Skip to content

gooseworks-ai/goose-skills / skills/marketing/work-with-creators

work-with-creators

Verified

Real creators and influencers, researched, priced and briefed. Use when the marketer says "find influencers", "which creators are worth paying, at what price?", "find real UGC creators to film for our ads", "who should we seed product to", "is this creator's engagement real", "what should I pay this creator" or "write a creator brief". Free research first; one credit yes only to vet deeper. Drafts only: contacts nobody, agrees nothing, moves no money. Not for AI-made UGC-style video — use goose-video; if unsure, ask: a real person filming, or AI-made?

references/vetting.md

View on GitHub
# Vetting: real engagement, audience fit, brand safety

Load at step 4. Every number here is an **approximate practitioner range (US, 2025–26)**, drawn from published benchmark reports and agency practice. Ranges drift every year — treat them as sanity checks, not facts, and date-stamp every verdict. There is no fake-follower API: all of this is *signals, not verification*. Say so in the roster.

## 0. Which depth, which rubric

| Role | Depth | What decides the verdict |
|---|---|---|
| Reach, affiliate, ambassador (paid) | **Full** | Sections 1–11 |
| Gift-only nano/micro in a seeding wave | **Light**: profile, last 12 posts, one comment pull, a caption scan for safety and competitor tags | Sections 2–4, 7, 8 (stated concern + comment self-descriptions, no audience pull), 9 (captions only), 10 (captions only), 11 — enough to avoid a bad fit at low cost; anyone promoted to paid gets full vetting |
| UGC-for-ads (content for the brand's own ads) | **Full, different rubric** | Section 13 — audience size and fit barely matter |
| LinkedIn voice, newsletter, podcast | **Full, B2B branch** | Section 14 |

## 1. The sample

- The last **12–20 posts** on the platform that matters.
- Leave out: pinned posts, giveaways ("comment to win"), and anything younger than ~48 hours (TikTok, Reels, X) or ~7 days (YouTube). They are missing data, not evidence.
- Score **sponsored** and **organic** posts separately.
- Use the **median**, never the mean. If mean > ~2× median, one viral post is carrying the profile — price and judge on the median and say so.

## 2. Formulas — never compare across them

- **By views** (TikTok, Reels, YouTube, X): (likes + comments + shares + saves) ÷ views, using whatever of those the data exposes. Say which fields were used, and compare only like with like — an ER without shares and saves reads lower than the bands below.
- **By followers** (IG feed images/carousels, LinkedIn): engagements ÷ followers, because views are not public there.
- **View rate**: median views ÷ followers.

## 3. Healthy ranges (approximate)

| Platform | Engagement | View rate | Red line |
|---|---|---|---|
| TikTok | ~4–10% by views | nano/micro ~10–50%; mid+ ~5–20% | ER under ~3%, or uniform low views on a big following |
| IG Reels | ~3–8% by views | micro ~20–60%, falling with tier | Reels plays far below followers with high likes |
| IG feed | by followers: nano ~3–6%, micro ~1.5–4%, mid ~1–2.5%, macro ~0.8–1.5%, mega ~0.5–1.2% | — | Likes high, comments near zero |
| YouTube | (likes + comments) ÷ views ~3–6% | median views of videos 30–90 days old ~10–30% of subscribers (don't mix in videos a few days or years old) | under ~3% of subscribers = stale subscriber base |
| X | likes ÷ views ~0.5–2% | — | replies with no substance |
| LinkedIn | comments ~5–10% of reactions is normal; judge pods by **who** comments, not the ratio | — | the same 10–30 people commenting "Great insights!" within the first hour on every post (an engagement pod) |
| Newsletters | not public — subscriber band from badge or media kit, likes/comments per post, a public "advertise" page | claimed opens ~40–60% plausible; above ~70% needs proof | no proof of size at all |
| Podcasts | chart rank, review count, views of the YouTube version | downloads cannot be verified — mark "creator-claimed, verify" | — |

Tiers by followers: nano <10k · micro 10–100k · mid 100–500k · macro 500k–1M · mega 1M+. B2B voices are judged on who engages, not size.

## 4. Read the comments (50–100 across ~5 posts)

Red flags: more than ~30% emoji-only or generic ("Nice 🔥", "Love this"); letter+digit handles with no profile picture; comments that do not refer to the content; comment language that does not match the creator's or the target market's; the same commenters on every post. Comment-to-like ratio ~1–5% is typical on Instagram; under ~0.5% there with high likes suggests bought likes (TikTok ratios run lower — don't apply this flag there); over ~10% suggests a giveaway or pod.

