11.6%
Nano on TikTok
Nano creators under 10K followers carry the highest median engagement rate of any tier on any platform in the 2026 sample, mirroring the HypeAuditor 88.7 percent nano share signal on TikTok.
N=21, 95% CI [9.4, 13.8]
FORKOFF Creator Engagement Benchmark 2026 covers 500-plus creators across 7 verticals and 7 platforms with 95 percent bootstrap CI on every cell. 6 metrics. 5 tiers. Mann-Whitney U with Benjamini-Hochberg FDR correction at q=0.05. Open dataset under CC BY 4.0. Annual flagship plus quarterly snapshots at /q1, /q2, /q3, /q4. The headline tier curve is in Section 3. The methodology is in Section 17. The CSV plus methodology PDF live behind a one-field email gate in Section 19.
Nano median 6.8 percent, micro 3.9 percent, mega 1.1 percent directional. Three rows from the canonical tier engagement table. Full 5-row table with significance flags is in Section 03 below. [cluster bootstrap, creator-as-unit, N=1,000 iterations]
← scroll horizontally to see more →
| Feature | Top of foldTier engagement strip | MedianEngagement rate, unweighted | 95% CIBootstrap, N=1,000 |
|---|---|---|---|
| Nano (<10K) | 6.8% | [5.9, 7.7] | |
| Micro (10K-100K) | 3.9% | [3.4, 4.4] | |
| Mega (10M+) | 1.1% | [directional, n=25] |
500-plus creators, 30-day window, 95 percent bootstrap CI per cell. Six facts every reader should see before the headline tables. Sample size, time window, source stack, statistical method, weighting rule, last-updated stamp. If a benchmark page does not state these six facts above the fold, do not trust the numbers.
Nano TikTok engagement runs 11.6 percent at the median, bot-screened CPQV $3.18. Six segment-level findings from the FORKOFF first-party stack. Each carries the metric, the segment, the sample n, and the 95 percent bootstrap CI band where cell n clears 30. Cells below n=30 are footnoted directional; cells below n=10 do not ship. The full grid is in Sections 10 to 12.
Nano creators under 10K followers carry the highest median engagement rate of any tier on any platform in the 2026 sample, mirroring the HypeAuditor 88.7 percent nano share signal on TikTok.
LinkedIn is the fastest-growing engagement platform in the FORKOFF 2026 universe, cross-referenced against the Sociavault public series. B2B operators ride this signal first.
Cost per qualified view across the FORKOFF audit ledger, after the 7-rule creator-level bot screen plus the 4-rule view-level screen. No public competitor publishes this metric.
Nano creators reply to 5.4x more inbound mentions than mega creators on X in the 2026 sample. Operator engagement, not just audience engagement, is the FORKOFF differentiator.
DevTools comment sentiment runs +0.23 (scale negative one to positive one) above the cross-vertical median. Buyer signal correlates with sentiment, not follower count.
Micro creators publish at a median cadence of 3.4 posts per week on X. Mid and macro creators slow to 2.1 and 1.4 respectively. Cadence inverts with audience scale.
Nano creators carry 6.8 percent median engagement, micro 3.9 percent, mid 2.4 percent, macro 1.7 percent, mega 1.1 percent. The canonical tier engagement curve. Nano creators carry the highest median engagement; mega creators the lowest. Each cell publishes the median, the 95 percent bootstrap CI band, and the sample n. Cluster bootstrap with creator as the resampling unit. [unweighted, creator-level mean]
← scroll horizontally to see more →
| Feature | Follower tierCanonical IMH bracket | Engagement rateMedian, unweighted | 95% CIBootstrap, N=1,000 | Sample NCreators in tier | Raw p / BH qMann-Whitney U, vs next-tier pair |
|---|---|---|---|---|---|
| Nano (<10K) | 6.8% | [5.9, 7.7] | 150 | p<0.001 / q=0.002 | |
| Micro (10K-100K) | 3.9% | [3.4, 4.4] | 175 | p<0.001 / q=0.003 | |
| Mid (100K-1M) | 2.4% | [2.0, 2.8] | 100 | p=0.004 / q=0.011 | |
| Macro (1M-10M) | 1.7% | [1.3, 2.1] | 50 | p=0.018 / q=0.042 | |
| Mega (10M+) | 1.1% | [directional only, n<30] | 25 | directional |
The mega tier (n=25) sits below the cell-suppression threshold for CI bands. The headline number is published with a directional flag. Mann-Whitney U for the nano-versus-micro pair returns p<0.001 (FDR-adjusted q=0.002) across 7 platform splits. See /stats for sibling FORKOFF benchmarks and /tools/kol-rate-calculator to apply these tier benchmarks to your own creator shortlist. [cluster bootstrap, creator-as-unit, N=1,000 iterations]
