TikTok Ads Targeting, Custom Audiences and Exclusions

#01What is a useful audience hypothesis?

Write who has the problem, why the offer is relevant and what would count as evidence of fit. "Finance staff who manage approval exceptions may value a practical mapping workshop" is a hypothesis. "People aged 25 to 54" is only a demographic setting.

Translate that hypothesis into available controls without pretending they identify people perfectly. Interests and behaviours are proxies. A person watching business content may be curious rather than a buyer. A broad audience may still find relevant people when creative and optimisation are clear, but it is not guaranteed to do so efficiently.

Define the business question before opening audience menus. Otherwise the available controls can dictate a strategy without anyone deciding what the test should learn.

#02How should you approach broad targeting?

Broad targeting is not the absence of strategy. Keep necessary location, language, age and policy controls where supported and appropriate. Then decide whether additional narrowing is justified by evidence or merely by a desire to feel in control.

Use creative that identifies the audience problem clearly. A specialist demonstration can help viewers recognise whether the offer is for them. That does not replace targeting or privacy rules, but it makes the message less dependent on a perfect audience proxy.

Check the actual automation settings. TikTok's unified Smart+ direction allows different degrees of control, subject to the account flow. Record which settings are constraints, suggestions or expansion mechanisms. Do not assume an audience entry means the platform will deliver only to that exact group.

Compare broad and more constrained approaches with a clear outcome and enough evidence. If one receives cheaper submissions but worse qualification, report both. Delivery volume alone does not settle the business question.

#03What are Custom Audiences and their constraints?

TikTok describes Custom Audiences built from sources including customer files, website or app activity, engagement and supported partner sources. Its current Help page requires 1,000 total matched users to use a Custom Audience in an ad group.

That is not 1,000 rows in a spreadsheet. Some records may not match, some may be ineligible and available reach can be affected by factors such as opt-in status, age and location. Keep source count, matched count and estimated delivery reach separate in the planning document.

Choose the source based on the action you want. Recent relevant visitors differ from all historical visitors. Qualified leads differ from every form submission. A seed dominated by students, competitors or low-quality enquiries can teach the system the wrong pattern.

Do not invent universal audience-retention windows. Check the selected source's current options in the account and record the duration you choose. Refresh, expiry and source behaviour can differ. A screenshot from another advertiser's interface is not a reliable specification for your setup.

#04What privacy and data controls belong before upload?

Confirm that your organisation has the appropriate rights, notices and legal basis for advertising use of the data. A business relationship or an email address in the CRM does not automatically authorise every audience use. This is operational guidance, not legal advice; involve the relevant privacy owner.

Do not use scraped personal data or sensitive information to create audiences. Review source fields, file contents and the purpose of the list. Hashing an identifier is not the same as anonymising it or obtaining permission to use it.

Maintain a source register: system, inclusion logic, exclusions, owner, refresh schedule and deletion process. If someone opts out or a retention period expires, the advertising workflow needs a way to reflect that decision. A static upload can become stale quickly.

Apply access controls to audience assets and source files. Do not put raw customer lists in a shared creative folder. Use sanctioned platform and business workflows, and retain only what is needed for the approved purpose.

#05How should exclusions and retargeting work?

Use exclusions to prevent obvious waste or inappropriate messaging where the selected setup supports them. Current customers may not need a new-customer offer; recent converters may not need repeated acquisition ads. But check the actual objective and business context before excluding everyone with any prior interaction.

Retargeting needs a distinct message. Someone who watched a broad educational video may not be ready for an aggressive sales request. Match the offer to the signal. Website visitors who inspected a detailed evaluation page may need different information from casual profile visitors.

Avoid tiny overlapping ad groups. They can fragment budget and complicate interpretation. Do not assume every overlap automatically creates a specific auction penalty; simply recognise that overlapping observational groups make business analysis less clean.

Monitor creative relevance and frequency using available metrics. A repeated message can become unhelpful even if the audience remains technically eligible. Change the offer or pause delivery when the business case no longer holds, rather than using retargeting as an indefinite pursuit mechanism.

#06Where does keyword targeting fit?

