← All updates

Engineering / Field note

Product feeds need identities, timestamps and permission

Product RSS, affiliate feeds, sitemaps and Merchant API serve different purposes. Five fictional examples show how to import useful data without confusing access, freshness or verification.

By Zyel · Published · 4 min read

Product feeds can make a catalog easier to maintain, provided the importer keeps three questions separate: which records can we access, when did their information change, and what have we actually verified? A file full of prices answers only part of that conversation. Choosing the right input saves downstream work; preserving its limits saves downstream embarrassment.

This guide explains existing interfaces, rather than announcing a new Google or Awin feature. The five examples use fictional sellers, products, prices and dates. The XML fragments illustrate public formats; the JSON and CSV examples are proposed internal records, not vendor API payloads or complete submission files.

Start with the access route

Google describes Merchant API as an interface for managing Merchant Center accounts, products and inventory. Its authorization documentation, updated October 26, 2026, ties permitted actions to roles within a Merchant Center account. An application must establish the appropriate account access; this is not a general discovery endpoint for arbitrary sellers’ Shopping listings. [2] [3]

A merchant supplying a product feed offers a different integration route. Before connecting it, establish the permitted use, the intended market and the update contract with that source. Our recommendation is to record those decisions alongside the feed configuration. A familiar XML format alone establishes neither access rights nor permission to republish its contents.

Example 1: Recognize a product-feed record

Google’s RSS product format extends RSS 2.0 with its Merchant Center namespace. The namespace declaration connects prefixed fields to that specification. This shortened item illustrates an identifier and a price with currency; a complete submission needs the other applicable attributes and the channel structure. [1] [8]

Fictional product-feed fragment: keep the namespace and currency · XML
<item xmlns:g="http://base.google.com/ns/1.0">
  <title>A5 notebook, single</title>
  <link>https://merchant-a.example/items/notebook-a5</link>
  <description>One fictional A5 notebook.</description>
  <g:id>SKU-7</g:id>
  <g:price>19.50 USD</g:price>
  <g:condition>new</g:condition>
</item>

The feed’s completeness contract matters too. Google instructs merchants to include their intended product set, rather than only recent changes: removing a product from that RSS source removes it from Merchant Center. An importer receiving a partial response should therefore avoid treating every missing row as a confirmed deletion. [1]

Example 2: Preserve the seller’s identity

Google requires stable product IDs, distinguishes variants and permits the same IDs in different Merchant Center accounts. That makes a useful engineering lesson: retain the source’s identity before attempting cross-seller matching. In this fictional pair, identical local IDs describe different pack sizes. Our proposed composite key preserves both observations without deciding that the products are equivalent. [7]

Fictional internal identity records: local IDs need a seller namespace · JSON
[
  {"seller":"merchant-a","source_id":"SKU-7",
   "catalog_key":["merchant-a","SKU-7"],"pack_count":1},
  {"seller":"merchant-b","source_id":"SKU-7",
   "catalog_key":["merchant-b","SKU-7"],"pack_count":12}
]

Keep the original identifier even if a matching process later links these rows to a shared product family. A title similarity score should not quietly erase differences in size, unit count or variant. That separation also makes a mistaken match easier to explain and reverse.

Example 3: Schedule from a change marker

Awin’s feed-list guide was published August 31 and updated November 4, 2026. It describes a CSV listing of feeds available to a publisher, with separate Last Imported and Last Checked columns. The guide recommends downloading feeds that have changed. Access requires a Publisher ID and a data feed key; that key differs from the Publisher API key. [4]

Fictional normalized manifest: a later check is not a later import · CSV
feed_id,advertiser_id,last_imported,last_checked
F-17,M-4,2026-11-20 08:00:00,2026-11-22 08:00:00

Here the checker ran later while the import marker stayed unchanged. Our suggested scheduler compares the relevant source marker and retains its last successful result. It should also distinguish an unchanged feed from a failed download. Confirm timestamp timezones before converting them; a displayed time without an offset should not acquire an invented UTC suffix. Treat credential-bearing download links as secrets when designing logs.

Example 4: Let sitemaps supply candidates

A sitemap identifies pages to consider. The protocol’s lastmod field describes modification of the linked page, not the time the sitemap was generated. Google says it uses that value when it is consistently and verifiably accurate. These are useful scheduling signals, but this record contains no price, pack size or stock claim. [5] [6]

Fictional sitemap: discover a page without inventing an offer · XML
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">
  <url>
    <loc>https://merchant-a.example/items/notebook-a5</loc>
    <lastmod>2026-11-20</lastmod>
  </url>
</urlset>

Route a discovered URL into an appropriately permitted inspection process. Avoid promoting it straight into a published offer. Google also notes that ordinary RSS or Atom sitemaps generally expose recent URLs; the word RSS therefore does not, by itself, promise a complete product inventory. [5]

Example 5: Keep freshness separate from verification

Google’s price specification requires the submitted amount and currency to match the landing page and checkout. For a comparison service or internal catalog, that suggests a separate review decision: confirm the relevant seller, product, pack, currency and current conditions before presenting an observation as verified. A successful download alone cannot perform that work. [8]

Fictional internal observation: three clocks preserve the evidence boundary · JSON
{
  "seller":"merchant-a", "source_id":"SKU-7",
  "source_updated_at":"2026-11-20T08:00:00Z",
  "fetched_at":"2026-11-22T08:00:00Z",
  "price":{"amount":"19.50","currency":"USD"},
  "availability_claim":"seller reports available",
  "verified_at":null, "publication_state":"needs_review"
}

The timestamps here are invented UTC values for this example, independent of the earlier manifest. The source update, our retrieval and a verification event remain distinct. An absent verification time is meaningful evidence of unfinished review; it should survive an import rather than being replaced by the latest download time.

Start an integration with a seller whose access and data contract are understood. Follow one record through identity, updates, matching and publication before adding more sources. Preserve failed fetches, uncertain matches and missing fields as visible states. A larger import is useful when the next person can still tell where each claim came from and what remains to be checked.

Sources + further reading

Source dates describe the original material. Publication above is the date this Zyel article went live.

  1. RSS 2.0 product data source specificationGoogle Merchant Center · Checked Nov 22, 2026
  2. Overview of Merchant APIGoogle for Developers · Checked Nov 22, 2026
  3. Merchant API registration, authentication and authorizationGoogle for Developers · Checked Nov 22, 2026
  4. Downloading a list of your product feedsAwin · Checked Nov 22, 2026
  5. Build and submit a sitemapGoogle Search Central · Checked Nov 22, 2026
  6. Sitemaps XML protocolSitemaps.org · Checked Nov 22, 2026
  7. Product ID attributeGoogle Merchant Center · Checked Nov 22, 2026
  8. Product price attributeGoogle Merchant Center · Checked Nov 22, 2026

Reader discussion

What would you ask next?

Questions, useful corrections and thoughtful disagreement are welcome. Comments are reviewed before they appear.

No reader comments yet.

10–2,000 characters. Plain text. Leave out passwords, private project data and personal contact details. A blank display name appears as Anonymous.

Complete the CAPTCHA before sending. Verification runs on your device.