Platforms in this story
In short
- Targeting is set at the ad group level: choices within one dimension add up (OR), choices across dimensions stack (AND).
- A custom audience needs 1,000↗ matched users to serve in an ad group, and the same base size to build a lookalike.
- TikTok recommends broad targeting: beyond 80%↗ of a country's users.
- In Smart+, automatic targeting is the default, and Symphony works on creative, not on audience.
Here is how I read TikTok's targeting documentation: what the platform states in black and white, what it presents as the result of its own tests, and what I would check in an account before building a recommendation for a B2B advertiser or an agency.
1,000
Custom audience: minimum matched users
TikTok for Business · checked October 7, 2026↗
80%
Threshold for "fairly broad" targeting
TikTok for Business · checked October 7, 2026↗
50languages
Symphony dubbing
TikTok for Business · checked October 7, 2026↗
Which targeting dimensions does TikTok offer?
The help page "Ad Targeting Dimensions" lists five families: demographics, audiences, advanced targeting, device type and smart targeting. Targeting is configured at the ad group level, and the available locations vary by market↗.
The combination logic is worth knowing before you build an ad group. Within one dimension, choices work as OR: if you select the Beauty interest or the #makeup hashtag, TikTok includes people who meet either criterion. Across dimensions, it is AND: "Female" and "25 to 44" only keeps people who meet both conditions↗. Stacking dimensions therefore shrinks the audience with each addition, while stacking choices inside one dimension widens it.
What is the difference between interests and behaviors?
TikTok separates the two on a dedicated page. Interest targeting is "predicted through machine learning based on long-term past behavior" and covers organic and paid content. Behavior targeting relies on recent actions, over a 7 or 15 day window, on organic content (for example completing, liking, commenting on or sharing a video in a given category)↗.
For a B2B advertiser the consequence is practical: a recent behavior signals a current interest, a predicted interest describes a lasting affinity. In the pages I read, I found no targeting dimension by job function or by industry: check the real list of options in your Ads Manager before promising that level of precision to a client.
How do custom and lookalike audiences work?
Custom audiences can be built from a customer file, engagement, app activity, website traffic, a lead generation form, a business account or shop activity, among others. They are used to include or exclude segments, for example to keep recent customers out of an acquisition campaign↗. The size condition is explicit: you need 1,000↗ matched users in the audience to use it in an ad group.
For a customer file, TikTok accepts email, phone number and mobile advertiser ID, as .csv or .txt, with hashed or original values↗. A lookalike audience starts from a source audience of at least 1,000 people and comes in three levels: narrow (closest to the source, but smaller), balanced and broad (larger reach, moderate similarity)↗. A B2B agency with only a few hundred contacts in its CRM will not reach the threshold: scope that before selling a retargeting plan.
Should you target broad or narrow?
TikTok is clear on this: ads reaching a "fairly broad" audience, meaning more than 80%↗ of a country's potential users, "tend to outperform all others by a wide margin". The page cites a CPA 15%↗ lower and a conversion rate 20%↗ higher than narrower audiences, and warns that targeting that is too narrow can keep an ad group from exiting the learning phase↗.
These are the platform's own figures: the page gives no sample, period or industry. I treat them as a direction to test, not a guarantee. For a B2B offer in a narrow market, a reasonable starting point is still broad targeting with strong creatives, then exclusions (customers, employees) rather than stacked interests.
What does Smart Targeting do?
Smart Targeting has two options, Smart Audience and Smart Interest and Behavior. It allows TikTok to deliver outside the interests, behaviors or audiences you selected when the model expects that to improve performance, without touching age, gender or location. TikTok states that it reduces CPA and ad fatigue "for most advertisers", with no figure↗. If you are testing a specific interest to learn something, keep in mind that this setting can dilute the test: compare two ad groups, with and without it.
Does Smart+ replace manual targeting?
In the upgraded Smart+ experience, automatic targeting is the default: TikTok identifies the best-performing audience with little input. You can set "audience controls" (location, minimum age between 18 and 25, languages, exclusions) or switch to custom targeting, which reopens age from 13 to 55+, gender, interests, behaviors, spending power, household income and device, at the risk of limiting delivery↗. The launch post of May 13, 2026 adds that you can guide the system with custom audiences or interests↗.
Smart+ campaigns cover the Traffic, Sales, Lead Generation and App Promotion objectives, and the help page warns that the feature "may not be available to you yet" on every account: ask your TikTok contact or support↗. The same May 2026 post cites a 29%↗ CPA improvement against manual campaigns, but for Smart+ Catalog Ads in closed beta, from TikTok's internal data: it does not apply to a B2B lead generation account.
Where does Symphony fit in targeting?
Symphony targets nobody: it is TikTok's creative suite. In Smart+, Symphony Automation offers two tools: Recommended Creatives, which generates videos from the destination URL (Smart+ web campaigns only at the date of the post), and Automatic Enhancements, which improves existing assets, refreshes music and translates and dubs videos in over 50 languages↗. At TikTok World 2026, the platform also announced improved Symphony image and video generation, powered by Dreamina Seedance 2.0 (May 13, 2026↗, according to the press).
My reasoning, to be read as such: when targeting is largely automated, creative becomes the main signal you control. The more clearly your videos speak to a specific job or problem, the more material the system has to find the right audience. To compare these video formats with Meta's, the guide Meta Ads in fall 2026 covers the attribution and creative diversity changes to know before reading results side by side.