Facebook Custom Audience Guide (2026): Sources, List Preparation, and Match Rate

Custom Audiences look simple from the campaign builder: pick a source, name it, use it. What decides whether the audience actually performs happens earlier — in how the source is prepared, how large the matched set turns out to be, and whether the audience is measuring what you think it is. This guide covers the four source types, the list preparation rules that determine match rate, and the mistakes that quietly make an audience useless.

The four source types, and what each one is really for

Customer lists. You upload identifiers you already hold. This is the only source you control end to end, and the only one usable for people who have never touched your site or app. Its ceiling is your data quality.

Website audiences. Built from pixel or Conversions API events — visitors, viewers of specific pages, people who completed specific actions. The most commonly used source and the one most affected by measurement gaps: if your tracking misses events, the audience is silently smaller than it should be. Getting the measurement layer right first is the prerequisite — see the Meta Conversions API setup guide and the Facebook Pixel setup guide.

App activity audiences. The mobile equivalent, built from SDK-reported app events. Requires the app to be registered and events to be defined before any audience exists.

Engagement audiences. Built from interactions with your assets on the platform itself — video views, form opens and submissions, page or profile interactions. These do not depend on your own tracking at all, which makes them the most reliable source available when your site measurement is incomplete, and the most underused for exactly that reason.

A practical ordering: engagement audiences work immediately, website audiences need working tracking, customer lists need clean data, and app audiences need SDK work.

Preparing a customer list so it actually matches

Uploaded identifiers are hashed before they leave your machine. Match rate — the share of your list that maps to real accounts — is decided almost entirely by preparation.

Use more than one identifier column. A list with email, phone and name matches better than a list with email alone, because any one column can fail on any one person.

Normalise before upload. Lowercase everything, trim whitespace, remove punctuation from phone numbers and prefix the country calling code, and use standard two-letter country codes. Inconsistent formatting does not error — it just does not match.

Split multi-value fields. One column per identifier type. A single "contact" column holding sometimes an email and sometimes a phone matches poorly.

Include country when you have it. It disambiguates phone numbers and common names, and it is cheap to add.

Expect a match rate well short of the full list. A realistic range for a well-prepared list is a large minority to a majority. A very low rate almost always means a formatting problem, not a data problem — check normalisation before concluding your customers are unreachable.

Minimum audience sizes apply before an audience can be used, and matched size, not uploaded size, is what counts.

Website audiences: the rules that decide quality

Retention window. How many days back the audience looks. Shorter windows mean more recent intent and smaller size; longer windows mean scale and staler signal. Pick per use case rather than defaulting to the maximum for everything.

Rule specificity. An audience of all visitors is a different asset from an audience of people who reached a specific step. The specific one is smaller and usually worth more, and it is what makes a sequence of audiences possible instead of one undifferentiated pool.

Refresh behaviour. Website audiences update continuously as events arrive; a customer list is a snapshot and goes stale unless you re-upload. Teams often forget the second half of that sentence and run a list for months after it stopped representing anyone current.

Exclusions matter as much as inclusions. Most wasted spend in audience targeting comes from not excluding people who already converted, or from overlapping audiences competing with each other. Structure audiences so each campaign has a clear inclusion and a clear exclusion.

Four mistakes that quietly make an audience useless

Building on broken tracking. A website audience is only as complete as the events feeding it. If measurement is lossy, the audience under-collects, and the effect looks like poor performance rather than a data problem.

Overlapping audiences bidding against each other. Two audiences containing largely the same people, running in parallel, raise your own costs. Check overlap before scaling, not after.

Never refreshing customer lists. A snapshot from six months ago targets who you served then, not who you serve now.

Treating a small matched audience as a big one. Uploaded size and matched size are different numbers, and only the second one is real. Decisions made on the first one are made on nothing.

A workable starting structure

1. One engagement audience as a baseline that works regardless of site tracking.
2. Two or three website audiences at different funnel depths, with distinct retention windows.
3. One customer list of existing buyers, refreshed on a schedule, used mainly as an exclusion and as a seed.
4. Explicit exclusions on every prospecting campaign so you are not paying to reach people you already have.
5. An overlap check before adding anything new.

That structure is small enough to reason about and covers the cases where audience targeting actually earns its keep.

Frequently asked questions

What is the minimum audience size? A floor applies before an audience can be used, and it is measured on matched size. Very small audiences also deliver poorly even when technically usable.

Why is my match rate low? Formatting, almost always: unnormalised phone numbers, mixed identifier columns, missing country. Check preparation before concluding anything about your customer base.

Do custom audiences work without a pixel? Engagement and customer-list audiences do. Website audiences do not.

How often should I refresh a customer list? On whatever cycle your customer base meaningfully changes. Monthly is a reasonable default; never is the common practice.

Should I use one big audience or several small ones? Several, at different funnel depths, with exclusions. One big audience gives you nothing to compare and nowhere to sequence.

The short version

Pick the source type by what you actually have working: engagement audiences need nothing, website audiences need reliable tracking, customer lists need normalised multi-column data. Judge lists on matched size rather than uploaded size, refresh them on a schedule, and set exclusions deliberately so your own audiences are not competing with each other.

DeepClick works with advertisers on the measurement and delivery layer that audience targeting depends on.