How WagerWard Detects Gambling Emails (It's Tricky)

When we first started building WagerWard, the task seemed straightforward: scan an inbox for gambling emails and flag them. How hard could it be? Search for "bet" and "casino," call it a day. (If you're curious why there are so many gambling emails in the first place, that's a whole separate story.)

It turns out this is a genuinely difficult challenge. The English language is full of gambling vocabulary used in non-gambling contexts, the gambling industry operates under thousands of domains and brand names, and the cost of getting it wrong goes in both directions—miss a gambling email, and a trigger slips through; flag a legitimate email, and you erode trust in the tool.

This post is a look behind the curtain at how our detection system works, why naive approaches fail, and what we learned building something that has to be right in a context where the stakes are real.

The Naive Approach (And Why It Fails Spectacularly)

The most obvious approach is keyword matching. Build a list of gambling-related words—bet, casino, poker, slots, jackpot—and flag any email that contains them.

Here's what happens when you do that:

An email from your friend: "You bet! See you at 7." Flagged.

A newsletter about the Natasha Lyonne TV show: "New episodes of Poker Face streaming now on Peacock." Flagged.

A scheduling email from your dentist: "We have time slots available next Tuesday." Flagged.

A community center announcement: "Join us for Bingo & Bowling night this Friday!" Flagged.

A travel newsletter: "Explore Venetian architecture on our guided tour of Venice." Flagged.

An email addressed to someone named Wynn: "Hi Wynn, here's the project update." Flagged.

Your aunt's email: "Bingo! I found the recipe you were looking for." Flagged.

A keyword-only approach generates so many false positives that it becomes useless. Users would spend as much time reviewing false flags as they currently spend dealing with actual gambling emails. The tool would create a new version of the same issue it's trying to solve.

The False Positive Hall of Fame

Over the course of building and testing the detection system, we've accumulated a collection of false positives that we refer to internally as the "hall of fame." A few highlights:

"You bet!" and "I bet you..." — Among the most common expressions in English. Appears in casual emails, marketing copy, newsletters—everywhere. The phrase "you bet" has essentially nothing to do with gambling in most contexts.

"Poker Face" (the TV show) — Natasha Lyonne's mystery series on Peacock. Streaming service newsletters mentioning this show would be flagged by any system that treats "poker" as a strong gambling signal. The show has nothing to do with gambling.

"Venetian" — This word refers to anything related to Venice. Art history, architecture, travel, restaurants, blinds, plaster finishes. It also happens to be the name of a casino in Las Vegas. A detection system has to understand that "Venetian plaster tutorial" and "Venetian Resort $200 free play" are completely different things.

"Time slots available" — The word "slots" in scheduling contexts has nothing to do with slot machines. Appointment emails, booking confirmations, event sign-ups—all use "slots" routinely.

"Wynn" — Both a common surname and a Las Vegas casino brand. An email to or from someone named Wynn should never be flagged. But an email from promotions@wynnlasvegas.com absolutely should be.

"Bingo!" — Used as an exclamation of success or surprise in everyday language. Also used for community social events—"Drag Bingo night," "Yacht Rock Bingo," "Bingo & Bowling." These are social activities, not gambling operations.

"Safer bet" / "Smart bet" — Marketing copy from all kinds of companies uses betting metaphors. "This jacket is a safer bet for unpredictable weather." A self-defense company promoting their product as "the smart bet for personal safety." None of this is gambling.

Each of these taught us the same lesson: individual words are not signals. Context is.

How Our Detection Actually Works

The system we built uses a layered approach. No single signal triggers a detection—instead, multiple signals have to converge before an email gets flagged.

Layer 1: Known Gambling Domains

The highest-confidence signal is the simplest: does this email come from a known gambling domain?

We maintain a curated list of gambling operator domains, broken into tiers. Tier 1 domains are definitive—if an email comes from draftkings.com or pokerstars.com, it's a gambling email. No further analysis needed.

This sounds easy, but the list requires constant maintenance. Gambling operators use dozens of domains (marketing subdomains, country-specific domains, promotional microsites). Affiliate marketers operate under their own domains. New operators launch regularly. The list grows every week.

Domain matching handles the clear-cut cases. But many gambling emails come from generic email service providers (Mailchimp, SendGrid, etc.) where the sender domain doesn't help. That's where the scoring system takes over.

Layer 2: The Scoring System

For emails that don't match a known domain, we use a weighted scoring system. The email's subject line and sender information are analyzed against several categories of patterns, each contributing a different weight to an overall score.

Strong gambling signals are terms with very little non-gambling usage: "sportsbook," "free spins," "deposit bonus," "parlay," "accumulator," "progressive jackpot." When these appear, they carry heavy weight.

Moderate gambling signals are terms that are common in gambling but also appear elsewhere: "bonus," "odds," "wager," "cashout." These contribute to the score but aren't enough on their own.

Promotional patterns work as multipliers. Gambling emails almost always contain promotional language—"exclusive offer," "limited time," "claim your bonus," "match deposit." When gambling vocabulary appears alongside promotional language, confidence rises significantly.

