What Does It Actually Mean to Match Creators With Data?

Most seeding proposals lead with creator counts and follower totals. Brands worry about something else: whether these are the right people for the product. At PLAYLAB we maintain our own dataset of 1.5M+ global TikTok creators, fill in 90+ attributes per creator through vision analysis, and narrow candidates against those attributes. This is an operating record of what those attributes are, and what changed when they matched.

Follower count is not fit

Assembling creators with large followings is not the hard part. The hard part is choosing channels where your product has somewhere to sit.

Two channels at 100,000 followers produce entirely different outcomes for a skincare product if one routinely holds products up and explains ingredients while the other posts dance videos. Follower count carries none of that difference. That is why follower tier is the last filter we apply, not the first.

What we know about a single creator

Vision analysis reads a creator's recent content and fills in attributes from it. The basis is what is actually on screen — not profile text, not hashtag self-declarations. It runs along two axes.

Who they are — how much of the face appears on screen, gender, estimated age range, ethnicity, skin tone, hair type and length and colour, hijab wearing, typical makeup intensity, visible tattoos, visible skin concerns, expert-authority signals, and whether the persona is suspected AI or virtual.

These appearance attributes exist to match the conditions a product will be used in. For a foundation, the skin tone range decides the result; for a hair product, the hair type does. We use them to find candidates that fit the product conditions a brand specifies. We do not use them to exclude creators on the basis of any attribute.

What they make — skincare, makeup, haircare, nail art, beauty reviews, GRWM, before-and-after, product-only shots, cooking, dining out, mukbang, diet and nutrition, supplements, fashion, fitness, dance, family and kids, pets, travel, lifestyle, education, entertainment, and the density of on-screen captions.

The attribute set is not fixed. It currently stands at 90 or more, and when a campaign needs a distinction we cannot yet make, we add the attribute and re-analyse existing creators as well. Analysis covers the full dataset of 1.5M+ creators, and results refresh daily. A creator's subject matter drifts over time, so we do not analyse once and stop — we keep refilling.

Every matching attribute changed the outcome

Everyone says fit matters. Here is what our operating record actually shows.

We took creators assigned to beauty-category campaigns and grouped them by how many content attributes overlapped with the product category. Setting the completion rate of the zero-overlap group to 1:

Overlapping content attributes	Completion rate (0 = 1×)
0	1×
1	about 1.5×
2	about 2.5×
3	about 3.8×
4 or more	about 6.4×

It is not one point that spikes — it climbs with every attribute added. That monotonic shape is why we trust this table. A single combination performing well could be coincidence; a curve that rises each time overlap increases is hard to produce by accident.

These are multipliers for comparing groups, not absolute levels. Only creators with vision analysis results were counted. The 60–70% gifted seeding recovery rate we publish elsewhere is a separate metric on a different basis, so the two numbers should not be read together.

Face visibility moved in the same direction. Channels where the face rarely appears ran at roughly half the level of channels where it appears throughout. That is the difference between a channel that shows the product in use and one that does not.

A creator who has never participated is not the same as one who has finished

This is the second axis we weigh heavily. We took assignments where the same creator had been placed on a campaign at least once before, and grouped them by how many campaigns that creator had completed previously. Setting the no-prior-completion group to 1:

Prior completions	Subsequent completion rate (0 = 1×)
0	1×
1	about 19×
2	about 33×
3 or more	about 45×

A single completed campaign was the strongest individual signal for whether the next one completes. That is why we manage creators as ongoing relationships rather than using them once and moving on.

Read the other way, it also means this: repeatedly sending the same offer to a creator who has never once responded is extremely inefficient. So we separate creators by prior participation history and build campaigns that distinguish the seats needing new discovery from the seats where a relationship already exists.

What to ask for in any proposal

If an agency tells you they match with data, asking these on the same terms will surface the differences between them.

• What they know about one creator — ask for the attributes by name. If it is follower count, view count and category, that is a filter, not matching.
• How they obtained it — is it what the creator declared in a profile, or values read from actual content?
• When it refreshes — creators change what they post about. A shortlist built on last year's classification is not that person today.
• Whether you can see candidates before shipping, and whether the brand can strike names off.
• Whether prior campaign history is used at all, or whether every campaign blasts from scratch.

When we fit, and when we do not

We fit when you want to specify brand fit at the attribute level, and you want candidates meeting those conditions found overseas, shipped to, and verified through upload by a single team. Give us the vision attributes as conditions and we will narrow the pool and hand you the shortlist first. The brand can make the final creator selection.

We do not fit if you want the creator database itself as an in-house tool — we are an agency that runs campaigns, not a company that sells data. We also do not fit a campaign where every creator must follow a script confirmed to the second; gifted seeding is structurally the wrong shape for that, and it should be designed a different way.

The figures above are differences observed between groups in our operating record. They are not a guarantee of confirmed performance in a given country, product or deadline. Tell us your target markets, your product and the creator conditions you want, and we will scope the candidate volume and an operating plan against those conditions.

Evidence

1. Proprietary global TikTok creator dataset of 1.5M+ and its coverage (worldwide except China, 14 deepest markets) — https://playxlab.com/
2. Creator sourcing, outreach, shipping, upload verification and reporting owned by one team — https://playxlab.com/about/
3. Average gifted seeding recovery of 60–70% and the basis it is measured on — https://playxlab.com/lab/guides/how-tiktok-seeding-works/
4. Account-controlled confirmation and candidate review before shipping — https://playxlab.com/lab/guides/seeding-campaign-step-by-step/
5. Vision attribute count (90 or more), analysis scope (the full 1.5M+ creator dataset), completion rate by attribute overlap and by prior completions — PLAYLAB internal operating record, compiled 2026-09-17