Inspiration

With the rise of misinformation, phishing, and politically manipulative emails, it can be difficult for users to quickly determine whether the information in their inbox is trustworthy. We wanted to create a lightweight tool that integrates directly into Gmail and helps users critically evaluate suspicious claims without needing to manually fact-check everything themselves. Our goal was to make fact-checking fast, accessible, and conversational.

What it does

Erm... Actually! is a Chrome extension that analyzes Gmail emails using AI to identify false, misleading, or unverifiable claims. When detection is enabled, the extension injects a floating chatbot directly into the Gmail page that:

Breaks down questionable claims into easy-to-read bullet points
Highlights misinformation, scams, or manipulative language
Allows users to ask follow-up questions about the email
Maintains conversational context for deeper clarification
Prevents spam requests with built-in flood protection and cooldown logic

The extension only activates when a valid Gmail email is open, ensuring it stays lightweight and context-aware.

How we built it

We built the project using:

Chrome Extensions Manifest V3
Vanilla JavaScript
Gemini 2.5 Flash API
Chrome Storage API
Chrome Tabs & Scripting APIs

The extension works by reading the currently opened Gmail email from the DOM, sending the content to the Gemini API for analysis, and then rendering the response inside a custom floating chatbot UI injected into the page.

We also implemented:

URL validation to ensure detection only works on actual emails
Persistent toggle state using chrome.storage.local
Cooldown and duplicate-request protection
Secure rendering with textContent to avoid XSS risks
Context-aware follow-up conversations

Challenges we ran into

One of the biggest challenges was working with Gmail's constantly changing DOM structure. Extracting email content reliably required careful handling of Gmail-specific class names and page states.

We also ran into issues with:

Preventing duplicate API calls when users spammed the toggle
Maintaining chatbot state across popup opens/closes
Synchronizing extension state with the currently active Gmail tab
Designing a chatbot UI that felt integrated without disrupting the Gmail experience

Another challenge was balancing responsiveness with API usage limits, which led us to implement multiple layers of flood protection.

Accomplishments that we're proud of

We're proud that we created a fully functional AI-powered Gmail assistant that feels seamless and interactive rather than just a basic popup tool.

Some accomplishments we're especially happy with:

Real-time fact-checking directly inside Gmail
Smooth floating chatbot interface
Persistent extension state across sessions
Follow-up conversation support with contextual memory
Strong security practices like avoiding innerHTML
Clean UX with clear validation and cooldown feedback

We also built the entire project without any frameworks, which helped us better understand Chrome extension architecture and browser APIs.

What we learned

Through this project, we learned:

How Chrome Extensions Manifest V3 works
How to inject and manage dynamic UI elements in existing webpages
Best practices for browser extension security
How to manage asynchronous communication between popup scripts and content scripts
How to integrate AI APIs into real-time applications
The importance of handling edge cases and preventing duplicate requests

We also gained experience designing AI systems that are constrained and context-aware instead of allowing unrestricted responses.