--- title: "Product Feed Generator: AI Fields, Rules & Validation - Serpwise" description: "Build channel-ready product feeds with AI enrichment, variant exports, reusable rules, 20 transformations, quality checks, and scheduled delivery." url: "https://serpwise.ai/ecommerce/product-feeds/" source: "https://serpwise.ai/ecommerce/product-feeds/" --- [Product feeds](#main) [E-commerce / Product feeds](/ecommerce/) # Your catalog. Every channel’s language. A product feed is more than a file export. Choose the products, shape every field, add AI-generated content, and validate the final result. Build a repeatable workflow for shopping ads, social catalogs, comparison sites, and custom destinations. [Get in touch](/contact/)[Explore the feed engine](#build) Illustrative feed configuration ## Summer collection / Google Shopping 01 Catalog One row per variant 02 Segment In stock + linen products 03 Title AI title → trim → limit length 04 Link Product URL → campaign parameters Preview → validate → publish Product feeds for the channels you sell on - Google Shopping - ![PriceRunner](/logos/commerce/pricerunner.svg) - ![Kelkoo Group](/logos/commerce/kelkoo.svg) - Meta - TikTok ## Start with the destination. Keep control of the details. Built-in templates give you a starting format, field map, and channel quality profile. Tailor the copied configuration to your catalog, then save your own reusable template. Your feed remains editable after you select a preset. ### Shopping & discovery Google Shopping · Google Free Listings · Microsoft / Bing ### Social catalogs Meta / Facebook & Instagram · TikTok · Pinterest ### Comparison shopping PriceRunner · PriceSpy / Prisjakt · Idealo · Kelkoo 01 / Build the right feed ## Choose the products. Then define the payload. Different destinations need different views of your catalog. Build each feed around its own sources, product selection, row structure, and output fields. 01 ### Select catalog sources Use all connected sources or select the ones a feed should include. Emit matched products once from the primary source, or keep one copy per selected source when your export needs that distinction. 02 ### Export products or variants Choose one row per product or one row per variant. Variant rows expose offer IDs and item-group IDs, so channels can distinguish individual offers while keeping related variants together. 03 ### Map fields your way Give each column its destination name. Read a catalog field, insert a fixed value, or reference a cached AI field. Map titles, descriptions, prices, availability, images, identifiers, taxonomy, attributes, and custom labels. CSV and TSV offer delimiter, quoting, header, and byte-order-mark options. XML supports root and item tags, namespaces, attributes, and wrappers. JSON can wrap the product array under your chosen root key. 02 / Rules & transformations ## Stop fixing the same data in every export. Combine nested AND / OR conditions to define which products belong in a feed. Filter by price, availability, brand, category, or other available fields, using numeric comparisons, text matches, sets, missing-value checks, and regular expressions. 01 ### Save a segment once Build an in-stock collection, a brand selection, or a price band and save it as a shared segment. Reference it from multiple feeds and Merchant Center. Updates are picked up on the next generation or sync. 02 ### Chain transformations in order Clean a value, add useful context, map it to a channel vocabulary, and apply the final length limit. Each step receives the previous step’s result, making the order explicit and repeatable. 03 ### Share the logic that matters Save a transformation chain and reference it across columns, feeds, and Merchant Center attributes. Keep an inline chain when a destination needs its own treatment. Reusable templates copy a configuration; shared rule references stay connected. Feed transformations change the exported values. Source write-back is a separate, deliberate action—adding a tracking parameter or shortening a feed title does not edit your webshop. ### 20 building blocks. Countless useful combinations. #### Clean text Trim · strip HTML · lowercase · uppercase · capitalize Remove storefront markup from a channel description. #### Control length Truncate · substring · pad Fit a title or identifier to the receiving format. #### Rewrite values Replace · regex replace · map value · default Translate stock labels into the channel’s accepted vocabulary. #### Build richer values Prepend · append · concatenate Combine a product name with useful attributes. #### Format numbers Format price · round · multiply · add Apply a numeric adjustment and format the price for the destination; this is not automatic currency conversion. #### Track destinations Append URL parameters Add a campaign or source parameter to product links. 03 / AI-generated feed content ## Create the fields your source catalog is missing. Define reusable prompts using product context, tone, target length, and campaign guidance. Generate synthetic product data such as short summaries, channel-specific titles, or descriptive labels, then use those values alongside ordinary catalog fields. 01 ### A field just for this feed Create a custom AI field and map it into any output column. For example, produce a concise product summary for a partner catalog while preserving the longer description used elsewhere. 02 ### An attribute used everywhere Target a shared attribute such as title, description, category, color, material, or product type. Choose fill-if-missing or overwrite. Shared enrichment can flow into feeds, Merchant Center, and supported storefront delivery paths. 03 ### Generate once. Reuse on export. Generated AI field values are cached per product. Rendering a feed reads the prepared values instead of making a new AI call for every export. Regeneration is a separate action, so scheduling exports does not itself regenerate the copy. Review AI-generated content against your product facts. Serpwise keeps source, manual, and AI provenance; Google XML can represent AI-origin titles and descriptions using the corresponding structured fields. 04 / Validate & improve ## Check the values the channel will actually receive. Quality checks run after mapping and transformation. A clean source value can still become invalid in an export; checking the final payload helps catch that before publication. 01 ### Check against the destination Versioned channel profiles can check required and recommended fields, accepted values, conditional requirements, lengths, URLs, prices, GTIN checksums, image dimensions when available, and consistency between variants. 