What a custom audience is and where the data comes from
Meta's documentation describes a custom audience as an audience built from people who have already had some contact with your business, as opposed to a core audience, which is assembled from Meta's own demographic and interest signals. The contact can be something you record (a purchase in your CRM, a visit to your site) or something Meta records on your behalf (a view of your video, a submitted lead form).
What all sources share is a matching step. Meta holds a graph of accounts and the identifiers attached to them. A custom audience is populated by comparing your data against that graph and adding the accounts that match. The quality of the audience therefore depends on two things you control: the quality of the identifiers you send, and whether the events reach Meta at all. The pixelAn ad platform's browser code that reports page views and events on a site back to the platform. and the Conversions APIServer-side event sharing with Meta. are the delivery mechanisms for website data, and the Meta pixel setup tutorial is the place to confirm they are sending what you think they are.
Custom audiences are created in the Audiences section of Ads Manager or Business Manager and attached to ad sets, where they can be included, excluded, or used as the seed for a lookalike. The audience itself is a definition, not a frozen list: website and engagement audiences are recalculated as events arrive and as people age out, while customer list audiences stay fixed until you replace the file.
Sources
Meta groups the sources into your own sources and Meta sources. The table below sets out the ones most advertisers use.
| Source | What it captures | Data needed |
|---|---|---|
| Website | People who triggered a pixel or Conversions API event on your site: page views, specific URLs, standard events, events with parameter rules | A working pixel or Conversions API connection with event matching parameters |
| App activity | People who took an action in your app recorded by the Meta SDK or app events API | SDK or server-side app events, with the app registered in your business |
| Customer list | People matched from identifiers you upload: email, phone number, name, postcode and others, with an optional value column | A file of hashed or hashable identifiers and a declaration of where the data came from |
| Catalog | People who interacted with items in your product catalog, used with catalog ads | A catalog in Commerce Manager connected to pixel or app events |
| Offline activity | People matched from in-store or call-centre events uploaded as an offline event set | Offline events with identifiers and timestamps |
| Video | People who watched a share of one or more of your videos on Facebook or Instagram | The videos, chosen by watch depth |
| Lead form | People who opened, submitted, or opened but did not submit an instant form | Instant forms on your page |
| Instant Experience | People who opened or clicked within an Instant Experience | The experiences you published |
| Facebook page | People who follow, visited, engaged with posts or ads, messaged, or saved content from your page | A page you administer |
| Instagram account | The same interactions on a professional Instagram account | A professional account connected to the business |
| Events | People who responded to or engaged with your Facebook events | Events hosted by your page |
| Shopping | People who interacted with your shop on Facebook or Instagram | A shop set up in Commerce Manager |
The exact list in Ads Manager shifts over time; Meta adds and retires sources, so check the Audiences interface rather than assuming a source still exists.
Website and app audiences: events, URL rules, windows
Website audiences are the core of most retargeting. They are built from the events your pixel and Conversions API send, filtered by rules you set. The simplest rule is all visitors within a window. More useful rules are built on specific URLs (contains or equals a string, which lets you capture a category, a pricing page or a checkout step), on standard events such as ViewContent, AddToCart, InitiateCheckout and Purchase, and on event parameters, so that an audience can be people who viewed content in a certain category or purchased above a certain value. Meta also documents audiences built from the most active share of visitors by time spent, which is a crude but useful proxy for interest.
Rules can be combined with include and exclude logic inside one audience (visited the basket but did not reach the confirmation page), which avoids the need to manage the subtraction at the ad set level. The exclusions guide covers when to subtract inside the audience and when to subtract at the ad set.
The retention window is the number of days back from today that a qualifying visit counts. Meta caps this window for website audiences; the maximum is documented in the Help Centre and is shorter than the maximum for engagement audiences. Within the cap, the right window depends on the buying cycle and on volume. A short window gives a hot, small audience that is quickly exhausted; a long one gives a larger, cooler audience whose conversion rate (CVRShare of visits or clicks that convert.) declines as the older cohorts dilute it. Most practitioners run several windows for the same behaviour and ladder them, which the uses section below describes.