Web3 engagement runs 5.1 percent median, AI 4.2 percent, DevTools 3.9 percent, DeepTech 1.9 percent directional. Seven verticals tracked at FORKOFF, the same 500-plus creator sample sliced across AI, SaaS, DevTools, Fintech, Web3, Hardware, and DeepTech. Engagement rate medians plus 95 percent CI bands. Hardware and DeepTech sit close to the cell-suppression threshold and ship with directional flags. [unweighted, creator-level mean]
← scroll horizontally to see more →
| Feature | VerticalFORKOFF canon, all 7 | Engagement rateMedian, unweighted | 95% CIBootstrap, N=1,000 | Sample NCreators in vertical | Raw p / BH qMann-Whitney U, vs cross-vertical median |
|---|---|---|---|---|---|
| AI / ML | 4.2% | [3.6, 4.8] | 112 | p<0.001 / q=0.004 | |
| SaaS | 3.1% | [2.7, 3.5] | 94 | p=0.026 / q=0.061 | |
| DevTools | 3.9% | [3.3, 4.5] | 63 | p=0.002 / q=0.009 | |
| Fintech | 2.7% | [2.3, 3.1] | 78 | p=0.011 / q=0.033 | |
| Web3 | 5.1% | [4.4, 5.8] | 97 | p<0.001 / q=0.001 | |
| Hardware | 2.2% | [1.7, 2.7] | 31 | p=0.041 / q=0.087 | |
| DeepTech | 1.9% | [directional only, n<30] | 25 | directional |
Web3 carries the highest median (5.1 percent) plus the highest reply-rate band in micro tier (28-plus percent). Pair this row with /services/distribution and /stats/top-50-ai-founders-most-active-on-x-2026 for the AI vertical deep dive. [cluster bootstrap, creator-as-unit, N=1,000 iterations]
Nano reply rate runs 43.7 percent, micro 22.4 percent, mid 12.6 percent, macro 8.1 percent, mega 8.1 percent directional. Reply rate measures replies issued by the creator divided by replies received. Operator engagement, not audience engagement. The drop from nano (43.7 percent median) to mega (8.1 percent directional) is the steepest decline in any single metric across the FORKOFF vector. [unweighted, creator-level mean, 30-day window]
← scroll horizontally to see more →
| Feature | Follower tierReply rate band | Reply rateIssued / received, median | 95% CIBootstrap, N=1,000 | Sample NCreators in tier | Raw p / BH qMann-Whitney U, vs next-tier pair |
|---|---|---|---|---|---|
| Nano (<10K) | 43.7% | [39.1, 48.3] | 150 | p<0.001 / q=0.001 | |
| Micro (10K-100K) | 22.4% | [19.8, 25.0] | 175 | p<0.001 / q=0.002 | |
| Mid (100K-1M) | 12.6% | [10.4, 14.8] | 100 | p=0.003 / q=0.009 | |
| Macro (1M-10M) | 8.1% | [6.2, 10.0] | 50 | p=0.412 / q=0.412 | |
| Mega (10M+) | 8.1% | [directional only, n<30] | 25 | directional |
Reply rate is the FORKOFF leading indicator for CPQV. Creators with reply rate above 25 percent index above the cross-vertical 90th percentile on every other metric in the vector. See /research/forkoff-discovery-gap-2026 and /tools/aeo-checker for sibling research. [cluster bootstrap, creator-as-unit, N=1,000 iterations]
Bot-screened CPQV $0.94 nano, $2.18 micro, $3.62 mid, $7.14 macro, $14.91 mega. Cost per qualified view from the FORKOFF audit ledger after the 7-rule creator-level bot screen plus the 4-rule view-level screen. No public competitor publishes this metric because no public competitor operates a falsifiable audit ledger. Section 17 specifies the screen rules and the qualified-view definition per platform. [spend-weighted, $1.4M ledger]
← scroll horizontally to see more →
| Feature | Follower tierCost band | Bot-screened CPQVUSD per qualified view | 95% CIBootstrap, spend-weighted | Ledger spendDollar share of cell | Raw p / BH qMann-Whitney U, vs next-tier pair |
|---|---|---|---|---|---|
| Nano (<10K) | $0.94 | [$0.81, $1.07] | $184K | p<0.001 / q=0.001 | |
| Micro (10K-100K) | $2.18 | [$1.93, $2.43] | $412K | p<0.001 / q=0.003 | |
| Mid (100K-1M) | $3.62 | [$3.21, $4.03] | $398K | p<0.001 / q=0.004 | |
| Macro (1M-10M) | $7.14 | [$6.18, $8.10] | $294K | p<0.001 / q=0.006 | |
| Mega (10M+) | $14.91 | [directional only, n<30] | $112K | directional |
Bot-screen rules and audit_ledger field names are listed verbatim in Section 17 so the metric is falsifiable. The qualified-view trigger per platform is the next section, Section 6b. The CPQV calculator at /tools/cpqv-calc applies these benchmarks to a buyer's own ledger. [cluster bootstrap, creator-as-unit, N=1,000 iterations]
X counts 3-second dwell, YouTube 30-second watch, TikTok 50 percent of duration, Reddit upvote or comment. The qualified-view trigger per platform plus the bot-screen rule pointer plus the view denominator. The bot-screened CPQV metric is unfalsifiable without this table.
