How to Vet Influencers: A Scoring Framework for Creator Selection at Scale

The most important vetting dimension is the one most teams skip: actually looking at the creator's content. Not just their bio, not just their follower count — their posts, their videos, their stories, their captions.

• Visual quality. Is the content well-produced, or consistently low-effort? This is relative to your brand's aesthetic standards, not an absolute measure.
• Topical relevance. Does the creator regularly produce content in your category, or would your product be a random departure from their usual feed?
• Voice and tone alignment. Would this creator's natural voice feel authentic promoting your brand, or would sponsored content feel forced?
• Content consistency. Is the creator posting regularly with a coherent aesthetic and message, or is their feed erratic?

Content quality is inherently subjective, which is why it requires structured scoring (not a binary yes/no). A creator might produce beautiful content that is topically irrelevant, or produce amateur-quality content that is perfectly aligned with your brand's raw, authentic aesthetic. The scoring framework needs to capture these nuances.

A creator with 500,000 followers and a 0.5% engagement rate is less useful than a creator with 50,000 followers and a 5% engagement rate — and both are less useful than a creator with 50,000 followers, a 5% engagement rate, and an audience that actually matches your target demographic.

• Follower geography. If you sell in the US, a creator whose audience is 80% based in Southeast Asia will not drive conversions, regardless of engagement rate.
• Audience age and gender distribution. Does the creator's audience match the demographic you are trying to reach?
• Follower growth patterns. Sudden spikes — gaining 50,000 followers in a week — are often indicators of purchased followers or viral content that attracted a transient audience.
• Bot and fake follower detection. Tools exist to estimate the percentage of a creator's followers that are inactive or fraudulent accounts. Any rate above 15–20% should trigger scrutiny.

Engagement rate is the most commonly cited vetting metric, and it is useful — but only if you dig beneath the headline number.

• Comment quality. Are comments substantive responses to the content, or generic reactions? A post with 500 comments that are all “fire emoji” is worth less than a post with 50 thoughtful comments.
• Saves and shares. These are high-intent engagement signals. A save means someone wanted to reference the content later. A share means someone found it worth distributing. Both are stronger indicators of content value than likes.
• Engagement consistency. Does the creator maintain steady engagement across posts, or do they have occasional viral hits surrounded by low-performing content?
• Engagement rate benchmarks by tier. Nano-influencers (1K–10K) typically see 5–15% engagement rates. Micro (10K–100K) see 3–8%. Mid-tier (100K–500K) see 2–5%. Macro (500K+) see 1–3%. A macro-influencer with a 4% rate is exceptional; a nano-influencer with a 4% rate is below average.

Brand safety in influencer marketing is not just “has this person posted something controversial.” It is a structured assessment of content risk across defined categories.

The Global Alliance for Responsible Media (GARM) framework, widely adopted by the advertising industry, defines content risk categories including adult and explicit content, arms and ammunition, crime and harmful acts, death or military conflict, hate speech and discrimination, obscenity and profanity, drugs and controlled substances, terrorism, and debated sensitive social issues.

Each creator's content history should be evaluated against these categories with a risk level: Low, Medium, or High. Low-risk creators proceed with standard review. Medium-risk creators should be flagged for manual content review. High-risk creators should be excluded unless there is a specific strategic reason to proceed — and that reason should be documented.

How much of a creator's content is sponsored? This ratio matters more than most teams realize.

• Balanced (under 25% sponsored): The creator's feed feels organic. Sponsored content is interspersed naturally. This is the sweet spot for most brand partnerships.
• High (25–50% sponsored): The creator collaborates frequently. Their audience may be experiencing “ad fatigue,” reducing the impact of each new partnership.
• Very High (over 50% sponsored): The creator's feed is primarily promotional content. Their audience expects sponsored posts, which can reduce perceived authenticity and engagement quality.

Beyond the ratio, examine which brands the creator has worked with. Have they promoted direct competitors? Have they endorsed products that conflict with your brand values? A creator's collaboration history stored in your CRM provides the relationship context that informs these decisions.

