TruthTap

Inspiration

AI hasn’t just accelerated creativity—it has super-charged misinformation.

• The World Economic Forum lists “misinformation & disinformation” as a top short-term global threat for 2025—second year running.
• The 2024 Edelman Trust Barometer found 63 % of people believe government leaders deliberately mislead the public, with similar numbers for business and media.

If bad actors are using AI to deceive, I wanted to use AI to defend. Perplexity’s Sonar Reasoning Pro already delivers lightning-fast, cited answers, so instead of building yet another chatbot or research agent, I embedded that power right where misinformation appears—inside the page.

Highlight ➜ Right-click ➜ Truth. That became TruthTap.

What I built

1. Chrome extension Highlight any claim → “Fact-check with TruthTap” → verdict card (True / Misleading / Unverified) with confidence, citations and chain-of-thought.
   
2. Cloudflare Worker Prompts sonar-reasoning-pro, sanitises the claim, caches results 24 h in KV and enforces rate limits.
   
3. Smart cache. SHA-256(claim) → KV; the extension also stores the latest response in chrome.storage for instant redraws.
   
4. Privacy-first. No PII is stored; only a hashed claim string reaches storage.

What I learned

In user tests, UX trumped raw tech—people trusted the tool when it felt effortless: one click, colour-coded verdict, recognisable favicons. I trimmed everything else to keep that “tap for truth” flow under a few seconds.

Challenges

Early prompts returned verbose narratives that were hard to parse. Switching to a strict, structured JSON schema solved everything—smaller payloads, faster rendering, easier error-handling.

What’s next

I’ll add semantic caching: near-duplicate claims (“COVID-19 vaccines cause X” variants) will hit the same cache entry, saving even more API calls and boosting speed.

Built With

• chrome
• cloudflare
• node.js
• perplexity
• typescript