Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

TikTok Marketing › TikTok Ads

TikTok Custom and Lookalike Audiences

How TikTok custom and lookalike audiences work: website, app, customer file and engagement sources, retargeting, exclusions, lookalikes and broad delivery.
Guideintermediate11 min readUpdated

What audiences do on a feed that finds people itself

On older ad platforms, targeting was how an advertiser found customers: choose the interests, demographics and lists, and the platform shows the ad to the people inside them. TikTok offers those controls, and the campaign structure guide lists them: location, language, age, gender, interests and behaviours, device, and custom or lookalike audiences built from pixel, app, customer file or engagement data, with exclusions for each, plus a targeting expansion setting that lets delivery go beyond the chosen interests when it predicts better results. The same guide makes the point that shapes this page: because the feed ranks every video for every viewer, the ranking does much of the audience finding, and a broad ad group with sensible exclusions usually gathers signal faster than a narrow one.

That leaves custom and lookalike audiences with three jobs. The first is sequence: reaching people who have already shown interest, on purpose, with a message that assumes it. The second is subtraction: keeping Prospecting budget away from people who have already bought. The third is instruction: a lookalike tells delivery whose characteristics to look for, which matters most when the account's own conversion history is thin. The TikTok Ads hub sets out targeting data alongside formats and bidding; this guide is about using the lists well once the first campaign is running, which the tutorial on launching a first TikTok Ads campaign deliberately keeps simple.

The See, Think, Do, Care framework is a useful frame for the rest of the page. See is the broad audience nobody has met yet; Think holds people who watched or visited; Do holds people close to acting; Care holds existing customers, who are usually excluded from acquisition and sometimes addressed on their own. Each audience below belongs to one of those stages and does one job there.

The sources and what each is good for

TikTok's Ads Help Center documents the sources, the size each audience needs before it can be used, and the retention periods each source allows. Those figures change and this page does not quote them; what follows is what each source is for.

SourceBuilt fromBest used forWatch for
WebsitePixel events such as page views, add to cart and purchaseRetargeting visitors and cart abandoners; excluding recent purchasers; seeding lookalikes from buyersOnly as complete as the pixel and the consent it runs under
AppActivity reported from the appRe-engaging lapsed users; excluding people who already installedDepends on app event measurement being in place
Customer fileA list from the business's own records, matched on hashed identifiersExcluding customers the pixel never saw; seeding from repeat or high-value buyersConsent, purpose and freshness of the list
EngagementPeople who watched videos to a chosen depth, or liked, commented, shared, followed or visited the profileCarrying viewers from a consideration flight to a traffic or conversion campaignShallow engagement makes a broad, weakly qualified list

Website audiences come from the pixel. The pixel and Events API tutorial ends by building the first of them, recent site visitors and past purchasers, so that retargeting and exclusions are ready before launch, and the traffic setup captures every visitor a traffic campaign brings as an audience for a later conversion campaign. A website audience can only hold the people the pixel recorded, so its size follows the consent rate as much as the traffic, as the privacy and consent guide explains.

App audiences come from app activity, and they matter most to the businesses the app installs setup is written for: people who installed and lapsed are worth a different message from people who never installed, and people who already use the app should not be shown an install ad.

Customer file audiences are the business's First-Party Data, uploaded and matched against TikTok's users. They reach customers the pixel never saw: people who declined consent, bought on another device or bought offline. Whether a list may be uploaded at all is a consent and purpose question, and the guide to customer lists across ad platforms covers the definition of a customer, the consent check, Hashing and match rates for every platform at once, including why a hashed identifier is still personal data.

Engagement audiences are TikTok's most distinctive source. The consideration setup explains that they can be built from people who watched a video to a chosen depth or who liked, commented, shared, followed or visited the profile, and that a list of people who watched most of a long explanation is more selective than one built from brief views.

A retargeting sequence from viewer to buyer

Remarketing on TikTok is less about chasing site visitors with the product they looked at and more about carrying people through a sequence the feed started. A workable sequence has three rungs.