Not all creators serve the same function in a campaign. Understanding creator archetypes helps you select creators who match your campaign objectives:

• Trendsetters: Early adopters who introduce their audience to new products and ideas. Best for launch campaigns.
• Educators: Creators who explain, review, and compare. Best for product-education campaigns where you need creators who can communicate complex value propositions.
• Lifestyle integrators: Creators who weave products into their daily content naturally. Best for product seeding and long-term brand awareness.
• Community builders: Creators whose audiences are highly engaged and interactive. Best for campaigns where comments, shares, and UGC generation matter.
• Entertainment-first: Creators who prioritize entertaining content. Best for brand awareness where the goal is reach and memorability rather than direct conversion.

The most effective approach is to define 2–5 evaluation questions that capture your specific vetting priorities. Each question maps to a dimension you care about, and each creator is scored on each question independently.

Example three-question framework for a D2C beauty brand:

• Q1: Content-brand fit. “How well does this creator's content aesthetic, topics, and audience align with our brand?” (Score: 1–10)
• Q2: Audience quality. “How authentic and demographically relevant is this creator's audience for our target market?” (Score: 1–10)
• Q3: Performance potential. “Based on engagement patterns, collaboration history, and content quality, how likely is this creator to deliver strong campaign results?” (Score: 1–10)

Each question gets a weight reflecting its importance: Content-brand fit at 40%, Audience quality at 35%, Performance potential at 25%. The composite score is the weighted average: (Q1 × 0.4) + (Q2 × 0.35) + (Q3 × 0.25).

This gives you a single number to rank creators, but also three component scores to understand why a creator ranked where they did. A creator might score 9 on content-brand fit but 4 on audience quality — valuable information that a single binary pass/fail would obscure.

Some criteria are non-negotiable. Before spending resources on detailed scoring, gate questions filter out creators who fail hard requirements:

• Keyword exclusion. Creators whose content contains specific terms (competitor brand names, banned topics) are automatically excluded.
• Minimum threshold on Q1. If a creator scores below a defined minimum on your most important evaluation question, they are excluded regardless of other scores.
• Safety gate. Creators flagged as High-risk in GARM content safety categories are excluded.

Gate questions reduce the evaluation pool before detailed scoring begins, saving computational and human review resources.

After a scoring run, the qualification rate — the percentage of evaluated creators who made the shortlist — is a diagnostic metric:

• High (>50%): Your search criteria are well-calibrated to your vetting standards. The discovery phase is surfacing relevant candidates.
• Moderate (20–50%): Normal for most campaigns. Discovery is casting a wide enough net, and vetting is applying meaningful differentiation.
• Low (<20%): Either your search criteria are too broad or your vetting standards are very high. Consider tightening discovery filters.
• Very low (<5%): Something is misaligned. Either the creator pool lacks qualified candidates, or vetting criteria are unrealistically strict. Revisit campaign requirements.

Tracking qualification rates across campaigns builds organizational knowledge about what conversion rates to expect from discovery to vetting, informing how many candidates to surface in the initial search.

The GARM Brand Safety Floor + Suitability Framework provides the industry-standard taxonomy for content risk classification. Rather than a vague “is this creator safe?” evaluation, GARM defines specific content categories with clear boundaries.

Each creator's content history — not just their most recent posts, but their account history — is evaluated against these categories. An AI-powered system can analyze hundreds of posts, captions, and visual content against GARM categories in seconds, surfacing specific flagged content with the category and risk level for human review.

• Low risk. No content flagged across GARM categories. Creator proceeds through vetting with standard review.
• Medium risk. Some content touches sensitive categories but does not constitute a pattern. Creator is flagged for manual content review with specific flagged content and category visible to the reviewer.
• High risk. Recurring content in flagged categories. Creator is excluded by default. Exception requests require documented justification.

The operational value is in the “medium” classification. Every vetting system can identify obvious high-risk creators. The hard work is efficiently handling the ambiguous middle — creators whose content occasionally touches sensitive topics but are otherwise excellent partners.

Research consistently shows that audiences disengage as sponsored content ratio increases. A creator whose feed is 60% sponsored content may still have a technically adequate engagement rate, but the quality of that engagement — measured by comment depth, saves, shares, and downstream conversion — degrades.

Tracking sponsored content ratio as a vetting signal, with threshold alerts at 25% and 50%, gives your team a data-backed reason to prefer creators with a balanced content mix.

