AI With Amirthan

← All guides

Content · July 11, 2026 · 3 min read

Fact-check your content with an agent before you post

Before a post ships, an agent fact-checks every claim against primary sources and flags overclaims. One wrong claim costs trust you cannot buy back.

The problem

A wrong claim in a post costs trust you cannot buy back. The correction never travels as far as the mistake, and the comment section remembers. The claims most likely to burn you are the almost-true ones: platform behavior you half remember, advice you saved months ago, numbers that were right last year. An agent fact-check caught one of mine before it went out.

What I do

Before a post ships, an agent fact-checks every factual claim in it against primary sources: the platform’s own announcements, official accounts, and documentation. Blogs repeating each other do not count. The agent returns a verdict for each claim, quotes the source’s exact wording next to mine, and ends with a source list and the date of the freshest one. Claims pile up fast when your research runs on autopilot, so this check is the gate between research and publish. Only what survives feeds my carousel pipeline.

The claim that almost shipped

My notes said that adding music to an Instagram carousel puts it in the Reels tab. The verification pass checked that against Instagram’s own announcement and came back with a sharper verdict: true, but overstated. The official wording is that carousels with music are eligible to appear in the Reels tab. Eligible is not guaranteed. Distribution is still algorithmic, and the post does not become a Reel.

The same pass caught a sizing recommendation in my notes that turned out to be best-practice advice, not a platform rule, with current sources openly disagreeing about it. The agent did not flatten that into a confident claim. It marked the verdict directionally correct with caveats and kept the disagreement visible.

Your audience fact-checks after you publish. An agent lets you do it before.

The fact-check prompt

Paste this into Claude, or any agent with web search, together with your draft:

Fact-check this draft before I publish it. For every factual claim:
1. Find the primary source: the platform's own announcement, official
   account, or documentation. A blog repeating the claim is not a source.
2. Give a verdict per claim: verified, partly true with caveats,
   or unverifiable.
3. Flag overclaims by quoting the source's exact wording next to mine.
   "Eligible to appear" is not "will appear".
4. Label each claim as a platform fact or as best-practice advice.
5. Where credible sources disagree, keep the disagreement visible.
6. End with a source list and the date of the freshest source.
Then rewrite every claim that failed, using the source's precise wording.

Do this now

  1. Run the prompt above on your next draft before it ships.
  2. Rewrite every flagged claim using the source’s exact wording, and cut any number the agent could not verify.
  3. Make it a standing rule: no factual claim ships without a verdict.

Claude Code

Turn this into a Claude skill

Paste this into Claude Code. It reads this guide and builds a skill shaped to how you work. 

Read the guide at https://aiwithamirthan.com/guides/fact-check-your-content-with-an-agent-before-posting. Then build me a Claude Code skill from it, personalized to how I actually work.

1. Fetch and read the full guide at the URL above.
2. Look at my setup so the skill fits me, not a generic template: skim my repos, my CLAUDE.md / AGENTS.md files, my existing skills, and my stack.
3. Create a new skill (a SKILL.md with a clear name, description, and steps) that lets me do what this guide teaches, adapted to my tools and goals.
4. Tell me where you saved it and how to invoke it.

If anything about my setup is unclear, ask me one or two questions first.

Want to build this yourself?

Join Claude Code Academy: courses, build alongs, and builders turning AI into real systems. And I post a new build on Instagram nearly every day.

Join Claude Code Academy → Follow @aiwithamirthan