The first rung is a broad prospecting ad group, or a consideration flight that the consideration setup describes, whose job is to be watched. Engagement audiences defined on the day it launches start collecting at once; defined afterwards, they collect nothing from the flight that has already run. The second rung is an ad group aimed at people who watched deeply or engaged, with creative that assumes they know the brand: a demonstration, an offer, an answer to the objection the comments raised. Its destination is the website, so it adds people to the website audiences. The third rung is aimed at website visitors and cart abandoners, close to the purchase, with the shortest message.

Each rung excludes the people the next rung holds, or two ad groups end up bidding for the same viewer and the comparison between them means nothing. Each rung also excludes recent purchasers. The rungs need not be separate campaigns; they are separate ad groups because each has its own audience hypothesis, which is the campaign structure guide's rule for when an ad group earns its place. If a rung's audience is too small to produce the results its ad group needs, merge it with the rung above rather than leave it learning.

A sequence also needs a measure of its own. Retargeting looks efficient in platform reports because it reaches people who were likely to act anyway, a point the prospecting and remarketing comparison makes in full. Judge it on what it adds in the business's own records, not on its reported cost per result beside prospecting.

Exclusions that stop paying for existing customers

Exclusions are the least glamorous use of custom audiences and the one that saves the most. A prospecting ad group with no customer exclusion spends acquisition money on people who already buy, and its reported results look better for it, because existing customers convert readily. The TikTok Ads launch QA checklist makes excluding recent purchasers a line to tick when the offer is for new customers.

The strongest exclusion combines sources. A website audience of purchasers holds the buyers the pixel saw; a customer file holds the ones it did not. Together they cover more of the customer base than either alone. Name each exclusion audience with its job and the date of its source, and refresh customer files on a routine the business owns, because a list uploaded once and forgotten slowly stops excluding new customers. The Meta guide to exclusions and overlap sets out the principle that each person should be eligible for one ad group at a time, following their stage; the principle transfers to TikTok unchanged, even though the controls have different names.

Exclusions have a visible cost in reports. With the easiest conversions removed, the reported cost per result of a prospecting ad group usually rises while the cost per new customer in the business's own records holds or falls. Read both, and use the CPA calculator on the business's own count of new customers, not on the platform's total.

Lookalikes and the seed they copy

A lookalike asks TikTok to find people who resemble a seed audience, and it copies whatever the seed has in common. A seed of every website visitor teaches delivery what a visitor looks like, including the many who browse and leave. A seed of repeat buyers, or the customers with the highest value, teaches it what a good customer looks like. The seed is the instruction, so choose it from the people the business wants more of, and make sure it is large enough for TikTok to use; the help centre gives the minimum and the size options.

The Meta lookalike guide explains the trade between a closer match and a wider reach, and why seed quality beats seed size; the reasoning carries over, though TikTok's options are its own. Three practical rules follow for TikTok:

  • Seed from buyers, not from engagement, when the goal is sales. An engagement audience copies people who watch, which suits a consideration flight and suits a sales campaign poorly.
  • Exclude the seed from the lookalike's ad group. The people in the seed are customers or near-customers and belong in an exclusion or a retargeting rung, not in prospecting.
  • Test a lookalike against broad delivery, not instead of it. A lookalike ad group that cannot beat a broad ad group with the same exclusions on cost per new customer is a narrower audience doing the same job.

Lookalikes earn most when the account's own conversion history is thin, because they give delivery a picture of the customer it has not yet had the chance to learn. As the account gathers conversions, broad delivery optimising for the same event often catches up, and the comparison is worth rerunning.

When broad delivery beats a defined audience

Broad delivery, in the campaign structure guide's sense of location, language and age plus sensible exclusions, is the default for prospecting on TikTok, and the launch checklist asks the builder to choose between automatic targeting and custom audiences deliberately rather than by habit. Broad usually wins when the budget is modest, because every narrow ad group splits the learning; when the optimisation event fires often enough for delivery to learn whom to show the ad to; and when the creative does the qualifying, naming who the product is for in its opening.