TikTok's dedicated Search Ads Campaign uses keywords to reach people searching for related terms on the search results surface. It is a different intent model from broad feed discovery or interest proxies.

Group keywords by the question the creative answers. A general brand video may be a poor response to a specific how-to query. Align the opening, proof and landing page with the intent. Inspect the current available keyword and reporting controls instead of assuming another search platform's match types and rules transfer exactly.

Check eligibility and availability for the account and market. Organic TikTok search access does not prove that a paid Search Ads flow is available. Do not publish a universal country list without current authoritative confirmation.

Use search tests to answer an intent question, not to disguise uncertainty about the broader audience. If the query is informational and the offer asks for a sales meeting immediately, the mismatch may be commercial rather than a targeting failure.

#07Worked example: Ledgerline

All list sizes, audience settings and Competitor A-D observations below are illustrative. They are not recommended minimum business sizes or performance forecasts.

Ledgerline considers a workshop campaign. Competitor A uses broad business messaging, Competitor B retargets content, Competitor C advertises queries and Competitor D uses customer-style creative. Ledgerline cannot see their actual private targeting and does not pretend otherwise.

Walkthrough

1. The team defines a fit hypothesis around approval-exception ownership. Its creative makes that problem explicit without claiming to identify a viewer's job privately.
2. A proposed CRM audience has an illustrative 1,800 uploaded records but only 900 matched users. It cannot be used under the currently documented 1,000-matched-user requirement.
3. Ledgerline does not supplement the file with scraped contacts. It reviews lawful source growth and uses another eligible test approach instead.
4. A broad test retains necessary market and policy constraints. A separate custom-source idea remains on hold until rights, quality and delivery eligibility are satisfied.
5. Recent workshop registrants are excluded where supported to avoid repeating the same acquisition request. Existing customers receive a different education plan.
6. The analyst reviews qualified leads, not only submissions, and records automation settings so later results remain interpretable.

The useful lesson is that audience eligibility and audience quality are distinct. A large matched list can still be a poor seed, and a small relevant list may not yet be deliverable.

#08Common mistakes

Calling uploaded rows matched users. Preserve the actual platform counts and definitions.

Treating first-party data as automatically permissioned. Confirm purpose, notices, rights and retention.

Over-narrowing to imitate another platform. TikTok's controls and audience signals are not identical to professional-network targeting.

Using the same message for every stage. Retargeting and search intent need appropriate offers and explanations.

#09Try it yourself

Write two audience hypotheses and the business evidence each would produce. For every proposed source, record rights, quality, matched eligibility, refresh and exclusions.

Map the message to the audience signal. Ask whether the CTA is reasonable for that level of intent. Then check the account's actual targeting and automation controls without launching. If a control is unavailable, revise the plan rather than describing a workaround as supported targeting.

#10Quick quiz

1. Is a file with 1,000 contacts necessarily a usable Custom Audience?

Answer: No. TikTok's requirement concerns matched users, with delivery eligibility also relevant.

2. Does hashing remove the need for an appropriate legal basis?

Answer: No. Hashing is not permission or complete anonymisation.

3. Does organic search availability imply Search Ads availability?

Answer: No. Verify the paid feature in the relevant account and market.

#11What should you check before the next lesson?

Confirm audience hypotheses, rights, source quality and actual delivery controls. Keep unsupported audience assumptions visible. Next, set budgets and bidding so the planned comparisons are affordable and the spending controls match the business risk.

Previous: TikTok Ads Objectives, Campaign Structure and Smart+ · Next: TikTok Ads Budgets, Bidding and Experiment Design

Or return to the TikTok Marketing for Marketers track.

Related: TikTok Ads Objectives, Campaign Structure and Smart+ · TikTok Ads Budgets, Bidding and Experiment Design · TikTok Pixel, Events API and Consent: Measurement Setup

#12Common questions

How large must a Custom Audience be? Current TikTok Help requires 1,000 matched users to use it in an ad group. Uploaded rows and available reach are different quantities.

Does hashing make customer data automatically usable? No. Appropriate rights, notices, purpose and privacy controls remain necessary.

Are targeting interests the same as verified job roles? No. Treat available interests and behaviours as proxies rather than proof of buying authority.