	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > AI Fact-checking Agent

AI Fact-checking Agent

Author: Venkata Sudhakar

In e-commerce and retail, incorrect product information can lead to customer disputes, returns, and regulatory issues. An AI fact-checking agent built with Google ADK and Gemini 2.0 Flash can automatically verify claims made in product listings, marketing copy, and customer communications against a trusted knowledge base before they are published.

The agent uses function tools to query a product database, cross-reference pricing records, and check specification sheets. Gemini evaluates each claim against retrieved facts and returns a structured verdict - verified, false, or unverifiable - with a confidence score and an explanation. This prevents inaccurate claims from reaching customers.

The below example shows ShopMax India verifying a product claim before it goes live on the website.

import google.genai as genai
from google.adk.agents import LlmAgent
from google.adk.tools import FunctionTool
from google.adk.sessions import InMemorySessionService
from google.adk.runners import Runner
import json

client = genai.Client(api_key="your-api-key")

# Trusted product knowledge base
PRODUCT_DB = {
    "SM-TV-55": {
        "name": "Samsung 55-inch 4K QLED TV",
        "price": 72000,
        "resolution": "3840x2160",
        "refresh_rate": "120Hz",
        "warranty": "2 years",
        "energy_rating": "4 star"
    },
    "AP-MB-M3": {
        "name": "Apple MacBook Pro M3",
        "price": 189000,
        "ram": "18GB",
        "storage": "512GB SSD",
        "battery": "22 hours",
        "warranty": "1 year"
    }
}

def lookup_product_specs(product_id: str) -> dict:
    # Retrieve verified specs from the product database
    if product_id in PRODUCT_DB:
        return {"found": True, "specs": PRODUCT_DB[product_id]}
    return {"found": False, "specs": {}}

def check_pricing(product_id: str, claimed_price: float) -> dict:
    # Verify claimed price against official pricing records
    if product_id not in PRODUCT_DB:
        return {"verified": False, "reason": "Product not found"}
    actual = PRODUCT_DB[product_id]["price"]
    diff_pct = abs(claimed_price - actual) / actual * 100
    return {
        "verified": diff_pct < 2,
        "actual_price": actual,
        "claimed_price": claimed_price,
        "difference_pct": round(diff_pct, 2)
    }

# Register tools
spec_tool = FunctionTool(func=lookup_product_specs)
price_tool = FunctionTool(func=check_pricing)

# Fact-checking agent
fact_checker = LlmAgent(
    name="fact_checker",
    model="gemini-2.0-flash",
    tools=[spec_tool, price_tool],
    instruction="""
    You are a fact-checking agent for ShopMax India.
    When given a product claim, use the available tools to verify it.
    Return a JSON verdict with fields:
    - verdict: "VERIFIED", "FALSE", or "UNVERIFIABLE"
    - confidence: 0.0 to 1.0
    - explanation: brief reason
    - corrected_claim: corrected version if false, else null
    """
)

session_service = InMemorySessionService()
runner = Runner(agent=fact_checker, app_name="fact_checker", session_service=session_service)

session = session_service.create_session(app_name="fact_checker", user_id="editor1")

claim = "The Samsung 55-inch 4K QLED TV (SM-TV-55) has a 144Hz refresh rate and is priced at Rs 72,000."

from google.genai import types
message = types.Content(role="user", parts=[types.Part(text=claim)])
for event in runner.run(user_id="editor1", session_id=session.id, new_message=message):
    if event.is_final_response():
        print(event.content.parts[0].text)

It gives the following output,

{
  "verdict": "FALSE",
  "confidence": 0.97,
  "explanation": "Refresh rate claim is incorrect. The SM-TV-55 has a 120Hz panel, not 144Hz. Price of Rs 72,000 is verified correct.",
  "corrected_claim": "The Samsung 55-inch 4K QLED TV (SM-TV-55) has a 120Hz refresh rate and is priced at Rs 72,000."
}

The agent returns a structured JSON verdict that upstream systems can parse programmatically. The corrected_claim field provides an immediate fix for editors, reducing the round-trip time for corrections. In production, integrate this agent into your content management system as a pre-publish hook so every product listing is verified before it appears on the ShopMax India website. Cache the product database locally and refresh it nightly to keep fact-checking fast and accurate.

Send your comments, suggestions or queries regarding this site to [email protected].