A defined audience earns its own ad group when there is a specific hypothesis: a retargeting rung, a market that needs creative in another language, a lookalike being tested against broad. Interest ad groups are the weakest case. They narrow what delivery can learn from and rarely beat a broad ad group with good creative, and an account with many of them is the pattern the diagnosis for TikTok ads that are not spending sees when ad groups stay in learning because each has too little budget or too few people. Audiences built in the business's own Business Center, as the Business Center setup tutorial recommends, stay with the business whichever of these approaches the account takes.

Advertisers who run Meta as well will recognise most of this; the Meta custom audiences guide describes Meta's equivalents, which differ in names and options but not in the jobs they do.

A running-sock brand's audience map (illustrative)

Illustrative example — not a real result

Felvrish Socks replaces interest ad groups with an audience map

Felvrish Socks, an invented running-sock brand selling through its own website, runs six interest ad groups (marathon, trail running, gym, cycling, hiking, sport fashion) at £20 a day each, all optimising for purchase. After a month, four are still learning and the blended cost per purchase in its own orders is £31. It rebuilds the account around stages instead of interests.

StageAd groupAudienceExcludes
SeeBroad prospecting, £85 a dayAdults in its delivery countryPurchasers from the pixel and the customer file
ThinkViewer retargeting, £20 a dayPeople who watched most of its long sock-fitting explainerSite visitors, purchasers
DoVisitor retargeting, £15 a daySite visitors and cart abandonersPurchasers
See (test)Lookalike, £20 a daySeeded from repeat buyers in the customer file, not all visitorsPurchasers, the seed itself

Over the next 4 weeks, judged on cost per purchase in the shop's own orders, the broad ad group lands at £22, the viewer rung at £18 and the visitor rung at £14. The lookalike comes in at £24 against broad's £22 and is paused, its budget moved to broad, with a note to rerun the comparison next quarter. The share of orders from new customers rises from 58% to 71%, which Felvrish reads as the purchaser exclusion working.

Common mistakes

  • Using audiences to narrow prospecting. Interest and audience splits divide the learning; broad delivery with exclusions usually finds new people better.
  • Defining engagement audiences after the flight. They collect only from the day they exist.
  • No purchaser exclusion on prospecting. The budget pays for customers who would have bought anyway, and the report flatters it.
  • Relying on the pixel alone for exclusions. Customers who declined consent or bought elsewhere are missing; a customer file fills the gap.
  • Seeding lookalikes from all visitors. The seed teaches delivery what a browser looks like, not a buyer.
  • Overlapping rungs. Without exclusions between them, retargeting ad groups bid for the same people.
  • Uploading lists without a consent check. Hashing does not make personal data anonymous; the business decides whether a list may be used before anyone uploads it.
  • Judging retargeting on its reported cost per result. It reaches people likely to act anyway; measure what it adds in the business's own records.

Frequently asked questions

Do I still need custom audiences if TikTok finds people on its own?

Yes, but for fewer jobs than on older ad platforms. Broad delivery is usually the better way to find new people, so custom audiences earn their place as exclusions that keep prospecting away from existing customers, as retargeting lists for people who have already watched or visited, and as seeds for lookalikes. An account that uses them only to narrow prospecting is spending effort where the feed's ranking already does the work.

How big does a TikTok audience need to be?

TikTok sets minimum sizes for audiences to be usable for targeting and for seeding lookalikes, and lists them with the retention periods each source allows in its Ads Help Center; check them there, because they change. The practical question is different: whether the audience produces enough results for its ad group to learn from. A retargeting list that cannot feed its own ad group belongs inside a broader one, or used as an exclusion instead.

Should I seed a lookalike from all website visitors or from buyers?

From the people you want more of, which is usually buyers and often repeat or high-value buyers. A lookalike copies whatever the seed has in common, and a seed of all visitors teaches delivery what a visitor looks like, including the many who never buy. A smaller seed of good customers is a clearer instruction than a large seed of mixed intent, provided it meets the minimum size TikTok sets.

Terms used on this page

Tools to try

Frameworks

Where this page is used

Sources

  1. TikTok Ads Help Center — TikTok for Business
  2. About TikTok Pixel — TikTok for Business

Related