Green flags: real questions about the product category, people describing themselves ("my two senior dogs", "as a nurse"), purchase intent ("where do I get this?") above ~2–5% of comments, creators replying to comments.

## 5. Signs of bought followers or engagement (inferred)

- Followers far above what views support (200k followers, a uniform ~2k views per post).
- Engagement suspiciously uniform post to post — real accounts have a long tail.
- Likes high, shares and saves near zero (where visible).
- Old posts that out-engage new ones by a wide margin while followers kept growing.
- Where TikTok audience countries exist: concentration in countries that do not match the creator's language or content.

## 6. Sponsored vs organic

- Sponsored median views ÷ organic median views should be ≥ ~0.7. Under ~0.5 = the audience tunes out ads; price on the sponsored median.
- More than ~1 in 3 recent posts sponsored = ad fatigue flag.
- The creator appears in a brand's Meta partnership ads = evidence a brand paid to amplify them (they convert well enough to scale) — and a possible competitor conflict.
- YouTube: check suspected sponsors per recent video (experimental signal, treat as a lead, not proof).

## 7. Cadence and consistency

Posted within ~14 days. Roughly ≥2 a week on TikTok, ≥1 a week on IG, ≥2 a month on YouTube, ≥1 a week on LinkedIn. Gaps over 30 days are a flag. A creator found through one viral outlier (research-market's saved outliers in the Inspiration library lean this way) must show a consistent median, not one hit.

## 8. Audience fit (state as a confidence level with evidence)

- **Comments are the audience sample**: self-descriptions, languages, places, purchase questions.
- **The creator's own content**: audiences roughly mirror the creator's niche and situation as the creator describes it; topic share ≥ ~50% of the last 30 posts on the niche.
- **TikTok audience countries** where pulled — countries only, and a pricier lookup: reach and install roles only, at most the top five finalists, shown as its own line at Gate 2.
- **B2B**: commenter headlines (function, seniority, company type) are the best evidence there is. A "LinkedIn influencer" whose engagement comes from other creators is not reaching buyers.
- Limits: commenters skew younger and more engaged than viewers; viral posts reach non-followers. **Never infer ethnicity, age, gender, skin tone or health from photos or names** — not the creator's, not their audience's. Use only what people say about themselves, quoted and counted ("14 of 80 commenters describe acne-prone skin").
- Confidence: **high** (direct evidence: audience countries + on-persona comments), **medium** (consistent content + comment signals), **low** (inference from content only — say so).

Category-specific fit: beauty needs the skin or hair concern the creator **states** in their own content; pet needs evidence the creator really has the animal; food needs regional availability and allergen awareness; apps need the creator's audience on the right store/region.

## 9. Brand safety (≥ ~12 months, ~50–100 posts including captions)

- Search the name with "scam", "apology", "drama", "controversy" (web search through the proxy); read call-outs in comment sections; notice deleted-looking gaps. If no search was possible, the verdict says "safety: captions and comments only" — never "clean".
- Politics, hate, explicit content, promotion of crypto schemes, gambling, MLMs, unlicensed medical or financial advice.
- Claims the brand could never make, made by the creator in past sponsored posts for the category (a supplement creator who promised cures).
- Transcripts of a few recent videos when captions are thin.

## 10. Competitor conflicts

Competitor partnership ads live in the Meta ad library; competitors' recent posts tagging the creator (the brand's own competitor organic feed is free); the creator's own posts tagging competitors in the last ~3–6 months. A recent competitor deal is either a conflict (possible exclusivity — the human asks) or evidence of category fit; say which and why. **Working for a rival or adjacent tool** (employee, founder, evangelist, ambassador — check the bio, LinkedIn headline and pinned post) is not a deal to ask about: it's a near-miss, with that reason.

## 11. Regulated categories and minors

- **Supplements, skincare, wellness**: signed-off structure/function or cosmetic claims only; no disease words ("cures", "treats", "heals", "pain relief"). **Acne**: "clears / treats / prevents acne or breakouts" are drug claims in the US (OTC acne products) — only if signed off in exactly those words. No unsubstantiated before/after.
- **Financial**: no promised returns, no "guaranteed"; crypto is high-risk; finance creators are regulated in several markets.
- **Alcohol**: creator 21+ (or local legal age) and an audience overwhelmingly of legal drinking age under industry codes; if it cannot be evidenced, the creator is out.
- **Minors**: a creator under 18 is out of scope for this playbook; kids regularly in content is a flag for regulated categories; "made for kids" YouTube channels are out.
- **Medical / pharmaceutical**: out unless the human confirms a compliance process exists.

