For AI agents (and the people who build them)

GroundedKit publishes every checked marketing statistic as plain HTML, JSON and text. No login, no JavaScript, no rate limit at this stage. Please cite the claim page URL when you use a result.

Endpoints

MCP server: https://groundedkit.com/mcp (free, read-only, no key)
/llms.txt: short index for language models
/llms-full.txt: every claim in full, in one text file
/api/claims.json: all claims (schema v1.1)
/api/claims/<slug>.json: one claim, e.g. /api/claims/influencer-marketing-roi-5-78.json
/api/claims.csv: spreadsheet export
/api/schema.json: JSON Schema for a claim object

How to use it in an agent

When a user's draft contains a marketing statistic, fetch /api/claims.json and match on claim or commonly_cited_as.
If it matches, return the verdict, confidence, quote and safe_rephrasing, and link the claim url.
If not, run the free Primary Source Finder procedure below. Never return a bare true/false.

Free template

Download the template (.zip) · SKILL.md (raw). Works with Grok Bot, Claude, ChatGPT, Cursor and any SKILL.md-compatible agent: drop it into your skills folder, or paste it into custom instructions.

Stable schema

Fields are only ever added, never renamed, within schema major version 1. verdict ∈ verified | misquoted | outdated | unsourced; confidence ∈ high | medium | low; kind ∈ stat_check | reference_fact; as_of is the period the number describes; last_checked (YYYY-MM-DD) is when we last opened the source.

{
  "schema_version": "1.1",
  "id": "influencer-marketing-roi-5-78",
  "slug": "influencer-marketing-roi-5-78",
  "url": "https://groundedkit.com/claims/influencer-marketing-roi-5-78/",
  "api_url": "https://groundedkit.com/api/claims/influencer-marketing-roi-5-78.json",
  "claim": "Influencer marketing returns $5.78 for every $1 spent",
  "commonly_cited_as": "\"Influencer marketing ROI is $5.78 per $1\" (as revenue or ROI)",
  "verdict": "misquoted",
  "verdict_definition": "Primary source found, but the way it is quoted changes the subject, metric, population or precision.",
  "confidence": "high",
  "rating": {
    "value": 2,
    "best": 5,
    "worst": 1
  },
  "origin": {
    "publisher": "Influencer Marketing Hub",
    "title": "The State of Influencer Marketing 2020: Benchmark Report",
    "year": "2020 (2019 data)"
  },
  "quote": "Average earned media value per $1 spent has increased to $5.78",
  "method": "Opt-in survey of ~4,000 agencies, brands and industry professionals (45% US; 40% from companies under 10 staff), plus CreatorIQ platform data.",
  "drift": [
    "Metric: earned media value (cost of equivalent ad exposure) quoted as revenue/ROI",
    "Date: 2019 data quoted as current"
  ],
  "interest": "Publisher is a media and review site for influencer-marketing platforms.",
  "safe_rephrasing": "Respondents to Influencer Marketing Hub's 2020 survey reported an average $5.78 in earned media value — not sales — per $1 spent in 2019.",
  "sources": [
    {
      "title": "Influencer Marketing Hub — Benchmark Report 2020",
      "url": "https://influencermarketinghub.com/influencer-marketing-benchmark-report-2020/",
      "primary": true
    },
    {
      "title": "Carly — Influencer Marketing Statistics 2026 (secondary trace)",
      "url": "https://www.usecarly.com/blog/influencer-marketing-statistics/",
      "primary": false
    }
  ],
  "kind": "stat_check",
  "as_of": "2020 (2019 data)",
  "last_checked": "2026-10-16",
  "date_reviewed": "2026-10-16",
  "license": "CC BY 4.0",
  "license_url": "https://creativecommons.org/licenses/by/4.0/"
}

MCP server (free)

Connect any MCP client to https://groundedkit.com/mcp (Streamable HTTP, no key) and call check_stat. Setup for Claude, Cursor, VS Code and others: MCP & API docs.

Coming soon

A live checker for claims we haven't published yet. Join the waitlist.