Related products that actually fit
Most "you may also like" widgets guess. Set related-product rules in plain words, like same brand or parts that fit, and measure what they really sell.
Feeds
The channel reads your feed, not your store. One prepared catalog, rules written once, outputs per channel, and labels built from what the pixel says sells.
Shoppers rarely see your store first. They see your feed: the file that tells Google Shopping, Microsoft, Meta, TikTok and now the AI shopping assistants what you sell. If the title is vague, the identifier is broken or the image is missing, the channel does not show the product, and no amount of bidding fixes it. This post explains how a store runs one prepared catalog for every channel, and why the feed gets sharper when it can read your sales data.
A product feed is a structured list: one row per product or variant, with a title, description, price, availability, image, brand, identifier and landing page. Each channel has its own required columns and its own rules about what it will accept.
The channel reads that row, not your product page. If the row says "Jacket M blk" and the shopper searched "black waterproof jacket medium", the match is weak. If the barcode is missing a leading zero because a spreadsheet stripped it, the product is disapproved. If the variant image points to the parent's generic photo, the black wheel shows a silver one.
Every one of those is a feed problem, not a store problem, and feed problems are quiet. Most channels will not alert you when a product is disapproved. It simply stops appearing.
Most stores end up with a separate app for each channel. Five apps, five sets of rules, five places a title can be wrong. When you fix the brand name in the Google feed, the Meta feed keeps the old one. When you add a new channel, you rebuild the same logic a sixth time.
Orkie's AI data catalog feeds take the other approach. Your catalog comes in once from Shopify or BigCommerce with read access, including collections, tags, metafields, variant images and custom fields. It becomes one prepared master. Rules run against that master. Then each channel gets its own output, shaped the way that channel wants.
Nothing in your store changes unless you approve a write-back.
There are two layers.
Preparation rules run before a product enters the master. Drop the boilerplate description that every product in the catalog repeats. Keep the description for clearance items. Never send a zero-price row. These rules can only touch plain text fields; they can never remove identity, images, variant SKUs or the fitment and attribute data that search depends on. A product is checked again after the rule runs, so an aggressive rule cannot break a row.
Output rules run per channel, for the things that genuinely differ:
| Channel | What its output tends to need |
|---|---|
| Google Merchant Center | The standard Shopping columns, custom labels, availability kept for out-of-stock items |
| Microsoft Merchant Center | Nearly the same columns, a few field differences, sale price sent so the strike-through shows |
| Meta catalog | Product sets, so dynamic ads can target a segment rather than everything |
| TikTok catalog | A focused feed, often top sellers, with proper IDs and tracking parameters |
| ChatGPT shopping | A product feed with clean landing URLs, delivered over the channel's file drop |
| Partners and marketplaces | A hosted CSV or URL with exactly the columns they asked for |
Each output can be one row per product or one row per variant. Each has its own field mapping with IF and ELSE cases: if the variant title is the default, use the product title, else combine them; if the variant has no image, fall back to the parent's. Each has a preview that shows the output row beside the source row before anything ships. Each has its own schedule.
Add a channel and you add an output. The master and the preparation rules do not change.
Shoppers are now asking ChatGPT and similar assistants what to buy. Those assistants read product feeds too, and they read them more like a person than a keyword matcher does. A title with the brand, product type, key attribute and size in plain language does well. A title stuffed with codes does not.
The same feed that powers paid placements can power free shopping results in those assistants. That is a reason to keep tracking parameters out of the feed URL itself and let the channel stamp paid and organic clicks differently, so the pixel can tell them apart when the visitor lands.
Titles and descriptions can be rewritten for how people search, and where a value is uncertain the rewrite leaves the field blank rather than guessing. A blank barcode is a missing product. A wrong barcode is a disapproved one, and often a warning against the account.
This is the part a standalone feed tool cannot do. A standalone tool has never seen an order.
Because the feed and the pixel live in the same workspace, an output can use sales data as a rule input. Examples of what that lets you do:
Each Pixel condition has its own lookback. One output can require more than five sales in the last 60 days. Another can use 14 days. A third can ignore sales entirely.
The Merchant Center connectors pull account issues and product status changes back into the workspace, and can receive real-time notices when a product is disapproved or approved. Those events show up where the team works, not in a channel dashboard nobody opens. The operators at Orkie watch them as part of running the feeds, and the AI can find the products that are seen often but rarely clicked compared with their peers, and name the four things a shopper sees to check: image, title, price, shipping.
The Pixel goes in first. The feed engine reads it from day one, but the labels get sharper after a few weeks of sales history. Then the ads run on the feed's labels, and the whole loop is measured on the pixel rather than on what each channel reports.
If you run one store on four channels, this is the difference between maintaining four feeds and maintaining one catalog. If you have ever found a product disapproved for a month without knowing, it is also the difference between finding out and not.
Want to see what your feed looks like with sales data in it? Talk to us.
Want this for your business? Talk to us How the pixel works
Most "you may also like" widgets guess. Set related-product rules in plain words, like same brand or parts that fit, and measure what they really sell.
What AI assistants read, how to feed them one clean catalog, and how to see the sales ChatGPT, Gemini and Perplexity send you.
The pixel steers the feeds, measures related products and tells the Outbox what each shopper left behind. CRM lists become ad audiences. The company brain keeps learning.
One or two a month. No hype, no spam.