Lifecycle Part 2: Vetting TikTok Creators (Signal vs Noise)

Part 1 ended with a 200-creator pool sitting in a KOL List. This chapter is about how to throw most of that pool away — quickly, confidently, and without burning your campaign budget on the candidates that look strong from a distance and fall apart on contact.

Vetting is where 80% of the campaign loss is prevented. It is also the stage that most teams skip because the dashboards make it feel optional — a 200k-follower account with a polished feed looks signed-deal-ready, and the marketing pressure to ship outreach this week is real. Skipping vetting is how you end up seven days later wondering why three of your ten picks never responded, two responded once and ghosted, and the one that actually posted got 8,000 views on an account with 480,000 followers.

The inflated 800k pattern

Every experienced influencer marketer has been burned at least once by what we call the inflated-800k pattern: an account with 800,000 followers, a beautiful feed, a celebrity-adjacent aesthetic, and a 2% engagement rate that quietly tells you 95% of those followers either do not exist or do not care.

The shape is consistent. The follower growth curve has one or two suspicious vertical cliffs (purchased batches), comments on recent videos are either generic emoji strings or in a language the creator does not appear to speak, the like-to-view ratio is healthy enough to pass a quick eyeball but the comment-to-like ratio is 1:300 or worse, and the audience geography snapshot shows a top country the creator has never visited. None of this requires forensic work. It requires looking at the right four numbers.

The four signals that do the work

Of all the data on a creator, four signals catch the majority of bad picks. The other twenty signals are either confirming these four or describing edge cases that mostly do not matter.

1. Engagement quality, not engagement rate

Raw engagement rate (likes + comments + shares / views) is the first number every tool reports and the easiest to fake. The more useful signal is engagement composition: how do the likes, comments, and shares relate to each other?

Healthy accounts cluster around a 1:20 to 1:40 comment-to-like ratio on recent videos. Coherent comments — questions about the product, replies to the creator, ongoing in-jokes — are nearly impossible to fake at scale, while like farming is cheap. A creator with 8% engagement rate but a 1:500 comment-to-like ratio is almost always a purchased-engagement account; a creator with 4% engagement rate and a 1:25 comment-to-like ratio is a genuine community.

KOLens surfaces both numbers on every creator card and on the public creator profile at kolens.ai/k/<username>. The dossier also includes per-video engagement so you can see whether the average is being skewed by one viral hit or held up across the last twenty videos.

2. Audience geography, not creator location

Where the creator lives is almost irrelevant. Where the creator's followers live is the entire game. A US-based creator can have an 85% Indonesian audience because that market discovered them a year ago and the algorithm compounded it. A Brazilian creator can have an 80% US audience because they post in English and a US Shop product went viral.

For paid placements, require an audience-country majority of 60% or higher in your target market — comfortable enough that the typical viewer of the sponsored video can actually buy the product. KOLens computes audience geography from real comment patterns and view-source signals; the snapshot lives on the dossier and we wrote a full piece on it in the audience-snapshot post.

3. Posting cadence and dormancy

A creator who posted four times a week six months ago and twice in the last 60 days is a silent-quit risk. The algorithm has already deranked them, the audience has already started forgetting them, and your sponsored post will get a fraction of what their averages suggest. None of this shows up in follower count, which is why posting cadence is the most under-weighted signal in vetting.

The KOLens dossier flags creators as dormant when the last post is more than 30 days old, and the cadence chart on the profile shows the rolling 12-week posting volume. For the full vetting recipe here see the posting-cadence metrics post.

4. The 8-signal authenticity score

The authenticity score collapses the above signals plus four more (follower growth shape, follower-account age, comment language coherence, engagement variance) into a single 0-100 number that lives on every creator card. It is not a magic verdict — it is a triage tool that lets you sort a 200-row list by trustworthiness in one click.

• 70-100: healthy. Worth contacting without a manual deep-dive on the trust signals. Spend your vetting time on fit and pricing instead.
• 50-69: borderline. Open the dossier and eyeball the comments, the growth curve, and the audience country. Often legit but worth ten extra seconds.
• 0-49: skip. Almost always a purchased- follower problem. Saving these wastes outreach slots.