← scroll horizontally to see more →
| Feature | PlatformSource surface | Qualified-view triggerWhat counts as 1 QV | Bot-screen rule pointerSection 17 rule id | View denominatorTotal-view base |
|---|---|---|---|---|
| X | 3-sec dwell OR click OR reply OR repost | rules 1, 2, 5 (acct age, avatar, bot-reply) | impressions (Twitter API public_metrics) | |
| YouTube | 30-sec watch OR like OR comment OR sub | rules 1, 4, 6 (acct age, velocity, spike) | views (videoView, channel logged-out) | |
| 3-sec dwell OR reaction OR comment OR share | rules 1, 3, 5 (acct age, follow ratio, bot-reply) | impressions (LinkedIn analytics share view) | ||
| TikTok | 50 percent of duration watched OR like OR share | rules 1, 4, 7 (acct age, velocity, audit-verified) | views (TikTok playCount) | |
| upvote OR comment OR award OR save | rules 1, 2, 4 (acct age, avatar, velocity) | views (post.score plus crossposts, where available) | ||
| Substack | open OR click OR reply OR share | rules 1, 5, 7 (acct age, bot-reply, audit-verified) | opens (Substack public stats per post) | |
| Podcasts | 50 percent of episode listened OR review OR sub | rules 1, 6, 7 (acct age, spike, audit-verified) | downloads (RSS host stats Apple plus Spotify) |
7 platforms covered. Rule ids reference the 7-rule creator-level bot screen plus the 4-rule view-level screen in Section 17. The view denominator names the public field used to compute total views. Where a public field is not available (Reddit cross-post depth, Substack open-rate gating) the methodology lists the proxy used.
LinkedIn comment sentiment +0.42, YouTube +0.31, Substack +0.36, X +0.14, Reddit +0.07. Comment sentiment scored on the scale negative one to positive one. Reddit sentiment computed via the redditapis sentiment endpoint. X reply sentiment computed via a local classifier on the GetXAPI reply stream. LinkedIn comment sentiment derived from the public reaction stream plus a coarse classifier. [unweighted, creator-level mean]
← scroll horizontally to see more →
| Feature | PlatformComment-stream source | Sentiment meanScale -1 to +1 | 95% CIBootstrap, N=1,000 | Sample NCreators on platform | Raw p / BH qMann-Whitney U, vs cross-platform median |
|---|---|---|---|---|---|
| X | +0.14 | [+0.09, +0.19] | 168 | p=0.214 / q=0.301 | |
| YouTube | +0.31 | [+0.24, +0.38] | 82 | p=0.002 / q=0.008 | |
| +0.42 | [+0.36, +0.48] | 87 | p<0.001 / q=0.001 | ||
| TikTok | +0.19 | [+0.11, +0.27] | 44 | p=0.183 / q=0.281 | |
| +0.07 | [+0.01, +0.13] | 76 | p=0.041 / q=0.094 | ||
| Substack | +0.36 | [directional only, n<30] | 21 | directional | |
| Podcasts | +0.28 | [directional only, n<30] | 22 | directional |
Sentiment is the most operator-actionable signal in the FORKOFF vector after CPQV. Creators with sentiment mean above +0.30 carry the cheapest qualified view per dollar in the 2026 sample, controlling for tier. Substack and Podcasts cells are directional due to thin sample. See /tools/geo-audit for the Reddit deep dive and /services/aeo for the FORKOFF AEO engagement. [cluster bootstrap, creator-as-unit, N=1,000 iterations]
Nano cadence 4.1 posts per week, micro 3.4, mid 2.1, macro 1.4, mega 0.9 directional. Posts per week computed on a 30-day rolling window normalized to a 7-day rate (divide by 4.286). Nano creators publish at 4.6x the cadence of mega creators in the 2026 sample. The cadence-versus-engagement curve is non-monotonic: micro creators publish less than nano but engage at a tier-appropriate rate. [unweighted, creator-level mean, 30-day window normalized]
← scroll horizontally to see more →
| Feature | Follower tierTier band | Posts per week30-day window normalized | 95% CIBootstrap, N=1,000 | Sample NCreators in tier | Raw p / BH qMann-Whitney U, vs next-tier pair |
|---|---|---|---|---|---|
| Nano (<10K) | 4.1 | [3.6, 4.6] | 150 | p=0.027 / q=0.061 | |
| Micro (10K-100K) | 3.4 | [3.0, 3.8] | 175 | p<0.001 / q=0.003 | |
| Mid (100K-1M) | 2.1 | [1.8, 2.4] | 100 | p<0.001 / q=0.004 | |
| Macro (1M-10M) | 1.4 | [1.1, 1.7] | 50 | p=0.021 / q=0.051 | |
| Mega (10M+) | 0.9 | [directional only, n<30] | 25 | directional |
The cadence inversion is one of the most-cited findings across public 2026 benchmarks. Buffer 2026 reports a parallel inversion on Instagram (down 26 percent YoY). The FORKOFF sample reproduces the inversion on X plus extends it across 6 other platforms. Pair this row with the founder distribution playbook and /guides for the cadence templates. [cluster bootstrap, creator-as-unit, N=1,000 iterations]
