How we score.
Every score PlutoBa shows you comes from the method on this page - the real signals, the real weights, the real thresholds. We publish it because you are making budget decisions on these numbers, and a score you cannot interrogate is just a vibe with a typeface.
Standard and Deep.
Standard
Programmatic signals only: engagement against benchmark, ratio checks, consistency, platform-specific patterns. Analyses up to 100 recent posts. Completes in under a minute.
Partnership Score = 100 minus the risk score.
Deep
Everything in Standard, plus an AI analyst reads the actual content and up to 300 comments: comment authenticity, brand safety, audience authenticity, geography and a written verdict. Completes in 2 to 4 minutes.
Pitch vets arriving through your intake link are Deep-shaped: same data fetch, same model, a lightened analysis pass.
What a vet actually checks.
Seven signals, each mapped here to precisely what runs underneath - including the two places where the honest answer is "that is a composite" rather than a tidy metric.
A composite, not a single number. We score the follow-back pattern (following-to-follower ratio against tier thresholds), engagement that is implausibly high for the audience size (bought engagement has a ceiling problem in the other direction), posting-consistency anomalies, and, on Deep vets, an AI read of whether the audience behaves like real people.
There is no single "fake follower percentage" - any tool quoting one is estimating. We score the pattern instead of inventing a decimal.
The creator's engagement rate measured against the published benchmark band for their follower tier and platform (tables below). Too far below the band flags Low; implausibly above it flags Suspicious - the ceiling scales with tier, because a nano account at 25% is normal and a mega account at 25% is not.
Deep vets only. The AI reads up to 300 real comments and scores 0-100 how much of the conversation is genuine humans versus bot loops, emoji walls and pods. This is the heaviest signal in the Deep risk model at 30 points.
Deep vets only. The AI reviews content themes, language and controversy signals, scoring 0-100. Deliberately kept OUT of the risk score so it cannot be double-counted - it feeds the Partnership Score directly at 40%.
Audience geography, inferred from content and audience signals, which drives the regional rate benchmark. Plus the audience-authenticity read above.
We do not claim age or gender splits. Public data does not support them honestly, so we do not sell them.
The asking rate compared against a benchmark band built from follower tier, platform and niche multipliers. You see below, within or above benchmark on every vetted pitch.
Niche detection - what the creator actually makes content about, not what their bio says. It sharpens the rate benchmark (niches price differently) and tells you at a glance whether the pitch belongs in your queue.
Risk model weights, by platform.
Each signal contributes its weight in points to a 0-100 risk score - a clean profile scores near zero, a suspicious one accumulates. Deep vets shift weight onto the AI's comment-authenticity read because it sees what ratios cannot. On Instagram, when likes are hidden the comment-to-like weight redistributes to the ratio and consistency checks rather than being silently dropped.
TikTok
| Signal | Std | Deep |
|---|---|---|
| Engagement vs benchmark | 35 | 30 |
| Comment-to-like ratio | 25 | 20 |
| Follow-back ratio | 20 | 10 |
| Posting consistency | 20 | 10 |
| AI comment authenticity | - | 30 |
| Signal | Std | Deep |
|---|---|---|
| Engagement vs benchmark | 30 | 30 |
| Comment-to-like ratio | 20 | 20 |
| Follow-back ratio | 15 | 10 |
| Posting consistency | 15 | 10 |
| Instagram-specific checks | 20 | - |
| AI comment authenticity | - | 30 |
YouTube
| Signal | Std | Deep |
|---|---|---|
| Engagement vs benchmark | 30 | 25 |
| Subscriber engagement | 25 | 20 |
| Posting consistency | 20 | 10 |
| Shorts-heavy composition | 15 | 10 |
| Family-safe signals | 10 | 5 |
| AI comment authenticity | - | 30 |
Brand safety appears in no risk table by design: it feeds the Partnership Score directly at 40% so a safety concern can never be diluted by an otherwise clean profile.
Engagement benchmarks, by tier.
The healthy band per follower tier. Below the band scores Low; far enough below scores harder. Above the band is High; implausibly above - the ceiling scales from 2x the band on nano accounts to 4x on mega accounts - is flagged Suspicious, because bought engagement overshoots.
| Tier | Followers | TikTok | YouTube | |
|---|---|---|---|---|
| Nano | under 10K | 8-20% | 3-12% | 4-12% |
| Micro | 10K-100K | 5-15% | 2-8% | 3-8% |
| Mid | 100K-500K | 3-10% | 1.5-5% | 2.5-6% |
| Macro | 500K-1M | 2-8% | 1-3.5% | 2-5% |
| Mega | 1M+ | 1.5-8% | 0.5-2.5% | 1.5-4% |
YouTube bands measure like-to-view ratio; a separate subscriber-to-view check runs alongside. Each flagged signal converts to a penalty on a fixed ladder from 0 (healthy) through 30, 50, 60 and 75 up to 100 (suspicious) before its weight is applied.
Published as ranges, deliberately: the exact follow-ratio cutoffs, comment-to-like boundaries and sponsored-saturation trip points sit inside the ranges described above. Publishing them to the decimal would hand a tuning manual to exactly the accounts we score against, so the precise values are the one thing this page withholds.
The Partnership Score.
On a Standard vet the Partnership Score is simply 100 minus the risk score.
On a Deep vet it is a weighted judgement:
where authenticity is 100 minus risk, and the verdict value maps the AI's recommendation: proceed 100, caution 65, avoid 25. The verdict then caps the score: an avoid can never exceed 34 and a caution can never exceed 64, so a chart-friendly number can never outrank the analyst's judgement.
Labels read from the score: 80 or above Excellent, 65 Good, 50 Fair, 35 Poor, below that Risky. The verdict word always accompanies its colour.
One pitch, end to end.
@stellarose_lifestyle - the specimen creator from our homepage, not a real person - pitches through your intake link: 248K Instagram followers, asking $1,800 per post.
248K followers lands in the Mid tier (100K-500K), so her Instagram benchmark band is 1.5-5%.
Authentic engagement rate measures 4.1% - comfortably inside the band, at its stronger end. No flag, no penalty.
Follow-back ratio is healthy and comments sit in normal proportion to likes. Posting rhythm shows a mild wobble - a small consistency penalty.
The comment analysis finds mostly genuine conversation with some repetitive pockets, and the risk model lands at 25 - authenticity 75. Content review scores brand safety 80: clear. Verdict: proceed.
Partnership Score = (75 x 0.4) + (80 x 0.4) + (100 x 0.2) = 82. PROCEED - and her $1,800 ask sits within the Mid-tier Instagram benchmark, so the rate line reads "within benchmark".
What this method cannot do.
- It works from public data. A vet is an informed judgement on observable signals, not an audit of a platform's private analytics.
- It scores probability, not certainty. A clean score is strong evidence, not a guarantee - and a flagged one deserves a human look before you write anyone off.
- It does not produce a fake-follower percentage, because one cannot be measured honestly from the outside.
- Audience demographics stop at geography and authenticity. We do not claim age or gender splits.
- It covers TikTok, Instagram and YouTube. A vet is a snapshot at the moment you run it, not ongoing monitoring.
See it score your first pitch.
Free plan, not a trial · No card required