 2026 change caught a large share of sellers with titles built to the old 200-character ceiling, and the ones without a governed source-of-truth title had to manually rewrite each listing rather than regenerating a shorter version from structured attribute data they already had.
4. Submit a correct identifier (GTIN/UPC) wherever one exists
Google reports that products submitted with a correct GTIN see an average 20% increase in clicks; an incorrect one causes outright disapproval. Never guess or fabricate an identifier — for genuinely identifier-less products, use the identifier_exists=false pathway instead of leaving the field blank or wrong.
5. Standardize attribute values before you scale, not after
"Navy," "navy blue," and "NVY" should resolve to one controlled value before a catalog reaches thousands of SKUs — retrofitting standardization across an existing large catalog is dramatically more expensive than enforcing it from the first import. Build the controlled vocabulary once, in the source system, not per channel. The cost asymmetry is worth internalizing: validating one new SKU against an existing controlled list takes seconds; auditing ten thousand existing SKUs for every variant spelling of every attribute value, after the fact, is a multi-week project that usually gets deprioritized indefinitely once the immediate pain (a channel rejection, an angry customer) passes.
6. Build taxonomy before you build attributes
Attributes are category-specific (a "sleeve length" field is meaningless for a blender), so a stable category taxonomy has to exist before attribute governance can be enforced consistently. Retrofitting taxonomy under an already-attributed catalog usually means re-mapping attributes twice — once to fit the new taxonomy, and again when the taxonomy itself gets refined based on what the first mapping pass revealed was actually needed.
7. Treat images as data, not decoration
Alt text, file naming, and minimum resolution requirements are catalog data with the same governance obligations as any text attribute — not an afterthought for the design team. Google disapproves listings for broken or non-compliant image links as routinely as it does for missing GTINs.
8. Set an explicit freshness SLA, especially for price and availability
Price and availability should sync in near real time or at minimum daily; a mismatch between your feed and your live checkout price is one of the most common causes of marketplace disapproval and account-level penalties. Descriptive attributes can run on a slower review cycle, but every field should carry a visible last-verified timestamp rather than an implicit, unknown age. A field with no timestamp at all is functionally unauditable — nobody can tell whether it's accurate today or was accurate three years ago and simply never revisited.
9. Give discontinued SKUs an explicit end state — never let them silently rot
A discontinued product needs a deliberate decision: redirect to its replacement, 404 cleanly, or convert to a "notify me" page — not an orphaned URL still indexed and still accumulating the wrong signals. Silent rot dilutes crawl budget and confuses both search engines and AI shopping assistants about what's actually purchasable. It also actively damages customer trust: a shopper who clicks through from a search result to a page that's quietly for a product that hasn't existed in eighteen months forms an impression about your entire catalog's reliability, not just that one SKU.
10. Score completeness numerically, not by gut feeling
A simple weighted score across required fields, per category, turns "our data quality feels okay" into a trackable number you can report against over time and use to prioritize which SKUs need attention first. Without a number, data quality work has no way to demonstrate progress — every conversation about whether the catalog is "getting better" stays anecdotal, and anecdote-driven prioritization tends to fix whatever broke most recently and loudly rather than whatever affects the most revenue.
11. Separate the system of record from the channel feed format
Your PIM's internal data model shouldn't be structurally identical to any single channel's feed spec — channels change their requirements (as Amazon just did) faster than you should want to restructure your core data model. A translation layer between source-of-truth and channel feed absorbs those changes without touching governed data.
12. Document exceptions instead of special-casing them silently
Every catalog accumulates edge cases — a category with no GTIN, a supplier who won't provide attribute X, a product that breaks the naming convention. Document why the exception exists and who approved it, in a place the next person auditing the catalog will actually find, rather than leaving a silent workaround that looks like an error six months later.
What a completeness score actually looks like
| Field group | Weight | Pass condition |
|---|---|---|
| Required channel attributes (title, price, availability, identifier) | 40% | Present, correctly formatted, matches live checkout |
| Descriptive attributes (material, size, color, specs) | 25% | Present in controlled taxonomy values, not free text |
| Imagery (primary + alt text) | 20% | Meets minimum resolution; alt text present and descriptive |
| SEO/structured data (meta fields, schema markup) | 15% | Present and populated from the same source record |
Weight these categories according to your own channel mix and category mix — a marketplace-heavy seller should weight required channel attributes even higher; a content-driven DTC brand might weight descriptive attributes and SEO more evenly. What matters is having a number at all, tracked over time, rather than debating data quality in the abstract every quarter.
A worked example: a SKU with a correct title, price, and identifier (40% earned in full), three of four expected descriptive attributes populated (25% × 0.75 = 18.75%), a primary image but no alt text (20% × 0.5 = 10%), and no structured data at all (0% of 15%) scores 68.75% complete. That single number tells a catalog ops lead exactly where to focus next — structured data and alt text, in this case — rather than a vague sense that "this product's page could probably be better."
A quick self-audit against these 12 rules
Before investing in new tooling, spend an hour checking where your current catalog actually stands. Pull 25 random SKUs and check each one against these questions:
- Does the product's title, price, and description match exactly across every channel it's listed on — or has it drifted?
- Is there a single place you could go to find the "official" current title for this product, or would you have to ask around?
- Does the SKU ID follow your stated naming convention, and is it still unique across your entire historical catalog, including discontinued items?
- If this product has a GTIN, is it validated (correct check digit, not a placeholder)?
- Is the "last verified" date on this record's price and availability less than 24 hours old?
- If this product is discontinued, does its URL do something deliberate (redirect, clean 404, notify-me page) rather than sitting live and orphaned?
A catalog failing more than one or two of these on a 25-SKU sample almost always has the same root cause: no single source of truth (Rule 1) and no completeness scoring (Rule 10) to make the drift visible before a channel penalty makes it visible for you.
What this looks like in practice
None of these 12 rules require exotic tooling — most catalog failures we see come from skipping rule 1 (no single source of truth) and rule 6 (attributes built before taxonomy), which then makes every other rule harder to retrofit. If you're starting a catalog management program from scratch, sequence matters: taxonomy first, source-of-truth governance second, channel-specific formatting last. Trying to fix channel-level formatting problems while the underlying taxonomy is still unstable just means redoing the formatting work twice.
We build the governance layer and translation pipeline that absorbs channel policy changes — like Amazon's new title cap — without touching your core product data. Talk to a team that's done this for 17+ years.
Talk to the Team →Frequently asked questions
What is the single most important catalog management best practice?
Maintain one system of record per SKU that every channel feed pulls from, rather than editing product data independently in Amazon Seller Central, Shopify, and Google Merchant Center. Without a single source of truth, the same product drifts into three different descriptions within months.
What is Amazon's current product title character limit?
As of July 27, 2026, Amazon requires product titles in all categories except media to be 75 characters or fewer, including spaces, down from a previous 200-character ceiling. A new 125-character "Item Highlights" field holds the details that no longer fit in the title, and is separately searchable.
Do I need a GTIN for every product on Google Shopping?
You need a GTIN for any product that has one assigned by the manufacturer; if you submit an incorrect one, Google disapproves the listing. For genuinely identifier-less products (custom, handmade, vintage), set identifier_exists to false and rely on brand plus MPN instead. Google reports that products submitted with correct GTINs see an average 20% increase in clicks.
How often should product data be refreshed?
Price and availability should sync in near real time or at minimum daily — stale pricing is a common cause of marketplace suspensions. Descriptive attributes and imagery can run on a slower cadence (monthly or quarterly review), but should still carry a "last verified" timestamp so staleness is visible rather than silent.