How Sift makes decisions.
Sift has two shields: scam protection and news source context. This page explains exactly how each one works, what data we use, and where our judgments come from. We think you should know.
Scam Detection
Sift's scam detection runs on every page you visit and checks three things independently. All analysis happens locally in your browser — no data is sent anywhere.
1. Known scam domain list
We maintain a list of domains confirmed to host scam content — tech support fraud, fake prize pages, IRS impersonation, and similar schemes. If you visit one of these domains, Sift immediately shows a high-severity warning.
2. URL structure analysis
Sift analyzes the URL of every page for patterns common in phishing attacks, even on domains not in our list:
- Brand impersonation in subdomains — e.g. a URL containing "paypal" or "microsoft" that does not actually belong to those companies
- Unusually deep subdomain structure — five or more subdomain levels is a strong phishing signal
- IP address URLs — legitimate services almost never use bare IP addresses instead of domain names
3. Page content phrase scanning
Sift scans the visible text on the page for language patterns characteristic of common scams, grouped by category:
Tech support fraud
"Your computer has been infected", fake Windows Defender alerts, instructions not to close the browser, phone numbers presented as Microsoft or Apple support.
Fake prizes & rewards
"You have been selected", "claim your free gift", gift card payment requests, sweepstakes that require personal information to claim.
Government impersonation
IRS arrest warrant threats, Social Security suspension notices, Medicare card update requests, fake tax debt demands.
Financial fraud
Requests to verify bank account or credit card numbers, suspicious SSN requests, get-rich-quick investment promises, fake account suspension threats.
Important limitation: Sift catches known patterns, not every possible scam. New scam tactics emerge constantly. Sift is a first layer of protection — if something feels wrong, always stop and call a family member before providing any information or payment.
News Source Context
When you visit a news site or social platform in our database, Sift shows a banner with the source's bias rating and factual accuracy — and links to how AP News, Reuters, and PBS are covering the same story.
How we classify sources
Each source is assigned a bias category and a factual accuracy rating by evaluating it against a fixed set of criteria (below). These ratings are Sift's opinion — not statements of objective fact — and reasonable people may disagree with any individual rating. We are automating this classification using AI so that ratings stay current as sources change and emerge.
| Category | What it means |
|---|---|
| Far right | Strong conservative bias; frequently publishes misleading or unverified claims |
| Right-leaning | Conservative bias in framing and story selection; factual accuracy varies |
| Left-leaning | Liberal bias in framing and story selection; factual accuracy varies |
| Far left | Strong liberal bias; frequently publishes misleading or one-sided claims |
| Conspiracy / unreliable | Regularly publishes false, fabricated, or pseudoscientific content |
| Foreign state-affiliated | Funded or substantially controlled by a foreign government and operating primarily as a vehicle for state messaging — distinct from editorially-independent public broadcasters |
| Social platform | User-generated content with no editorial oversight — anyone can post |
How ratings are generated
Ratings are produced by evaluating publicly available information about each outlet against a consistent set of criteria. This classification is being automated using AI so that ratings can be refreshed as sources change and emerge:
- Factual track record — history of publishing claims later shown to be false, fabricated, or retracted.
- Sourcing and corrections — whether the outlet cites primary sources and visibly corrects errors.
- Framing and story selection — the degree and direction of political slant in what is covered and how.
- Separation of news and opinion — whether commentary is clearly distinguished from reporting.
- Transparency — disclosure of ownership, funding, and editorial standards.
The system is calibrated against the public methodologies of the established media-analysis organisations below, and the ratings are refreshed regularly so they keep pace with new and changing sources. As a safeguard, major outlets with consistently high factual-accuracy records are never automatically flagged.
Why we don't flag centrist or mainstream sources
Sources like CNN, NBC, the New York Times, and the Washington Post have acknowledged left-leaning tendencies in framing and story selection. We do not currently flag them because their factual accuracy ratings remain high across the analysis organisations we reference, and because the goal of Sift is to protect against misinformation — not to flag every source with a viewpoint.
We may add context for centrist sources in a future update, presented differently to reflect the distinction between bias and inaccuracy.
What Sift doesn't do
- For most flagged pages, Sift shows a dismissible banner and never gets in the way. For pages on Sift's known scam list, and for pages where multiple high-severity scam signals are detected at once, Sift covers the page with a clearly-labeled warning. Known-scam pages cannot be bypassed; very-likely-scam pages include a "continue anyway" button for cases where the user is certain they trust the site.
- Sift does not tell you what to think or what to believe.
- Sift does not collect, store, or transmit any of your browsing data.
- Sift does not flag satire sites, opinion-only publications, or sources outside our current database.
- Sift's scam detection is pattern-based and will occasionally produce false positives on legitimate pages.
Suggesting a correction or addition
If you believe a source is incorrectly rated, missing from our database, or if you've encountered a scam pattern we're not catching, we want to hear from you.