02 ### Fix the affected destination Review errors and warnings by product and attribute. Preview a supported fix, apply it to the selected feed or Merchant Center account, and rerun the checks. A destination-specific fix does not silently rewrite the shared catalog. 03 ### Protect the last good version A generation produces a candidate before it becomes live. With blocking enabled, a failing candidate remains available for review and the last good feed stays live. In shadow mode, findings are recorded while candidates still publish. Preview shows a bounded sample through the same mapping and rendering path without publishing. It helps you inspect values; it is not a full-catalog quality verdict. Generic feeds can also use advisory checks for required columns, lengths, and patterns. 05 / Put it to work ## Four feeds you could build from the same catalog. Illustrative configurations using the feed engine’s existing controls. Expand a workflow to see how the pieces fit together. ### A variant-aware shopping feed \+ − Give a shopping channel distinct offers for sizes and colors, with more descriptive titles. 1. 01 / Choose Start from Google Shopping and use variant rows. Keep offer IDs and item-group IDs in the mapping. 2. 02 / Shape Use an enriched title, trim whitespace, and apply the destination’s length limit. Map price, availability, image, and identifiers from the catalog. 3. 03 / Deliver Preview sample rows, review the channel quality report, and publish a passing candidate. ### An in-stock campaign catalog \+ − Create a focused selection for a seasonal campaign without rebuilding your source catalog. 1. 01 / Choose Save a segment combining in-stock availability with your selected category, brand, or price range. 2. 02 / Shape Map a campaign label as a literal or AI field. Append campaign parameters to product URLs. 3. 03 / Deliver Reuse the segment across Meta, TikTok, or Pinterest feeds and give each its own schedule. ### A comparison-shopping export \+ − Use the same catalog for a comparison site with its own field vocabulary. 1. 01 / Choose Start with PriceRunner, PriceSpy / Prisjakt, Idealo, or Kelkoo and select the relevant source catalog. 2. 02 / Shape Map categories and availability values, format prices, and include the required product identifiers. 3. 03 / Deliver Inspect warnings and blocking errors for that profile, then provide the published URL to the receiving service. ### A custom partner feed \+ − Deliver an agreed data format when a standard channel preset is not the right fit. 1. 01 / Choose Choose CSV, TSV, XML, or JSON and define the partner’s required output fields. 2. 02 / Shape Add a short AI-generated summary, reorder columns, and configure delimiters, XML wrappers, or a JSON root. Use a custom Jinja template for a bespoke document structure. 3. 03 / Deliver Add advisory required-field, length, and pattern checks; preview the serialized output and schedule generation. 06 / Delivery & operations ## A feed workflow you can keep running. Move from an occasional export to a maintained delivery process. Choose the cadence, inspect each generation, and keep a clear view of what was published. 01 ### Manual or scheduled generation Run a feed when you need it, or choose hourly, every six hours, daily, weekly, or monthly presets. Custom cron schedules are available when your destination needs a different cadence. 02 ### Publication with a history Use the published feed URL for the destination’s fetch workflow. Generation history records product counts, output size, warnings, quality results, and whether the candidate published, so a completed run is not confused with a live update. 03 ### Alerts when the state changes For scheduled targets, owners and admins receive an email when a target first becomes blocked and when it recovers. Repeated failures do not produce the same alert on every run. Channel templates shape an export; they do not connect every advertising account automatically. Google Merchant Center also has a direct integration, using the same reusable segments and transformations. ## Flexible enough for your technical team, too. ### Custom document templates When column mapping is not enough, a custom Jinja template can arrange the already filtered and transformed records into a bespoke output structure. Keep your product selection and transformation logic in the same feed workflow. ### Agent-assisted configuration The MCP interface exposes the feed schema, fields, operators, templates, and examples. Authorized agents can inspect, create, update, and preview feed configurations. Creating a configuration does not itself generate or publish a feed. ## Product feeds, explained. 01 How is a feed different from source write-back? + - A feed exports a selected view of your catalog. Its column transforms and feed-only AI fields affect that destination. Source write-back applies supported fields to your connected store through a separate action. Product JSON-LD can be delivered at the edge independently of both. 02 Can one AI field be used in several feeds? + - Yes. AI fields are named prompt templates with cached values per product. Multiple feed columns can reference the same field. A prompt can also target a supported shared catalog attribute instead of remaining feed-only. 03 Can I export only some sources or variants? + - You can select source connections, choose product or variant rows, and filter the resulting catalog selection with rules or a reusable segment. Matched products can be emitted once or separately per selected source. 04 Do templates guarantee channel approval? + - No. Templates and channel validation help you prepare the right payload. The receiving channel makes its own eligibility and approval decisions. Required fields can also depend on the product and destination. 05 What happens when the next feed has errors? + - With blocking enabled, a candidate containing blocking errors is held for review while the previous live feed remains unchanged. In shadow mode it still publishes with findings recorded. Warnings can publish. Generation history shows the outcome. 06 Does every scheduled feed run spend AI generation credits? + - The feed renderer uses cached AI field values and does not call an AI model during export. Generating or regenerating AI content is a separate action from rendering the feed. ## Bring your catalog. Bring your hardest feed requirement. We’ll walk through your destinations, mappings, AI fields, quality checks, and delivery workflow with you. [Get in touch](/contact/)[Explore E-commerce Mode](/ecommerce/)