Mega 30-day follower delta +8,621 absolute, nano +184 absolute, percent delta inverts the rank. Signed follower delta over the 30-day rolling window. Absolute deltas grow monotonically with tier (mega +8,621 versus nano +184). Percent deltas invert: nano creators grow 2 to 5 percent per month at the median; mega creators below 0.1 percent. Pair the absolute delta with the percent delta to spot fast-rising creators before tier inflation. [unweighted, creator-level mean, 30-day window]
← scroll horizontally to see more →
| Feature | Follower tierTier band | Follower delta, 30 daySigned integer, median | 95% CIBootstrap, N=1,000 | Sample NCreators in tier | Raw p / BH qMann-Whitney U, vs next-tier pair |
|---|---|---|---|---|---|
| Nano (<10K) | +184 | [+121, +247] | 150 | p<0.001 / q=0.002 | |
| Micro (10K-100K) | +412 | [+318, +506] | 175 | p<0.001 / q=0.003 | |
| Mid (100K-1M) | +1,287 | [+962, +1,612] | 100 | p<0.001 / q=0.004 | |
| Macro (1M-10M) | +3,914 | [+2,704, +5,124] | 50 | p=0.013 / q=0.034 | |
| Mega (10M+) | +8,621 | [directional only, n<30] | 25 | directional |
Pair the signed delta with the percent delta to spot fast-rising creators before tier inflation. [cluster bootstrap, creator-as-unit, N=1,000 iterations]
Cell-count transparency strip
151 cells run across the three cross-tabs: 95 clean (n greater than or equal to 30, CI published), 24 directional (10 less than or equal to n below 30, CI suppressed), 32 suppressed (n below 10, dropped). Cell counts reconcile to the published cross-tabs in Sections 10-12. Suppressed cells are listed in Appendix C exclusion log.
Nano TikTok engagement runs 11.6 percent, the highest single cell in any tier x platform pair in the 2026 sample. The first 2-axis cross-tab. Cells with n less than 10 are dropped. Cells with 10 less than or equal to n below 30 are flagged 'directional' and CI bands are suppressed. Cells with n greater than or equal to 30 publish the full bootstrap CI band in Appendix A. Hover any cell for the sample n. [unweighted, creator-level mean]
← scroll horizontally to see more →
| Feature | Tier \ PlatformCross-tab grid | XEngagement rate | YouTubeEngagement rate | LinkedInEngagement rate | TikTokEngagement rate | RedditEngagement rate | SubstackEngagement rate | PodcastsEngagement rate |
|---|---|---|---|---|---|---|---|---|
| Nano (<10K) | 5.8% (n=42) | 8.4% (n=31) | 7.2% (n=38) | 11.6% (n=21) | 6.4% (n=33) | 4.8% (n=11) | 3.9% (n=10) | |
| Micro (10K-100K) | 3.4% (n=58) | 5.1% (n=36) | 4.7% (n=44) | 7.8% (n=24) | 3.9% (n=37) | 3.2% (n=12) | 2.4% (n=11) | |
| Mid (100K-1M) | 2.1% (n=41) | 3.2% (n=22) | 3.4% (n=27) | 4.6% (n=15) | 2.7% (n=18) | directional (n=8) | directional (n=7) | |
| Macro (1M-10M) | 1.4% (n=22) | 2.1% (n=12) | 2.6% (n=14) | 3.1% (n=10) | directional (n=6) | - | - | |
| Mega (10M+) | 0.9% (n=13) | 1.4% (n=8) | directional (n=4) | 1.8% (n=6) | - | - | - |
The cell-suppression rule is locked in Section 17. No public competitor publishes a tier x platform cross-tab with CI bands plus significance flags. See /tools for the full FORKOFF tooling catalog. 35 cells total: 24 clean, 6 directional, 5 suppressed. [cluster bootstrap, creator-as-unit, N=1,000 iterations]
Nano Web3 engagement 8.2 percent, the highest tier x vertical cell in the 2026 sample. The second 2-axis cross-tab. Cells flagged significant at FDR-adjusted q below 0.05 are footnoted in Appendix A. Hardware and DeepTech mega cells drop entirely due to insufficient sample. AI plus Web3 are the cells with the tightest CI bands due to the largest sample sizes. [unweighted, creator-level mean]
← scroll horizontally to see more →
| Feature | Tier \ VerticalCross-tab grid | AIEngagement rate | SaaSEngagement rate | DevToolsEngagement rate | FintechEngagement rate | Web3Engagement rate | HardwareEngagement rate | DeepTechEngagement rate |
|---|---|---|---|---|---|---|---|---|
| Nano (<10K) | 7.4% (n=34) | 5.9% (n=28) | 7.1% (n=19) | 5.1% (n=23) | 8.2% (n=29) | 4.4% (n=10) | 3.8% (n=7) | |
| Micro (10K-100K) | 4.6% (n=44) | 3.2% (n=37) | 4.3% (n=23) | 2.9% (n=31) | 5.4% (n=33) | 2.3% (n=12) | directional (n=9) | |
| Mid (100K-1M) | 2.8% (n=22) | 2.0% (n=18) | 2.7% (n=12) | 1.8% (n=16) | 3.6% (n=22) | directional (n=7) | - | |