App audiences work the same way with app events in place of web events. They require the Meta SDK or app events API to be sending events from the app, and they have their own set of standard events for installs, launches and in-app actions. If an app and a website both exist, the two are separate audiences unless you deliberately combine them in an ad set.
Customer list audiences: matching and hashing
A customer list audience is built by uploading a file of identifiers that Meta compares against its accounts. Meta documents a set of accepted identifier types; email and phone number are the ones that match most reliably, and adding more identifier columns for the same person (first name, last name, city, postcode, country, date of birth) improves the chance of a match because Meta can use the combination. The upload template in Ads Manager shows the accepted columns and the formatting each expects. The tutorial on uploading a customer list takes a file from the CRM export through formatting, hashing and the data origin declaration to a scheduled replacement.
Meta states that identifiers are hashed before they leave your environment. You can hash them yourself using the documented algorithm and formatting rules, or let the upload tool hash them in the browser. The hashed values are compared with Meta's own hashed identifiers, matches are added to the audience, and Meta documents that the uploaded hashes are not retained once matching is complete. This is the mechanism that lets a list be used without disclosing readable data, but it does not relieve you of the obligations that come with the data: Meta's Custom Audience terms require that you have the right to use the data for this purpose and, on upload, that you declare whether it was collected directly from customers, from partners, or both.
Match ratesShare of the records in an uploaded customer or account list that an ad platform matched to its own users or companies. are not shown as a precise figure. The audience shows an estimated size, and small audiences display a floor label rather than a count. In practice, the share of a list that matches depends heavily on the identifier quality: a list of business email addresses matches poorly because people register on Meta with personal addresses, while a consumer list with phone numbers and emails matches far better.
A customer list can carry an optional value column, which Meta uses to build value-based audiences and, more usefully, value-based lookalikes. The lookalike audiences guide covers what the value column changes about the model; here it is enough to say that if you have lifetime value or margin per customer, include it.
Engagement audiences: page, Instagram, video, lead forms
Engagement audiences are built entirely from Meta's own records of how people interacted with your presence, so they need no pixel and raise no matching question: everyone in them is already a known account. They are the easiest audiences to build and the most underused.
Page and Instagram account audiences can be defined by anyone who engaged, by people who visited the profile, by people who engaged with a post or ad, by people who sent a message, and by people who saved content. Followers are a separate option. The engagement definitions include reactions, comments, shares and clicks, so they capture a wide and shallow set of people.
Video audiences are defined by watch depth across one or more chosen videos, from a brief view up to near-complete viewing. Deeper watch thresholds produce smaller and warmer audiences. Because video views accrue quickly on paid reach, a video audience can grow to a useful size within days of a campaign starting, which makes it a natural second step after an awareness buy.
Lead form audiences are built from people who opened a form, submitted it, or opened it without submitting. The last of those is one of the sharpest retargeting audiences available: people who got as far as the form and stopped.
Engagement audiences have a longer maximum retention window than website audiences, which makes them a good fallback where consent rules or ad blockers thin out the pixel data. Their limitation is intent: reacting to a post is a weaker signal than adding a product to a basket, and the audiences should be treated as consideration-stage rather than conversion-stage material.
Uses: retargeting ladders, exclusions, seeds for lookalikes
Custom audiences do three jobs, and a mature account uses all three.
Retargeting ladders. Rather than one retargeting ad set, build a sequence of audiences by recency and intent and give each its own message. A typical ladder, moving down the funnel: engaged with content but never visited; visited but viewed no product; viewed a product; added to basket; started checkout. Each rung excludes the rungs below it so a person sits in exactly one. The messaging shifts from reminder to reassurance to incentive as intent rises. The See-Think-Do-Care framing maps directly onto the rungs and is a sensible way to name them. Because each rung is small, FrequencyAverage times a person saw an ad. builds fast; the reach and frequency calculator is a quick way to see how long a budget can run against a given audience before the same people see the ad too often.
