First, confirm the size and the status
Open the Audiences table and read three things for the audience in question: its status (still populating, ready, or flagged), the estimated size or floor label, and when it was created or last updated. A website audience created this morning is not small, it is new; it fills as qualifying events arrive within its window. A customer list updated months ago is small because it is stale. An audience that shows a floor label rather than a number is below the size Meta reports precisely, which is a signal in itself.
Then trace the source. For a website audience, open the event behind the rule in Events Manager and see how many times it fired over a period equal to the window. For a customer list, check which identifier columns were uploaded and how the file was formatted. That one comparison usually tells you whether the problem is the data or the definition, and it keeps the fix away from the audience-source catalogue that the custom audiences guide already covers.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Customer list shows a floor label despite a long file | Identifier columns uploaded; formatting against the template | List poorly formatted or carrying few identifiers | Add identifier columns and reformat to the template |
| Website audience stays tiny while overall pixel traffic is healthy | Count of the rule's event over a period equal to the window | Rule built on an event or URL that rarely fires | Rebuild the rule on an event that fires reliably, or fix the event |
| Audience shrinks on quiet weeks and grows on busy ones | Window length against weekly qualifying visitors | Retention window too short for the traffic | Lengthen the window within Meta's cap, or merge rungs |
| Audience much smaller than analytics visitors for the same pages | Consent acceptance and blocked-request share | Consent and browser restrictions removing visitors | Accept the floor; add server events for consented actions |
| Audience exists but ad sets struggle or a lookalike cannot be created | Floor label; lookalike source requirement | Audience below what Meta needs to deliver or seed | Merge audiences, widen the source, or choose a different seed |
Fix, verify and prevent each cause
Customer list with few identifiers
A file of email addresses alone, or of work addresses that people do not use on Facebook or Instagram, matches poorly. Formatting errors make it worse: country codes missing from phone numbers, names in one column, extra spaces.
Fix: add every identifier you lawfully hold for each person, such as phone, first name, last name, city, postcode and country, in the columns and formats the upload template shows.
Verify: the replacement audience shows a larger estimated size than the original list.
Prevent: export customer lists from a saved report that already uses the template's columns, as the matching section of the custom audiences guide describes.
A rule that rarely fires
An audience built on a URL that changed in a redesign, a custom event nobody sends any more, or a parameter condition too strict to match many visitors, will stay small however much traffic arrives.
Fix: rebuild the rule on a standard event that fires reliably, or repair the event itself following the Meta Pixel setup tutorial.
Verify: the event count in Events Manager over the window is in line with the visits you expect, and the audience grows over the following days.
Prevent: link each website audience to a named event in the tracking plan, so a change to the event is reviewed against the audiences that depend on it.
A window shorter than the traffic can fill
The retention window keeps visitors for a set number of days. A short window suits a busy site and starves a quiet one, and a RemarketingAdvertising to previous visitors or engagers. ladder with many short rungs splits a modest flow into pools too small to deliver.
Fix: lengthen the window within Meta's documented cap, or merge adjacent rungs into one audience.
Verify: the audience holds a steadier size across busy and quiet weeks.
Prevent: choose windows from weekly qualifying visitors rather than habit, using the guidance on windows in the custom audiences guide.
Consent and browser limits
Visitors who decline advertising tracking produce no pixel events and never enter website audiences, and blocked requests have the same effect. That is a floor, not a fault.
Fix: confirm that consented visitors are captured reliably, and send server events through the Conversions APIServer-side event sharing with Meta. for consented actions so blocked browsers do not also remove them.
Verify: the audience size moves closer to consented traffic for the same pages.
Prevent: expect regional differences and plan audiences around them, as the privacy and consent guide explains; better matching also helps, which the low event match quality page covers.
Below what Meta needs
Some audiences are small because the business is small or the behaviour is rare. Meta shows a floor label below a threshold and documents a minimum seed size for lookalikes.
Fix: merge related audiences into one source, use a broader behaviour as the seed, or choose a value-based customer list.
Verify: the lookalike can be created, or the ad set delivers without stalling.
Prevent: plan seeds that will stay above the minimum, following the seed advice in the lookalike audiences guide for every Lookalike AudienceAudience similar to a seed list. you depend on.
Prevention: keeping audiences fed
Audiences shrink silently. A pixel event breaks on a template change, a customer list is never refreshed, or a window that suited peak season starves the quiet months, and nobody looks until an ad set stops delivering. A light routine catches each of those early.
Keep a register of the audiences in use, with the source, the rule or file, the window and the ad sets that depend on each. Once a month, read the Audiences table for status and size and compare the size with the previous month; a sharp fall points at the source, not the audience. Refresh customer lists on a schedule that matches how quickly the customer base changes, replacing the file rather than appending to it so that lapsed customers leave; the step-by-step customer list upload sets out the export, column formatting and replacement routine that keep a list matching well. After any site release, check that the events behind website audiences still fire, using the same test you run for conversion events. Before launching a new ladder, add up the expected qualifying visitors per rung and merge rungs that would sit at the floor. Finally, if an audience is used as an exclusion, its size matters in the other direction: a stale purchaser list lets existing customers back into prospecting, a problem covered on the Advantage+ audience reaching existing customers page.
A yoga studio whose retargeting pool kept vanishing
A yoga studio builds a website audience of people who viewed the class timetable in the last 7 days. The site gets about 1,900 visits a week, but the audience shows a floor label and the ad set barely spends. Events Manager shows the custom event behind the rule fired only 140 times last week: a redesign moved the timetable to an embedded booking widget, so most visits no longer load the tracked URL. The studio rebuilds the audience on the standard ViewContent event, which the booking widget sends, and lengthens the window to 30 days. Within three weeks the audience reports a usable size, and the uploaded member list, re-exported with phone and postcode columns, grows from a floor label to a figure large enough to seed a lookalike.
Common mistakes
Judging a newly created audience before it has had time to fill. Appending to an old customer file instead of replacing it. Building ladders with so many rungs that each one starves. Blaming consent for a gap that is really a missing event. Forgetting that a shrinking exclusion audience lets the wrong people back in.