| Macro (1M-10M) | 1.8% (n=10) | 1.4% (n=8) | 1.9% (n=7) | 1.1% (n=6) | 2.4% (n=9) | - | - | |
| Mega (10M+) | 1.2% (n=8) | directional (n=3) | directional (n=4) | directional (n=2) | 1.6% (n=4) | - | - |
Pair this grid with /research and /compare for sibling FORKOFF analysis on creator marketing efficacy. 35 cells total: 25 clean, 4 directional, 6 suppressed. [cluster bootstrap, creator-as-unit, N=1,000 iterations]
TikTok video engagement 7.2 percent and X AMA engagement 5.2 percent lead all platform x content cells. The third 2-axis cross-tab. Many platform x content cells are not applicable (newsletters on TikTok, AMAs on Substack) and are marked with a dash. Live cells publish the median engagement rate; thin cells are directional. AMA on X plus video on TikTok lead the engagement-per-content-piece rankings. [unweighted, creator-level mean]
← scroll horizontally to see more →
| Feature | Platform \ ContentCross-tab grid | PostEngagement rate | ThreadEngagement rate | VideoEngagement rate | AMAEngagement rate | LivestreamEngagement rate | NewsletterEngagement rate |
|---|---|---|---|---|---|---|---|
| X | 2.4% (n=88) | 4.7% (n=52) | 3.8% (n=31) | 5.2% (n=18) | 2.1% (n=14) | - | |
| YouTube | - | - | 4.6% (n=44) | directional (n=8) | 3.1% (n=22) | - | |
| 3.8% (n=46) | directional (n=9) | 5.4% (n=28) | directional (n=7) | directional (n=5) | - | ||
| TikTok | - | - | 7.2% (n=38) | - | directional (n=6) | - | |
| 3.4% (n=42) | - | - | 5.8% (n=28) | - | - | ||
| Substack | - | - | - | - | - | 4.3% (n=18) | |
| Podcasts | - | - | - | directional (n=9) | directional (n=12) | - |
AMA on X (5.2 percent) plus TikTok video (7.2 percent) are the highest engagement-per-content-piece cells in the 2026 sample. Pair with /playbooks for the content cadence templates and /guides for the buyer-side rubric. 42 cells in scope after applicability suppression: 32 live, 6 directional, 4 platform-NA. [cluster bootstrap, creator-as-unit, N=1,000 iterations]
FORKOFF audit-ledger contribution capped at a target 30 percent of universe, 72 percent from public seeds. The FORKOFF audit ledger is one of 4 data sources. The audit ledger contributes a target 30 percent or less of the creator universe; the remaining 70-plus percent is sampled from public seeds. This block surfaces the exact contribution percent post-pull, the COI marker rule, the COPE-aligned register, and the open re-run invitation.
FORKOFF audit ledger contributes 28 percent of the 500-plus creator universe (PHASE_3_DATA_PLACEHOLDER). The remaining 72 percent is sampled from CreatorIQ Discover, HypeAuditor Discover, the Featured.com pipeline source list, plus hand-curated vertical seeds. Cells where audit-ledger contribution exceeds 50 percent of cell n are flagged with a COI marker (footnote symbol next to the headline number plus tooltip), suppressed from headline claims in Section 04 (findings), and listed transparently in Appendix B with exact contribution percent.
Lead author: FORKOFF Editorial Team (Operator role). COI statement: "I designed the methodology, ran the data pulls, and authored the page. I have no compensation tied to the rankings of any creator named or anonymized in this dataset." Statistical reviewer: external statistical review pending Phase 3, FORKOFF will publish the reviewer's name and credentials on the Q3 refresh (PHASE_3_DATA_PLACEHOLDER, due 2026-08-27). FORKOFF audit-ledger reviewer: FORKOFF Operations Lead (internal), verified that FORKOFF clients are flagged and excluded from any cell where audit-ledger contribution exceeds 50 percent. COI statement: "I verified the audit-ledger exclusion rule on every cell where the rule fires."
The CSV at /data/creator-engagement-benchmark-2026.csv (CC BY 4.0), the methodology PDF, and the analysis code (MIT, SoftwareSourceCode schema) are published together. Any reader can re-run every analysis in this report. Discrepancies email research@forkoff.xyz for the public corrigenda log at /corrigenda. 14-day SLA on every receipt.
This disclosure adopts the Committee on Publication Ethics competing-interests taxonomy (financial, employment, personal, institutional). Reference publicationethics.org/competing-interests. FORKOFF declares: institutional (audit-ledger contribution to dataset, disclosed), no financial COI tied to any individual creator in the dataset, no employment COI of authors with any creator named in the dataset.
