Skip to content

Guide

Meta CAPI Deduplication: Why Events Count Twice and How to Test a Pair

  • capi deduplication
  • meta event deduplication
  • event_id
  • conversions api
  • tiktok events api deduplication
  • pinterest conversions api
  • reddit conversions api
  • duplicate conversions

When you send the same purchase from the browser pixel and from your server, you are relying on one small matching rule to keep it from being counted twice. When that rule works, you get the best of both: the server copy fills in what ad blockers and Safari removed, and the platform keeps one event per purchase. When it does not, your conversions quietly inflate, and nothing in the ads interface tells you why.

We covered the symptoms and the full diagnosis in Facebook CAPI deduplication: why your conversions doubled. This post is the practical companion: the exact matching rule, the six ways a pair fails it, how the other ad platforms differ, and how to test a single pair of events before you trust a whole integration. The test uses our free CAPI deduplication checker, which runs in your browser and uploads nothing.

How Meta matches a browser event and a server event

Meta treats two events as the same action when both of these are true:

  1. The event names are identical. Purchase from the pixel and Purchase from the Conversions API. The comparison is exact, so purchase is a different event.
  2. The event IDs are identical. The pixel sends it as eventID (it appears as eid in the request), the server sends it as event_id. Again an exact string comparison.

The browser call:

fbq('track', 'Purchase', { value: 129.0, currency: 'EUR' }, { eventID: 'order-10482' });

The matching server call to the Conversions API:

{
  "event_name": "Purchase",
  "event_id": "order-10482",
  "event_time": 1727700000,
  "action_source": "website",
  "user_data": { "fbp": "fb.1.1727600000000.1234567890" },
  "custom_data": { "value": 129.0, "currency": "EUR" }
}

The two copies also have to arrive within 48 hours of each other. After that, Meta no longer looks for a match and the later copy is counted as a new event. Separately, Meta rejects any server event whose event_time is more than 7 days in the past, so a queue that falls far behind loses events outright.

Meta also documents a fallback that matches on fbp or external_id when no event_id is present. It is less reliable, because one browser can produce several different events of the same name, so treat event_id as the rule and the fallback as a safety net.

The best event_id is one you already have and that both sides can see: the order number for Purchase, a generated ID for everything else. Generate it once, in the browser or on the server, and pass the same value to both calls.

The 6 ways deduplication breaks

Almost every broken pair we see comes down to one of these:

  1. The event names differ. The pixel sends Purchase, the server sends purchase or CompletePayment. Often caused by one side using a platform app and the other a custom integration.
  2. The event IDs differ in format. Same order, different string: 10482 versus order-10482, ABC123 versus abc123, or a trailing space from a template. The strings must be byte for byte identical.
  3. The event ID is missing on one side. Usually the browser, because the pixel was installed by a theme or app that never passed eventID. Without it there is nothing to match.
  4. The copies arrive too far apart. The server call is queued and sent in a nightly batch, or retried after an outage. Anything later than 48 hours is counted again. Sending within minutes of the browser event is the safe target.
  5. The fbp value differs. The server sends an fbp it generated itself instead of reading the visitor’s _fbp cookie. This does not stop event_id matching, but it weakens identity matching and breaks the fallback.
  6. external_id does not match after hashing. The pixel hashes external_id with SHA-256 after normalising it. If the server sends it unhashed, hashes it without trimming and lowercasing, or hashes a different identifier, the two copies describe different people.

How TikTok, Pinterest and Reddit differ

The idea is the same on every platform: one shared ID sent by both sides. The field names and details differ, and that is where cross-platform setups slip.

Platform Browser field Server field Also compared
Meta eventID (sent as eid) event_id event_name, 48 hour window
TikTok event_id in the ttq.track options event_id event name, 48 hour window
Pinterest event_id in pintrk('track', ...) event_id event_name
Reddit conversionId in rdt('track', ...) metadata.conversion_id (API v3) event type

Two things catch people out:

  • Event names are not shared across platforms. A Meta Purchase is a TikTok CompletePayment (or Purchase in newer setups), a Pinterest checkout and a Reddit Purchase. Map the names per platform, and make sure the browser and server integrations use the same map.
  • Reddit uses a different field name on each side. conversionId in the pixel and conversion_id inside metadata on the server (Reddit Conversions API v3; the older v2.0 called the object event_metadata). Checking only one of them is a common gap. Reddit also spells event names differently on each side: Purchase in the pixel, PURCHASE in API v3.

Test your pair in the browser

Before you look at a week of numbers, test one pair. You need two things:

  1. The browser event. Open DevTools, Network tab, filter by facebook.com/tr (or your platform’s endpoint), trigger the event, and copy the request parameters. For Meta, Events Manager, Test Events also shows the browser event’s parameters.
  2. The server event. The JSON body your server sent to the Conversions API. Take it from your server log or from your integration’s debug output. Events Manager, Test Events shows server events too.

Then paste both into the CAPI deduplication checker, pick the platform, and run the check. It compares the event names, the event IDs (and tells you when they differ only by case or spaces), the time between the two events, fbp, and external_id after hashing. Each problem comes with the fix.

The checker runs entirely in your browser. Nothing you paste is sent to our servers, which matters because real payloads contain hashed customer data. You can confirm it yourself: the Network tab stays empty when you click the button.

If you do not know yet whether your browser events carry an event ID at all, the Meta Pixel and CAPI checker scans your store and reports it, without you copying anything.

Fix checklist

  • One ID per event, generated once, passed to both the pixel and the server call.
  • Order number (or checkout ID) used as the ID for Purchase on both sides.
  • Event names identical, including case, and mapped per platform.
  • No trimming, prefixing or reformatting of the ID on only one side.
  • Server events sent within minutes, and never more than 48 hours after the browser event.
  • fbp and fbc read from the visitor’s cookies and forwarded, not generated on the server.
  • external_id normalised (trimmed, lowercased) and hashed with SHA-256 the same way on both sides.
  • One pair per event type tested with the checker, and Test Events showing the event as deduplicated.

If this list looks like more plumbing than you want to maintain, that is the problem server-side tracking platforms exist to solve. AxiTrace, for example, sends the browser and server copies of every order with matching IDs out of the box. You can read more about event deduplication in our glossary.

Frequently asked questions

How do I know deduplication is working in Meta?

In Events Manager, open your dataset and look at the event’s details. When browser and server copies are matched, the event shows both sources and Meta reports it as deduplicated. In Test Events, send a pair with a test code and check that it appears once with both connection methods.

Should the browser or the server generate the event_id?

Either works, as long as the other side receives the same value. For Purchase, use the order number, which both sides know. For browser-first events such as ViewContent or AddToCart, generate the ID in the browser and pass it to your server along with the event.

Can I send only server events and skip the pixel?

You can, and then there is nothing to deduplicate. You lose the browser-only signals the pixel collects, such as the _fbp cookie being set for new visitors, so most setups keep both and deduplicate.

Why did my conversions double after I enabled the Conversions API?

Almost always because the pixel events carry no event_id, or the IDs do not match exactly. Test one pair with the checker and the cause usually shows up in the first result.

Written by

Chris Gaudy

Founder & CEO

Building accurate conversion tracking for e-commerce teams.

Get tracking insights in your inbox

Tips on server-side tracking, conversion optimization, and product updates. Delivered weekly.

No spam. Unsubscribe anytime.