The scoring thresholds are calibrated so that a single moderate signal alone won't trigger a detection. You need convergence: multiple gambling signals, or a gambling signal plus promotional language, or a moderate signal plus a known gambling-adjacent domain.

Layer 3: The Fast Gate

Analyzing every email against the full scoring system would be wasteful. Most emails in anyone's inbox have nothing to do with gambling—newsletters from clothing brands, social media notifications, work emails, shipping updates.

The fast gate is a lightweight first pass. It checks for the presence of a small set of cheap triggers—terms that could indicate gambling content and warrant further analysis. If none of these triggers are present, the email skips the full scoring pipeline entirely.

This is an important performance optimization, but it also serves a detection purpose. It means that words like "venetian" or "wynn"—which only appear in our detection database as domain entries, not as cheap triggers—will never cause a non-gambling email to enter the scoring pipeline in the first place.

Layer 4: Innocent Phrase Detection

This is where the system gets nuanced. Even when an email contains a gambling-related word, the surrounding context can indicate non-gambling usage.

We maintain a library of "innocent patterns"—phrases where gambling vocabulary is being used in a clearly non-gambling way. Some examples:

• "you bet" / "we bet" — Idiomatic expressions
• "time slots" — Scheduling language
• "Poker Face" near "streaming" or "Peacock" or "Natasha" — The TV show
• "Bingo" near "bowling," "drag," or "community" — Social activities
• "safer bet" / "smart bet" — Metaphorical usage
• "lottery" near "visa," "housing," "school," or "admission" — Non-gambling lotteries

When an innocent pattern is detected, the email's score is reduced. This doesn't guarantee the email won't be flagged—if there are strong enough gambling signals elsewhere, they'll override the innocent pattern. But it provides a meaningful counterweight against false positives.

Layer 5: The Recovery App Whitelist

This is the layer we consider most important to get right.

WagerWard exists to help people in gambling recovery. Other tools exist for the same purpose—Gamban, GamStop, BetBlocker, Bet Breaker, Gamblers Anonymous, the National Council on Problem Gambling. These tools and organizations sometimes send emails that contain gambling-related terminology, because that's their entire domain.

An email from Gamban saying "Your gambling blocking software has been activated" contains the word "gambling." An email from GA about an upcoming meeting might mention "recovery from gambling." An email from the NCPG helpline confirming your chat session could contain a dozen gambling-related terms.

Flagging these as gambling emails would be deeply counterproductive—even harmful. Someone in recovery receives a supportive email from their blocking software, and the tool they're using to clean up their inbox flags it as a threat. That's the kind of ironic false positive that undermines trust completely.

So the recovery app whitelist runs first, before any other detection logic. If an email comes from a known recovery tool or support organization, it's immediately cleared. No scoring, no analysis, no possibility of a false flag. Recovery support emails are sacred.

The Ongoing Challenge

Building the initial detection system was hard. Keeping it current is harder.

Gambling operators change domains regularly. They launch new brands. They acquire competitors and rebrand them. They use affiliate marketers who operate under constantly shifting domains. What was a comprehensive detection list three months ago has gaps today.

The promotional language landscape shifts too. Operators test new copy, invent new offer structures, use different terminology. "Risk-free bet" becomes "bonus bet" becomes "second chance bet" as regulations and marketing trends evolve.

And the false positive landscape is equally dynamic. New TV shows, new businesses, new cultural phrases—the world constantly generates new non-gambling content that happens to use gambling vocabulary.

This means the detection system is never "done." It's a living system that requires regular updates—new domains, new patterns, new innocent phrases, new recovery apps to whitelist. We track gambling industry changes, review detection accuracy, and push updates regularly.

Honest Limitations

We think our detection system is good. We don't think it's perfect, and we don't expect it ever will be.

Here's what we're honest about:

Affiliate emails are hard. Emails from gambling affiliates often come from generic domains with promotional language that could apply to any industry. Without a known gambling domain to anchor the detection, these rely entirely on content signals, which makes them harder to catch reliably.

We don't read email bodies. WagerWard only analyzes metadata—sender address and subject line. We never access or store email body content. This is a deliberate privacy decision, but it means we're working with less information than a system that reads full email content. A gambling email with a vague subject line ("Your exclusive offer awaits") and a generic sender domain is harder to detect from metadata alone.

New operators take time to detect. When a new gambling platform launches or an existing one rebrands, there's a gap between when they start emailing users and when they appear in our detection database. We work to close that gap quickly, but it exists.

Privacy First, Always

Why This Matters

Detection engineering might seem like a dry, technical topic. But for the people who use WagerWard, it's the difference between an inbox that feels safe and one that doesn't.

Every false negative—a gambling email that slips through—is a potential trigger in someone's recovery journey. Every false positive—a legitimate email incorrectly flagged—is a reason to lose trust in the tool.

Getting this right isn't an academic exercise. It's the core of what we do, and we take it seriously. The system will keep getting better, because it has to.

If you're curious how this works on your own inbox, you can test it with a free scan—it covers the last 60 days and shows you what WagerWard finds without deleting anything.

For a complete walkthrough of all your options for dealing with gambling emails, check out our guide to stopping gambling emails in Gmail. And for a broader look at email's role in your digital environment, see our guide to managing your digital environment in recovery.