## 12. The verdict line

> "@handle · TikTok · micro (62k) · pulled 2026-10-16 · last 18 posts (2 sponsored) · median views 21k (34% view rate) · ER 7.1% by views (likes+comments+shares) · sponsored median 17k (0.81× organic) · comments: 14 of 80 describe their own acne-prone skin, 9 ask where to buy, ~12% generic · posts 3×/week · audience fit medium (self-descriptions + on-niche content; no audience data pulled) · safety: captions, comments and a name search, 12 months, nothing found · conflicts: a competitor partnership ad, July 2026 (exclusivity unknown — ask)." (Fictional.)

A verdict without these numbers is not a verdict.

## 13. UGC-for-ads rubric

The content runs on the brand's own ad account (or through partnership/Spark access), so the creator's audience barely matters. Judge:
- **On-camera delivery** — natural, specific, believable; holds attention in the first 2–3 seconds. The MCP cannot show you a creator's video, so the verdict says **"delivery not watched"**; judge what you can from transcripts (first lines, pacing of the words), ad-library reuse and their portfolio, and put 2–3 video links per UGC finalist in the roster for the human to watch before approving names.
- **Hook craft** — the first line of their best videos; variety of openings across posts.
- **Product handling and demo** — shows the thing clearly; good light and sound on a phone.
- **Past brand-ad work** — their content in brands' partnership ads (ad library) or a portfolio on their link-in-bio; did brands run it as ads (a sign it performed).
- **Turnaround and reliability** — posting cadence; stated turnaround on their portfolio.
- **Honesty risk** — creators who present every product as a years-long favourite. For a new product or app, the creator must actually use it before filming.
Skip audience fit and audience countries for this role.

## 14. B2B branch (LinkedIn voices, newsletters, podcasts)

- **LinkedIn**: the post sample from the person's recent posts (post search is keyword-based and may not return a clean last-12–20; record the real sample size and lower confidence when it is thin); reactions and comments per post; **who** comments — commenter headlines (function, seniority, company type) when the data exposes them; if it doesn't, say "commenter mix not checked" and lower the confidence. Pods are judged by repeat commenters, not ratios.
- **Newsletters**: size is creator-claimed unless a public badge or media kit shows it — mark "size creator-claimed — request the media kit"; sponsor history from past issues; topic fit.
- **Podcasts**: chart presence, review count, views of the YouTube version, past sponsors (spoken reads in transcripts).
- Fit is buyers, not reach: a 6k-follower LinkedIn voice whose commenters are finance leaders beats a 60k voice whose commenters are other creators.

## 15. What public pages show (the free pass)

Read in the host browser as a person would — a profile, a dozen posts, a few comment threads per creator; no scripted paging. What's visible logged-out changes often; check, and list what was hidden.

| Platform | Usually visible logged-out | Often hidden → paid top-up |
|---|---|---|
| TikTok | Followers, total likes, **views on every video in the profile grid**, likes / comments / shares / saves on a video page, many comments | Full comment threads, transcripts, audience countries |
| YouTube | Subscribers (rounded), views and upload date on every video, likes, comments | Transcripts at volume, suspected sponsors at volume |
| Instagram | Followers, post count, bio and link-in-bio, captions, sometimes reel plays and likes | Reel plays and likes on many accounts, most comments, tagged posts |
| LinkedIn | Headline, some recent posts with reaction and comment counts | Commenter headlines at volume, full post history |
| Newsletters, podcasts | Advertise / sponsor pages, Substack badges, chart positions, YouTube versions | Subscriber and download numbers (creator-claimed; ask for the media kit) |

The free verdict uses what was visible, labelled "public-page read <date>"; a creator whose key number is hidden is still shortlisted with that gap named, and the top-up fills it.

**No browser, only a fetch tool** (it often fails on TikTok and gets followers-only from Instagram): use, in this order, public creator-stats pages found by search (label "third-party stats, data date <date>" — they lag and round), the numbers in search-result snippets for a specific post (label them), YouTube pages (they fetch well), and the creator's own media kit. Never call a top-5 or a single post the median; say "typical views not read". Say in the report that the free read was fetch-only.

**Check every handle before listing it**: open the profile and match the name, niche and location to the source that named it — roundups and search snippets often give a wrong or stale handle. A mismatch is dropped, not shortlisted.

add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks.Stars: 115kUpdated Claude CodeCodexVerifiedInstall