Exclusions. Purchase-event audiences, customer lists of existing customers, and recent lead submitters are excluded from prospecting and from the rungs of the ladder they have passed. This is the cheapest improvement most accounts can make and is the subject of the exclusions and overlap guide.
Seeds for lookalikes. The best custom audiences are also the best seeds: purchasers, high-value customers from a list, lead submitters. The retargeting value of an audience and its value as a seed are different things, and an audience that is too small to retarget profitably may still be an excellent seed.
A furniture retailer rebuilds its retargeting around windows
A furniture retailer had a single website audience of all visitors in the last 30 days and one ad set showing a generic sale message to it. The audience held about 60,000 people, frequency climbed past 6 within two weeks, and the cost per purchase was roughly £95. The rebuild split the audience into three rungs: product viewers in the last 7 days (around 9,000 people), basket abandoners in the last 14 days (around 2,500) and a 30-day catch-all that excluded both. Purchasers from the last 90 days, pulled from a customer list with order values, were excluded from all three. The basket rung carried a delivery-cost message, the viewer rung carried reviews, and the catch-all carried the sale. Across the first month the blended cost per purchase fell to about £62, frequency on each rung stayed under 3, and the purchaser list with its value column became the seed for a value-based lookalike for prospecting.
Size, freshness and privacy constraints
Three constraints shape what custom audiences can do.
Size. Meta shows an estimated audience size and, below a threshold, only a floor label. An audience can be used in an ad set below that threshold, but delivery will be constrained and the ad set may struggle to leave the learning phase. Lookalikes also need a seed above a minimum count, which Meta documents in the lookalike Help Centre pages. Very small audiences are better merged with a neighbouring rung than run alone.
Freshness. Website and engagement audiences refresh as events arrive, but only as long as the events keep arriving. A pixel that breaks on a site redesign stops feeding the audience, and the audience shrinks silently as people age out of the window. Customer list audiences are static and need a replacement upload on a schedule. Meta also documents that audiences unused for a long period may be flagged or removed, so check the Audiences table for availability notices.
Privacy and consent. Where consent regulation applies, visitors who decline tracking do not generate pixel events and do not enter website audiences; Meta's documentation on consent and data-processing options covers how to configure limited data use. Customer lists must be collected with a lawful basis that covers advertising use, and Meta requires you to declare the data origin at upload. Certain ad categories (Meta's special ad categories covering credit, employment, housing and social issues) restrict which audience options are available, including some custom audience and lookalike uses; the restrictions are documented and are enforced at ad set creation. Health is a further sensitive case, and the healthcare provider marketing guide explains why audiences built from condition or treatment pages need particular care.
None of these constraints stops custom audiences being the backbone of retargeting. They do mean the audiences need checking on a schedule rather than being set once, which is the habit that the targeting hub recommends for every audience type and the author's remarketing expertise page describes in practice.
When an audience refuses to grow large enough to deliver or to seed a lookalike, the diagnosis page for a custom audience that is too small traces the shortfall to the list, the rule, the window, consent or the minimum size.
Engagement audiences are the main output of a consideration campaign, and the Meta Ads setup for engagement and consideration shows how to build them and follow them up.
Common mistakes
- One window for everything. A single long-window audience mixes hot and cold intent and gets one message; ladder it.
- No exclusions between rungs. Without exclusions, a basket abandoner sits in the viewer audience and the catch-all too, and sees three competing ads.
- Trusting the audience without checking the pixel. A website audience is only as complete as the events feeding it; confirm event coverage before blaming the audience.
- Uploading a list without a value column. The column is optional for matching and essential for value-based lookalikes; include it whenever the data exists.
- Ignoring engagement audiences. They need no pixel, survive consent refusals, and give a usable consideration-stage audience within days of a video campaign starting.
- Letting customer lists go stale. A list uploaded once contains customers from that date and nobody since; schedule the replacement.
- Using a tiny audience as if it were a large one. Below Meta's size floor, delivery is constrained and frequency runs away; merge small rungs.
