Choose creators using evidence about the people who watch them, the problems those people discuss, and the markets you can serve. Record where each claim came from before scoring it. A creator's follower count can help you describe account size, but it cannot answer whether those followers are your customers. Start with audience fit, leave missing evidence marked unknown, and compare account size after a candidate clears your requirements.
Define the audience before you inspect the account
Write a short audience requirement that another person could apply without guessing. "People interested in food" leaves too much room for interpretation. A hypothetical brief for a meal-planning product might read:
We want to reach English-speaking adults in the United Kingdom who regularly cook weekday meals at home. The content should address planning, shopping, or preparing those meals. We need evidence of a UK audience before approving paid work.
That brief separates four questions: geography, language, topic, and use situation. It also names a requirement that cannot be traded away by a high total score.
Choose those requirements before looking at attractive profiles. Otherwise, a candidate with strong production skills can tempt you to redefine the customer around their content. If the brief is still broad, use the customer-problem search worksheet to turn buying situations into terms you can investigate.
Label the evidence before you score it
Use the following ladder as a recommended review order. These levels describe how much a piece of evidence can establish about your question. They are not universal accuracy ratings.
| Evidence class | What to record | What it can support | What remains unresolved |
|---|---|---|---|
| Platform-reported audience information shared by the creator | Account, field name, audience population, date range, capture date, filters | What the platform reports about that defined audience | Future campaign delivery, purchasing intent, and any platform measurement limits |
| Provider-modeled estimate | Provider, measurement date, audience definition, published method and coverage | A directional audience-fit hypothesis | Model error and differences from platform-reported data |
| Creator claim | Exact scope of the claim, when it was made, supporting material requested | What the creator says about their audience | Whether the claim is measured, current, or representative |
| Public content and visible interactions | Post links, dates, sampled posts, relevant questions or discussion | Observable content themes and examples of audience interest | The composition of the whole audience |
An observed analytics report is evidence of what the report says. Do not rename it a verified census of customers. YouTube describes unique viewers as an estimate, so even platform-reported data needs its original definition. A provider estimate should remain labelled as an estimate even when it contains a precise percentage. A creator's statement may be useful context while you request supporting evidence.
Modash describes collecting public information and applying machine-learning models to produce creator and audience insights. That supports labelling its modeled outputs as estimates. The overview does not provide a field-level error range, so do not invent one.
Keep these labels in the selection sheet. Removing them turns unequal evidence into numbers that look comparable.
Keep creator location separate from audience location
A public profile that says London gives you a location claim about the creator. It supplies no denominator for the audience. A London restaurant video also gives you no percentage of viewers who live nearby.
YouTube lets a creator choose a country or region in channel settings. Separately, its audience analytics reports the geographies contributing watch time. These fields answer different questions.
Use those clues to decide whom to investigate. Request audience geography before treating local reach as established. For a city campaign, country-level evidence still leaves the city question open. The local-creator checklist helps distinguish a creator's local connection from a visit, while audience reach needs its own evidence.
Ask for comparable information
Before requesting a screenshot or report, specify the population you need. Followers, viewers, reached accounts, and people who interacted answer different questions. Do not silently substitute one for another.
TikTok Studio documents separate viewer and follower demographics. YouTube's top geographies report describes watch time by geography, not a count of local followers. A UK watch-time share therefore needs that label. YouTube also warns that some audience data may be limited.
For shortlisted candidates, use this request:
Please share the audience information available for the account and content format proposed for this campaign. Include the account name, the field labels, the selected date range, and any filters. We are assessing UK audience geography and interest in weekday home cooking. Please flag information that is unavailable and explain whether any figures come from an external estimate.
Request only the aggregate information needed for the decision. You do not need a list of individual audience members.
When a report arrives, check its scope before copying a number. Does it describe the right account? Does its date range match the period you requested? Is the audience group named? Keep the original field label beside your interpretation.
If two reports disagree, first check their dates, populations, filters, and methods. Do not average a modeled follower estimate with a platform-reported viewer figure. They may describe different groups.
Modash's vetting guide recommends reviewing audience questions about products alongside demographics. Treat those comments as examples of interest. They do not establish income, purchasing power, or the proportion of the audience likely to buy.
Use a weighted rubric that preserves unknowns
The following rubric is an editorial recommendation for the hypothetical meal-planning brief. The weights and scoring thresholds are planning choices, not validated predictors of sales.
| Criterion | Weight | Evidence question |
|---|---|---|
| Serviceable geography | 35 | What UK share does the specified audience report show? |
| Relevant content and discussion | 30 | Do the reviewed posts and interactions address weekday home cooking? |
| Language | 15 | Does the reviewed content use the language required by the brief? |
| Product-use context | 20 | Can the proposed content show a credible planning, shopping, or cooking situation? |
Rate each supported criterion from 0 to 4. Use 0 for a documented mismatch, 1 for weak fit, 2 for mixed fit, 3 for strong fit, and 4 for direct fit across the reviewed evidence. Define what those descriptions mean for your brief before scoring candidates. Use U when evidence is missing or cannot answer the criterion. Language and use context score the content reviewed; neither establishes the language or buying intent of every viewer.
For a known criterion, calculate points as weight multiplied by rating divided by four. Keep unknown criteria out of the known-point total, then report their weight separately. Never turn U into a zero rating or silently drop its weight and rescale the result to 100.
A worked comparison
Every creator and figure below is hypothetical. Assume comparable evidence supports the numeric ratings. For this example, score UK audience share as 4 at 60% or above, 3 from 40% to below 60%, 2 from 20% to below 40%, 1 above zero but below 20%, and 0 at zero. Use the same audience definition for all candidates and unrounded values at each boundary. Creator C has no usable audience-geography evidence.
| Candidate | Followers | Geography | Topic | Language | Use context | Known points | Unknown weight |
|---|---|---|---|---|---|---|---|
| A | 18,000 | 4 | 3 | 4 | 3 | 87.5 | 0 |
| B | 140,000 | 1 | 4 | 4 | 2 | 63.75 | 0 |
| C | 52,000 | U | 4 | 4 | 4 | 65 | 35 |
Creator A has the strongest supported fit under this rubric. Creator B's larger follower count does not resolve the weaker geography fit. Creator C has 65 supported points, with another 35 unresolved. Their possible score spans 65 to 100; it is not a measured 100 or a confirmed 65.
Because the brief requires UK audience evidence, put C on hold pending that evidence. Do not reject C for a mismatch you have not established. Do not approve C because all the available ratings are high.
The rubric also does not prove that A will generate more sales than B. Fit is one selection decision. Cost, delivery terms, creative quality, and the measurement plan still need review.
Make the decision traceable
For each candidate, save the criterion, rating, evidence class, source reference, date, and unresolved question. Record one of three decisions: advance, hold for evidence, or reject for a documented mismatch. Add a short reason that another reviewer can check.
Move candidates who advance into a defensible creator shortlist, preserving the evidence labels alongside the scores. For your next three candidates, fill the geography row first. If geography is required and unknown, request that evidence before spending time ranking follower counts.
