Fact-check a claim against the evidence
A claim checked against the web, the news, research papers and Wikipedia, with the original source found in the Internet Archive.
Steps
A fact-check traces a claim to its origin and weighs the evidence for and against it. With Agentik, your AI searches the web with Exa and the news with Serper, checks background facts on Wikipedia and research on OpenAlex, and finds archived copies of the original source in the Wayback Machine.
Step 1: Find the origin and evidence
Exa searches the web for the claim’s origin and evidence, with the pages’ text, and Serper searches the news.
Step 2: Check facts and research
Wikipedia checks background facts, and OpenAlex finds research on the topic.
Step 3: Find the original source
The Wayback Machine finds archived copies of the original source, in case it changed or disappeared.
Step 4: Write the verdict
Your AI writes a verdict with the evidence for and against, where the claim started and what is still uncertain.
Your agent does this part, with no tool call.
What you get
A verdict with evidence for and against.
You provide
- The exact claim
- Where you saw it
Your agent returns
- A verdict with its reasoning
- Evidence for and against, with links
- Where the claim started
- What is still uncertain
- The claim
- Where it started
- Evidence for
- Evidence against
- Verdict and what is uncertain
Placeholders. Your rows come from your run.
Each provider does one part
No single provider covers this job. Your agent calls each one for its part, from one connection.
Good to know
- Your AI separates what the sources say from its own reading.
- A claim with no evidence either way is unproven, not false.
When to use something else
For professional fact-checking with editorial review, rely on a fact-checking organisation. This playbook gathers the evidence fast so you can judge.
Questions
Does it decide for me?
It gives a verdict with its reasoning and separates what the sources say from its own reading.
What if there is no evidence either way?
The claim is unproven, not false, and the verdict says so.
What does the example cost?
At most $0.07, reserved before each call and settled at the measured usage.
Connect your agent to 2,400+ tools for free
Sign in with Google or GitHub. No card required.