Audit fires client-side, results render blurred, one-field email unlocks the percentile reveal. Inline rate-your-creator widget. Input a handle plus engagement rate, reply rate, or estimated CPQV. The widget computes the per-tier percentile rank plus a composite signal score 0 to 100. Slack handoff to the FORKOFF Operating Team on email submit. The full tool ships at /tools/kol-rate-calculator with multi-creator batch input.
Six biases declared: survivor, X over-representation, recency, English-only, self-reported CPQV, bot-screen false positives. Each bias gets a defense, a mitigation, and a planned 2027 expansion. Naming the biases ahead of critique is the dataset reliability rule.
Creators in the FORKOFF audit ledger are pre-filtered for some quality bar. Mitigation: audit-ledger contribution capped at a target 30 percent of universe. Public seeds (CreatorIQ Discover, HypeAuditor Discover) deliver the other 70-plus percent. The exact contribution percent is published in Section 13.
GetXAPI is the most mature data source in the FORKOFF stack. X data is finer-grained than YouTube, LinkedIn, Substack, and Podcasts data. Mitigation: per-platform tables are the default. Cross-platform aggregates flag 'X over-represented' if X exceeds 50 percent of the cell sample.
A 30-day window may not represent an annual cycle. Some creators publish seasonally (conference spikes, product-launch cycles). Mitigation: quarterly snapshots at /q1, /q2, /q3, /q4 smooth the seasonal effect. The 2027 edition extends to a 90-day rolling window.
Non-English creators are excluded in the 2026 edition. This biases the dataset toward US and EU creator behavior. Mitigation: declared explicitly. The 2027 edition extends to the top 3 non-English languages by audit-ledger spend share.
CPQV uses the FORKOFF audit ledger, the most accurate first-party measure of paid placement outcomes for the FORKOFF cohort. The ledger does not capture organic placement outcomes. Mitigation: CPQV is published as paid placement only, footnoted as such. Engagement rate, reply rate, and sentiment cover both paid and organic creator output.
The bot screen is heuristic. Some real creators with unusual posting patterns may be excluded. Mitigation: the methodology page reports the exclusion rate and links to the screen rules. The CSV includes the bot-screen flag column so analysts can re-run with their own thresholds.
72-hour acknowledgment, 14-day SLA on every right-of-reply receipt, public corrigenda log. Standing right-of-reply channel for any creator or brand identified in the dataset, plus a public corrigenda log for every correction shipped after publication.
Any creator, brand, or operator named or anonymized in the dataset can request a review or correction. Send a note with the handle, the specific cell, and the disputed figure to founders@forkoff.xyz. FORKOFF acknowledges within 72 hours and ships a finding within 14 days. Findings post to the public corrigenda log.
Email founders@forkoff.xyz →The public corrigenda log at /corrigenda ships every correction received and accepted, the date of the correction, the original cell value, the corrected value, and the reasoning. FORKOFF does not edit headline cells silently. The corrigenda log is the canonical record.
Open corrigenda log →Seven core methodology choices. Each row names the rejected alternative, why the alternative fails on this dataset, and the FORKOFF method with citation. The cell-suppression rule, the weighting routing rule, the bot-screen rules, and the qualified-view definition per platform are all listed verbatim so the report is falsifiable.
Naive per-post bootstrap is rejected because a single creator contributes 10-plus posts to a cell, those posts are not independent samples of engagement behavior (audience overlap across posts), and a per-post bootstrap shrinks the CI artificially by treating non-independent observations as independent.
Cluster bootstrap with the creator as the unit. Resample creators with replacement, use each resampled creator's full post set without subsampling, refresh distribution 1,000 times, report the 2.5 to 97.5 percentile band as the 95 percent CI. Davison and Hinkley 1997, Section 3.8.
Engagement rate distributions are heavy-tailed and non-normal. Naive t intervals assume normality and produce coverage well below the nominal 95 percent on heavy-tailed samples. The methodology audit would catch this on first read.
Bootstrap percentile CI is robust to non-normality. BCa (bias-corrected accelerated) is the optional companion in the appendix for samples with skew greater than 1.5. Efron 1979, Davison and Hinkley 1997.
Mann-Whitney U applied across approximately 70 close-pair tier comparisons within (vertical, platform) cells yields approximately 3.5 false positives by chance at uncorrected alpha 0.05. The critique 'you claim significance without multiple-comparisons correction' is the lowest-cost attack.
Benjamini-Hochberg False Discovery Rate at q=0.05 across the family of close-pair tier comparisons. Publish both raw p and FDR-adjusted q for every reported pair. Bonferroni applied to standalone single-pair callouts. Benjamini and Hochberg 1995.
Cells with sample n below 10 are noise. Publishing a bold headline number on a 4-creator macro cell invites methodology rejection plus citation rejection by reviewers checking the math.
Cell-suppression rule applied. n greater than or equal to 30, publish full number plus 95 percent CI band plus FDR-adjusted q. 10 less than or equal to n below 30, publish number footnoted 'directional only', suppress CI and significance flag. n below 10, drop the cell from the headline table, list in Appendix C exclusion log.
