Draft for review, not yet in force.
Second Look checks posts you share with it (an image, some text or a link) and tells you what's true, what's missing and where the evidence is. This page says exactly what happens to what you send.
To check a post we send it to these services. Each processes it to answer our request.
| Service | What it receives | When |
|---|---|---|
| Anthropic (Claude) | The image, the extracted claim and the evidence we found | Every check |
| Google Cloud Vision | The image: a safety screen (SafeSearch), then a search for where it appeared online | Every image check |
| Google Fact Check Tools | A few keywords from the claim | Checks with a claim |
| OpenAI (content provenance check) | The image | Images without embedded content credentials, to see if an OpenAI tool made them |
| Websites you link to | A normal page request from our server | Link checks only |
An AI-image detector (Sightengine) is being evaluated but is off; this page will be updated before it is switched on.
Images that appear to contain sexual content are refused before any check runs, and nothing about them is kept except a safety-screen result tied to the image hash. To report illegal content, email secondlook.beta@gmail.com. We act on reports promptly and pass them to the authorities where the law requires.
Verdicts are our best reading of public evidence and can be wrong. Health-related checks carry a reminder that they are not medical advice.
One cookie, which keeps you signed in with your access code. No tracking or advertising cookies, and no analytics.
Draft of 22 November 2026. Back to Second Look