Meta purchases vs Shopify orders: a cross-check in Claude

Shopify says one number for last week. Claude, reading your ad account, says another. The tempting move is to decide which one is lying and stop trusting it.

Neither is lying. They are counting different things, on different dates, under different rules. The useful question is not which number is right, but where between the checkout and the report the orders went missing — and that has an answer you can find in about ten minutes, if you split the gap in two.

Two gaps, not one

A Shopify order becomes a Meta purchase in two steps. First the event has to reach Meta, from the pixel in the shopper’s browser, from Shopify’s server, or both. Then Meta has to credit it to an ad the shopper saw or clicked inside the attribution window. Anything lost in step one is tracking. Anything lost in step two is attribution.

A reply in a Shopify Community thread (started 3 January 2026, on a store where Ads Manager showed fewer purchases than Shopify) put it in one line: “Events Manager shows raw events received by Meta. Ads Manager only shows purchases that Meta can attribute to an ad.” The connector has a separate tool for each of those, and that is what makes the split possible from one chat.

• Events received: ads_get_dataset_stats returns event volume for a pixel or dataset, filterable to a single event such as Purchase and to WEB_ONLY (browser) or SERVER_ONLY (Conversions API). Its default window is the last 7 days and its schema caps the look-back at four weeks.
• Purchases attributed: omni_purchase from the entity reader, which the field catalogue describes as “the number of purchase events attributed to your ads”. The narrower offsite_conversion_fb_pixel_purchase counts only purchases tracked on your website by the pixel or Conversions API.

If you only sell through your own store, the website field is the closer match to Shopify’s online store orders. omni_purchase also includes purchases that happen directly on Meta’s own surfaces.

Checking a number against its source before acting on it is part of the weekly routine the Claude Ads Operator course teaches, alongside the habits in catching wrong numbers in Claude’s reports.

What each side counts

Two traps sit in the defaults. The entity reader measures the last four weeks when you name no dates, while the analytics examples in the Shopify connector’s tool description use a thirty-day window, so two “last month” answers can cover different days. And value is not revenue by another name: Shopify’s data sharing page says events that carry an order value use the order’s total price, “including duties, taxes, and discounts.” Compare that to Shopify’s total sales, not net sales.

The cross-check, in order

1. Fix one closed window and state it first

   Pick a full past week that ended a few days ago, so late conversions have landed. Give both connectors the same explicit dates. Never let either one default.

2. Shopify orders against Purchase events received the tracking gap

   Ask for the Purchase count from the dataset, then again with SERVER_ONLY and WEB_ONLY. Shopify says its Enhanced and Maximum sharing levels send the purchase event server to server through the Conversions API. If the server-only count is zero on either level, the integration is not doing what the setting says, and no amount of attribution analysis will fix that. Whether the stats count events before or after deduplication is not stated in the tool’s schema, so do not read a browser and server count of similar size as proof of double counting.

3. Events received against purchases attributed the attribution gap

   This gap is supposed to exist: organic orders, email orders and purchases outside the window all reach Meta and are never credited. Read the ad set’s attribution_setting before judging its size — it is an ad set field only, and a one-day-click ad set will credit less than a seven-day-click one.

   The trade: without an order ID you cannot say which orders are in the gap, only how big it is.

4. Compare the value last, against total sales if counts line up

   Once the counts make sense, check value. A value gap with matching counts is usually a definition problem — taxes, shipping, currency — rather than missing orders.

A vendor blog from TrackBee (Frank de Vos, 27 June 2026) tells readers that “a gap of 20–35% is normal with the default 7-day click + 1-day view window”. A single normal range blends the two gaps. A modest total can hide a broken server integration that attribution happens to paper over, and a large one can be entirely organic orders. Split it before deciding whether it is normal. The same thinking applies to the step after this one — attributed is still not caused, which is a separate question with its own tool.

The gap between Shopify and Meta is two gaps. Only one of them is a problem.

• Comparing yesterday to yesterday. The two sides date purchases differently, and the connector cannot switch Meta to purchase date. Daily mismatches are noise you cannot remove.
• Letting each tool pick its own window. Four weeks on one side and thirty days on the other produces a gap made of calendar.
• Asking Claude to match orders to ads. There is no order identifier on the Meta side of the connector. An answer that names which orders came from which ad is invented.
• Fixing attribution when tracking is broken. If events never arrive, changing windows only changes which missing numbers you look at. Check the dataset’s match quality once volume looks right.

Try this tonight Take last full week. Write down Shopify’s order count, then ask Claude, with the connector set up, for the Purchase events your dataset received in the same dates, split server and browser. If the server count is zero, open Shopify’s Facebook & Instagram app and check the data sharing level before touching anything in Ads Manager. If both arrive and the attributed count is far lower, that is attribution — note the ad sets’ attribution setting and leave it alone for now.