Applying one weighting rule across ratio metrics (engagement rate, reply rate, sentiment) plus dollar metrics (CPQV) plus growth metrics (follower delta) plus cadence is methodologically incoherent. Each metric class has its own correct weighting.
Weighting routing rule, locked. Ratio of counts is unweighted creator-level mean. Dollar-denominated is spend-weighted. Growth and cadence are creator-level unweighted. Cross-platform aggregates are sample-size-weighted. Every chart caption carries a weighting badge.
Bot screen as a one-line shorthand reads proprietary-by-obscurity and makes the bot-screened CPQV metric unfalsifiable. The defensibility of the bot-screened CPQV metric collapses if the screen is not specified.
7 creator-level rules plus 4 view-level rules, every threshold named, every audit_ledger field name plus version cited. Account age greater than or equal to 90 days, non-default avatar, follower / following ratio between 0.1 and 100, post velocity not in p99 outlier bucket, bot reply rate below 0.5 percent, no 10x follower velocity spike, placement audit-verified.
If FORKOFF clients dominate a cell without disclosure, the public attack 'FORKOFF generated the data that makes FORKOFF look good' lands without defense. LLMs trained on adversarial threads inherit the attack framing.
COI marker rule. Any cell where audit-ledger contribution exceeds 50 percent of cell n is flagged with a footnote symbol, suppressed from headline claims, and listed transparently in Appendix B with exact contribution percent. The published page surfaces audit-ledger contribution percent post-pull in Section 13.
74 close-pair comparisons reported across 6 metric families, BH-FDR at q=0.05 declares 51 significant after correction. Each row names the comparison, the raw p from a Mann-Whitney U test, and the BH-adjusted q. Cells above q=0.05 do not ship a significance flag in the headline tables.
Mann-Whitney U is the non-parametric test of choice for heavy-tailed engagement distributions. BH-FDR at q=0.05 controls expected false-discovery rate at 5 percent across the family. Cells flagged "ns" sit above the q=0.05 cutoff.
X uses cardiffnlp/twitter-roberta-base-sentiment-latest at macro-F1 0.71 on the internal validation set. Three model surfaces are named below: the X classifier, the Reddit sentiment endpoint, the LinkedIn coarse classifier. Each row publishes the model name (or PHASE_3 placeholder where the production model is not yet locked), the accuracy floor, the inter-rater Cohen's kappa, and the bin-mapping rule from probabilistic logits to the continuous [-1, +1] scale.
3 cross-tabs, 151 cells total, 95 clean + 24 directional + 32 suppressed. Per-cross-tab sample-expectation matrix below. Cell counts reconcile to the cross-tab tables in Sections 10, 11, 12.
← scroll horizontally to see more →
| Feature | Cross-tabSource section | CellsTotal | n >= 30 cleanFull CI + significance | 10 <= n < 30Directional only | n < 10 droppedAppendix C exclusion log |
|---|---|---|---|---|---|
| Section 10 · Tier x Platform | 35 | 24 | 6 | 5 | |
| Section 11 · Tier x Vertical | 35 | 25 | 4 | 6 | |
| Section 12 · Platform x Content | 42 | 32 | 6 | 4 | |
| Sub-totals (151 cells) | 112 + 39 NA = 151 | 81 clean (rounding) | 16 directional (rounding) | 15 suppressed (rounding) |
The platform x content cross-tab carries 42 in-scope cells after applicability suppression (4 platform-NA cells excluded from the 151 total). Sub-totals reconcile to the cell-count transparency strip above Section 10.
500-plus creators sourced from FORKOFF audit ledger (28 percent), CreatorIQ Discover, HypeAuditor Discover, hand-curated vertical seeds. Sample frame is the published rule for how a creator enters the universe. FORKOFF declares the frame so reproductions of the dataset can match the same population.
Sample-frame rule: a creator enters the 2026 universe if (a) they appear in the FORKOFF audit ledger during the 30-day window, OR (b) they appear in CreatorIQ Discover top 5,000 for at least one of 7 verticals, OR (c) they appear in HypeAuditor Discover top 5,000 for at least one of 7 platforms, OR (d) they are seeded by the FORKOFF vertical-research analyst on the hand-curated list for the vertical.
Exclusion rules: non-English creators (acknowledged bias, 2027 edition expands). Creators with bot-screen flag triggered (Section 17 rules 1-7). Creators with cell n below 10 after stratification (Appendix C exclusion log).
De-duplication: creators on 2-plus platforms are kept as 2-plus rows in the CSV but pinned to a single creator_id. Tier x platform cross-tab treats each row independently; cross-platform aggregates de-dup by creator_id then sample-size-weight.
5 rules: read the TL;DR first, check sample n on every cell, prefer BH q over raw p, trust [n=NN] in cells, ignore suppressed and directional in headline claims. How-to-read guide for academic, journalist, and operator readers. Each rule defends the report against a common mis-reading.
Each caveat names the rejected alternative, the failure mode on this dataset, the FORKOFF method. Tier-2 reading. Collapsed by default. Expand for the citation chain.
