How to Find Influencers with an AI Agent
The reliable way to find influencers is to search by what people create, not how many followers they have: pull the accounts that repeatedly post about your topic, then rank them by how their audience responds to those specific posts. That inverts the usual directory approach, and it is exactly the kind of multi-step search-and-score job an AI agent can run from a single question.
This tutorial uses influencer-discovery, an open-source skill from our collection at Xpoz. The data layer is the Xpoz index of Twitter/X, Instagram, Reddit, and TikTok (billions of posts, free tier available); the skill encodes the discovery, scoring, and vetting method below.
Content-First Discovery
A follower count tells you an account is big. It doesn't tell you whether the account actually talks about your niche, whether its audience engages with that content, or whether the followers are real. Content-first discovery starts from topic searches ("AI agents" OR "agentic AI", or your category's equivalent) and returns the accounts authoring matching posts, with aggregates attached: how many on-topic posts each account made in the window, and the total likes and impressions those specific posts earned.
Those aggregates change the question from "who is big?" to "who moves this conversation?" An account with 40K followers whose topic posts average 900 likes outranks a 400K-follower account that mentioned the topic once. The window defaults to the last 30 days, so the ranking reflects who is active now, not who mattered two years ago.
How the Skill Scores Candidates
Every candidate gets an influencer score out of 100, built from five weighted factors:
| Factor | Weight | What it measures |
|---|---|---|
| Relevance | 30% | How often they post about the topic |
| Engagement | 30% | Average likes per on-topic post |
| Reach | 20% | Followers, on a logarithmic scale |
| Authenticity | 10% | Bot-likelihood signals inverted into a trust score |
| Consistency | 10% | Posting frequency across the window |
Reach is deliberately logarithmic and capped at 20 points: it prevents one mega account from drowning out the micro creators who typically drive niche engagement. The authenticity factor draws on Xpoz's inauthenticity scoring for Twitter/X accounts, so audiences padded with bots cost a candidate rank instead of inflating it.
Candidates are then classified two ways. By tier: mega (1M+), macro (100K-1M), micro (10K-100K), nano (1K-10K), because each tier suits a different goal (awareness, credibility, engagement, affordability). And by voice type: analyst, builder, educator, news, commentator, or community, read from the bio and recent posts, because a builder's endorsement and a news account's mention do different work for a brand.
Setup and First Run
Install the skill (Claude Code shown; the folder works in any SKILL.md-compatible agent):
git clone https://github.com/XPOZpublic/xpoz-agent-skills.git
cp -r xpoz-agent-skills/skills/influencer-discovery ~/.claude/skills/
Connect data access once, via the remote MCP server (OAuth, no keys) or a free SDK key from xpoz.ai/get-token:
claude mcp add --transport http --scope user xpoz https://mcp.xpoz.ai/mcp
Then ask in plain language:
"Find the top 20 AI agent influencers on Twitter"
"Who are the micro-influencers talking about sustainable fashion?"
"Find developer advocates who post about coding agents,
ranked by engagement"
"Crypto KOLs with high engagement but under 100K followers"
The report lists ranked candidates with follower counts, on-topic post counts, average engagement, tier, voice type, and an authenticity check, plus detailed profiles of the top 10 with sample posts and a one-line "why they matter". From there the conversation continues: "drop anyone below 0.9 authenticity", "show me only builders", "pull the last month of posts for #3".
Try this with Xpoz
No API keys needed. Query Twitter, Reddit, Instagram & TikTok with natural language.
Vetting Before Outreach
Two checks are worth running on every shortlist. First, authenticity: an account whose bot-probability score is high, or whose engagement is thin relative to its follower count, is a paid-reach risk regardless of how good the content looks. Second, consistency: a creator with three on-topic posts a week is a durable channel; one viral post is a moment. The skill surfaces both signals in the ranking, but the judgment call about brand fit stays yours, made faster by having sample posts and voice type in front of you.
FAQ
How do I find micro-influencers in my niche?
Search by content, not follower count: find accounts that repeatedly post about your topic, then rank them by engagement on those posts. An agent with a discovery skill runs the search from one question and classifies results into tiers; micro is roughly 10K-100K followers.
How can I tell if an influencer's audience is real?
Look at engagement relative to reach and at bot-likelihood signals. The skill folds an authenticity probability score into every ranking, so inflated accounts sink automatically instead of leading the list.
Are micro-influencers better than bigger accounts?
They serve different goals: mega and macro buy awareness, micro typically shows the strongest engagement and niche authority, nano is the most targeted and affordable. Tier classification exists so you can match tier to campaign goal.
Does this work for Instagram and TikTok too?
Yes. The skill defaults to Twitter/X and extends to Instagram creator search on request; the same content-first method applies to TikTok through hashtag and keyword search.
Next Steps
- The manual version of this method, query patterns included, is in How to Find Micro-Influencers Using Topic Keywords.
- TikTok specifics are covered in TikTok Influencer Discovery: Finding Creators by Niche.
- Turning a shortlist into a program: Building Influencer Lists That Actually Convert.
- The full skills collection and install pattern: How to Add Social Media Skills to Your AI Agent.