• Verified status. Platform-verified accounts have passed identity verification. Useful signal, but not sufficient for brand safety.
• Business account status. Business accounts have access to analytics and creator tools. Generally indicates a professional creator.
• Account age. Recently created accounts (under 6 months) should be scrutinized. New accounts are a common pattern for purchased or fraudulent profiles.
• Growth trajectory. Steady, organic growth over months versus sudden follower spikes. Growth rate is a more revealing metric than follower count.

No single indicator is decisive. The value is in the combination: a verified business account with steady growth and a low sponsored content ratio is a substantially different risk profile than an unverified personal account with recent follower spikes and a high sponsored content ratio.

Graph relevance measures how closely a creator is connected to your brand's category ecosystem. This is not just topical relevance but structural relevance — is the creator embedded in the network of creators, brands, and audiences that define your category?

A creator who regularly interacts with other creators in your category, whose audience overlaps with your existing customers, and who has worked with brands adjacent to yours has higher graph relevance than a creator who happens to post about similar topics but is disconnected from your category's creator ecosystem.

Persona fit evaluates alignment between the creator's content and audience profile and your campaign's target persona. This goes beyond demographic match to include psychographic alignment: interests, values, content consumption patterns, and purchasing behavior.

A creator might reach the right demographic (25–34 year-old women in the US) but attract them for the wrong reasons (comedy content when your brand is positioning as premium and aspirational). Persona fit scoring captures this distinction.

Based on the creator's collaboration history, posting frequency, and reliability patterns, how likely are they to respond to outreach, complete the collaboration, post within the expected timeframe, and follow campaign guidelines?

For returning creators (those who have participated in your previous campaigns), this score is informed by actual historical data. For new creators, it is predicted based on patterns from similar creator profiles.

The final dimension synthesizes all signals into a predicted performance range: expected engagement rate, expected reach, expected content quality, and confidence level.

The confidence level is critical. A prediction based on extensive historical data and strong signal alignment carries more weight than a prediction based on limited data for a creator with an unusual profile. Making confidence explicit prevents over-reliance on predictions where the underlying data is thin.

An interactive review deck presents scored creators to human reviewers one at a time, with context: why this creator was selected, what their scores are, and what their key content looks like.

Exploration mode (early rounds): The system presents a diverse set of creators representing different segments, archetypes, and score profiles. The goal is to learn what the human reviewer values — not just what the scoring model thinks is good, but what the reviewer's taste, intuition, and brand knowledge consider valuable.

Exploitation mode (later rounds): Having learned the reviewer's preferences from exploration rounds, the system narrows the selection to creators who match the emerging preference pattern. Each subsequent round gets closer to the reviewer's ideal creator profile.

This exploration/exploitation pattern is borrowed from recommendation system design, applied to creator selection. It produces better shortlists than either pure AI scoring or pure human review.

In a spreadsheet-based review, the reviewer marks creators as “approved” or “rejected” with no structured reason. This is fast but produces zero learning signal.

A structured review captures three pieces of information for each creator:

• Decision: Yes, no, or maybe.
• Reason: Selected from a set of predefined reason categories relevant to this campaign (“great brand fit,” “wrong audience demographic,” “too much sponsored content”).
• Optional free text: For nuances that predefined categories don't capture.

The reason data is what makes review productive beyond the immediate campaign. When you can aggregate that 40% of rejections in beauty campaigns cite “wrong audience demographic” as the reason, you know to tighten demographic filters in the discovery phase for future beauty campaigns.

Each review round refines the system's understanding of what your brand considers a good creator. Round 1 might present 10 diverse creators and receive feedback that 3 were “yes” and 7 were “no.” Round 2 uses that feedback to present 10 creators more closely aligned with the “yes” profiles. By round 3, the system is surfacing creators that human reviewers approve at high rates.

This compounding effect means that campaign #5 benefits from the review feedback of campaigns #1 through #4. The vetting system does not start from zero each time — it starts from a calibrated understanding of your brand's preferences.

When you run multiple rounds of shortlisting — iterating on discovery instructions, adjusting scoring weights, refining vetting criteria — you need a way to evaluate whether the changes improved the output.

• Statistics: How did the total pool, shortlist size, and average composite score change between runs?
• Overlap: How many creators appear in both runs? High overlap suggests minor refinements. Low overlap suggests a significant shift in selection criteria.
• Score distribution: Did the average quality of shortlisted creators improve?

This meta-evaluation prevents the common failure mode where a team iterates on vetting criteria without knowing whether the iterations are actually improving outcomes.