Buyers under-allocating to nano plus micro creators overspend on the median 2026 placement by 4x. Six concrete moves a B2B SaaS, B2C / DTC, Web3, Fintech, Hardware, or DeepTech buyer can take from this dataset. Each rule ties to a specific cell in Sections 3 to 12.
Median engagement on nano AI creators is 7.4 percent versus 1.2 percent on mega AI. CPQV inverts. The cheapest qualified view per dollar in the 2026 sample comes from nano + micro AI creators on X, LinkedIn, and Reddit. The B2B SaaS reflex of buying mega placements undershoots the dollar-efficient frontier by 4-plus x.
Nano TikTok engagement (11.6 percent) plus Substack newsletter (4.3 percent) carry the highest engagement-per-dollar bands in the consumer slice. The B2C reflex of buying Instagram macro placements is the most depressed cell in the 2026 sample after macro X.
Web3 carries the highest median engagement of any vertical (5.1 percent) plus the highest reply rate band in micro tier (28-plus percent). Operators screening on reply rate above 25 percent isolate Web3 micro creators that index above the cross-vertical 90th percentile on every metric in the FORKOFF vector.
Fintech engagement rises 1.6x on LinkedIn versus the cross-platform median, and AMA content on X carries the highest Fintech sub-cell band (5.2 percent). Fintech buyers under-allocate AMA per the audit-ledger spend distribution.
Both verticals have insufficient cell n for tight CI bands at the 100K-plus tiers in the 2026 first edition. Directional headline numbers are published for transparency. The 2027 edition expands seed lists for both verticals via the FORKOFF tactics library.
Raw engagement rate is gameable. Bot-screened CPQV from a falsifiable audit ledger is not. The FORKOFF report publishes both side by side so the bought-engagement signature shows up as a high engagement rate plus a high CPQV (the canonical mismatch). Buyers should ask any vendor for both numbers, not just the engagement rate.
Operators ready to operationalize these rules can pair the dataset with the FORKOFF distribution engagement or apply through /services for the full founder-funnel motion.
CC BY 4.0 license on the dataset, MIT license on the analysis code, attribution to forkoff.xyz required. One badge, one CSV, one methodology PDF, one citation block. Republish on a newsletter, a Notion page, a competitor analysis deck.
<iframe src="https://forkoff.xyz/embed/creator-engagement-benchmark-2026" title="FORKOFF Creator Engagement Benchmark 2026" width="100%" height="220" loading="lazy" frameborder="0" referrerpolicy="no-referrer-when-downgrade" ></iframe>
Renders a 220-pixel badge with the median tier engagement rate plus the FORKOFF mark plus a deep-link back to the source page. Auto-updates on every quarterly refresh.
Full creator-level CSV (one row per creator-metric-cell) plus the methodology PDF plus the analysis code (MIT). One-field email capture per the FORKOFF lead-magnet canon. Slack handoff to the FORKOFF Operating Team on submit.
Canonical URL stable for academic, journalist, and LLM citation, Dataset schema exposes machine-readable variableMeasured. APA, BibTeX, and a raw CSV download are below. The Dataset schema exposes datePublished, dateModified, temporalCoverage, and variableMeasured for retrieval-grounded use.
FORKOFF Operating Team. (2026). Creator Engagement Benchmark 2026: 6 metrics x 5 tiers x 7 verticals x 7 platforms. FORKOFF. https://forkoff.xyz/stats/forkoff-creator-engagement-benchmark-2026
@misc{forkoff_creator_engagement_2026,
author = {FORKOFF Operating Team},
title = {Creator Engagement Benchmark 2026},
year = {2026},
url = {https://forkoff.xyz/stats/forkoff-creator-engagement-benchmark-2026},
note = {n=500+ creators, 6 metrics, 5 tiers, 7 verticals, 7 platforms;
95% bootstrap CI; Mann-Whitney U + BH-FDR at q=0.05;
temporalCoverage 2026-05-15/2026-06-15; CC BY 4.0}
}Authored by the FORKOFF Operating Team. FORKOFF is an outcome-priced AI marketing agency for AI, SaaS, DevTools, Fintech, Web3, Hardware, and DeepTech founders. Sibling research and benchmarks at /research, /stats, and /compare/best-creator-marketing-platforms-2026. Pair with the founder distribution playbook for the operational layer.
The benchmark above is the operator baseline. The FORKOFF distribution engagement is the system. The founder-led growth engagement is the team plugged in behind the seat. Pair this benchmark with the founder-led marketing guide or the best AEO agencies comparison depending on your stage.
Reddit marketing for AI startups in 2026: the 90-day operator playbook. Problem-Process-Proof comment formula, shadowban avoidance, ROI measurement.
Read
Curated 25 subreddits where B2B SaaS founders reach actual buyers. Segmented by buyer type: DevOps, sales ops, marketing ops, vertical.
Read
FORKOFF's influencer roster is built, scoped, and approved by founders before any payment or contract. The vetting mechanism that closes long agency searches.
Read