For the full 8-signal breakdown and the formula behind each component see the authenticity-score deep-dive.

The vetting loop — 30 seconds per creator

Once the four signals are surfaced, vetting collapses to a ten-minute pass over a 200-row list. The loop:

1. 1

   Sort the list by authenticity score descending.

   The bottom of the list — anything under 50 — is gone in one select-and-delete. That is typically 15-25% of the pool.
2. 2

   Apply two table filters: engagement rate ≥ 4% and last post inside 14 days.

   This usually halves the survivors. The dormant creators and the polished-but-quiet accounts both fall away here.
3. 3

   Open the audience-country column.

   For paid placements, drop anything where the target country is under 50%. For brand-awareness plays or content licensing, keep this filter looser.
4. 4

   Click into the 10-15 dossiers worth a deeper look.

   The public dossier at kolens.ai/k/<username> shows the brand-fit assessment, recent sponsored posts (so you can check competitor overlap), and the audience persona. Two minutes per dossier; 30 minutes total for the deep cut.

Where KOLens enters the workflow

• /tiktok-audit — the audit surface. Paste a handle, get the full authenticity score, engagement composition, audience country, and dormancy flag in one screen. Useful for the 10-15 dossier-level checks per launch.
• kolens.ai/k/<username> — the public dossier page. Same data as the audit tool but shareable with anyone on the team (no login required). Often the link you drop into Slack when asking a colleague to double-check a candidate.
• /tiktok-creators — the cross-creator browse surface. Useful for comparing two candidates side by side when you can only contract one and the metrics are close.

A worked example — the 134 becomes 38

Picking up the phonecase brand from Part 1: discovery handed them 134 candidates inside a KOL List. The vetting pass took 14 minutes.

Sorting by authenticity score and dropping everything under 50 removed 22 creators — most of them the polished mid-tier accounts that had bought a follower batch six months earlier. The engagement-rate filter at 5% removed another 41 (mostly creators who looked fine but had a slow recent quarter). The 14-day cadence filter cut another 19 (silent-quit risks). The audience-country filter at 60% US removed another 14 (creators with US handles and Southeast Asian audiences).

The 134 became 38. Dossier review on the survivors flagged four for competitor overlap (recent sponsored posts for direct rivals) and two for off-brand voice. The final list to hand to outreach: 32 creators, every one of which the team could ship product to with high confidence.

The previous Excel-based workflow stopped at "looks fine to me" for most of the candidates and shipped product to about 18. Eight of those eighteen never posted. The new workflow ships to 12 and 11 of them post. The 200-creator funnel converts to eleven actual posts where the old one converted to three.

Common anti-patterns in vetting

• Trusting follower count. Follower count is the worst signal and the easiest to forge. Sort by engagement, never by followers.
• Vetting one creator at a time in a new tab. Vetting is a sort-and-filter operation on the whole list, not a slow tour of individual dossiers. Save dossier reads for the final ten.
• Skipping audience geography. This is the single most expensive miss. A creator with a perfect score on every other signal but an off-market audience is still a bad pick.
• Ignoring competitor overlap. Vetting must include a 30-second scan of recent sponsored posts. A creator who posted for a competitor last week is not available to you this week.

vs the alternatives

In the next chapter

Vetting collapses your pool to a short list of creators worth contacting. Part 3 — Library is about what you do with that list after the launch ends. Ad-hoc Excel does not scale beyond three launches; you need a durable creator-asset layer that survives team turnover, lets you re-find a perfect candidate from six months ago, and grows automatically between launches via Discovery Plans. That is the third stage of the lifecycle.

Continue to Part 3 — Building Your KOL Library →

The full lifecycle series

1. Part 1 — Discovery: Finding TikTok candidates at scale
2. Part 2 — Vetting (you are here): Separating signal from noise inside the funnel
3. Part 3 — Library: Turning finds into a durable asset
4. Part 4 — Monitoring: Tracking the picks you made
5. Part 5 — AI / MCP: Claude integration for natural-language research