How Publishers Game the Pay-Per-Call System (And How to Catch Them)

Pay-per-call is a high-trust business. You trust your publishers to send real consumers with real intent. You trust that the calls hitting your buyers' lines are genuine leads, not manufactured ones. And you pay out accordingly — $50, $100, $150 per qualified call depending on the vertical.

That trust gets exploited every single day.

Fraudulent publishers have developed sophisticated playbooks to game pay-per-call networks, and most brokers are catching only a fraction of it. If you're only sampling a small fraction of calls for manual QC, you're leaving the door wide open.

Here's how it works, what it costs, and what you can do about it.

The Coached Call Farm

This is the most damaging fraud tactic in pay-per-call. A publisher recruits callers (or runs a small call center) and feeds them scripted answers designed to hit every qualification checkpoint on the buyer's side.

The caller doesn't actually need Medicare Advantage. They don't actually have water damage. They aren't actually interested in debt relief. But they've been coached to say the right things — the right zip code, the right qualifying conditions, the right level of urgency — so the call registers as a qualified lead.

A single coached call that pays out at $100 is $100 of pure loss. That caller will never convert to a customer for the buyer. Consider the scenario: a publisher sending 50-100 coached calls per day at that rate. That's $5,000-$10,000 in daily fraud from a single source. Over a month, the damage adds up fast — potentially six figures before anyone notices.

The reason it works: on the surface, these calls sound qualified. The caller says the right things. The duration looks normal. The disposition comes back as "qualified." It takes careful listening — or pattern analysis across many calls — to detect the script.

Velocity Manipulation

Legitimate publishers ramp traffic gradually as they optimize campaigns. Fraudulent publishers spike.

The pattern is recognizable once you know what to look for: a publisher sends 5-10 calls per day for a week, building a clean track record. Then suddenly they're sending 80-120 calls per day. The quality hasn't changed on a per-call basis (each call still sounds plausible), but the volume is designed to maximize payouts before the inevitable clawback.

Some publishers run this across multiple campaigns simultaneously. They'll push volume on three or four verticals at once, collect payouts, and disappear before monthly reconciliation.

The financial math is straightforward. Say a publisher pushes 500 fraudulent calls at a $75 average payout before getting caught — that's $37,500 in losses. Net 30 payment terms make this worse — by the time the invoice is paid and the fraud is discovered, the publisher has already cashed out.

Repeat and Recycled Callers

In this scheme, the same callers are routed through different publisher IDs, sub-IDs, or campaigns. A caller who already called in for a Medicare quote last week calls again this week through a different traffic source. The call looks new, but it's the same person — and they're either not a real lead or they've already been paid for.

More sophisticated operators rotate caller IDs using VoIP services, making number-level deduplication unreliable. They recycle callers across verticals (the same person calls for auto insurance on Monday and home warranty on Wednesday), or across networks entirely.

This is harder to catch because each individual call looks legitimate. The fraud only becomes visible when you analyze patterns across your entire call volume — same voices, same background noise, same qualifying answers across different "leads."

Short-Duration Gaming

Some payout structures trigger at a duration threshold — say, 90 seconds or 120 seconds of connected time. Publishers who know the threshold will coach callers to stay on the line just past the mark, then disengage.

You'll see a cluster of calls from a publisher with durations that land suspiciously close to the minimum — 92 seconds, 95 seconds, 121 seconds. Legitimate call durations follow a natural distribution. Gaming produces an unnatural spike right at the payout threshold.

This one is easier to detect with basic analytics, but most brokers aren't looking at duration distributions by publisher. They're looking at aggregate numbers.

Why Manual QC Fails

The standard approach to call quality control is human review: hire QC analysts to listen to a random sample of calls and flag issues.

The math doesn't work in your favor. If a publisher is sending 100 calls per day and 30 of them are coached, and you're only reviewing a handful, there's a real chance you miss the bad ones entirely on any given day. The publisher continues unchecked.

But the bigger problem isn't sample size. It's pattern recognition. A human QC analyst listens to calls one at a time. They can tell you if a specific call sounds suspicious. What they can't do is correlate patterns across 500 calls from the same publisher — the same phrasing, the same background noise, the same unnatural pauses before qualifying answers. That requires analysis at scale.

Manual QC also doesn't scale economically. A QC analyst can review maybe 40-60 calls per day (listening, noting issues, documenting). At $15-20/hour depending on market, the per-call cost adds up fast. If you're processing thousands of calls per day, reviewing all of them manually is a non-starter.

What to Actually Look For

If you're a broker or network operator, here's what should trigger deeper investigation:

At the publisher level:

• Sudden volume spikes without corresponding campaign changes
• Qualification rates significantly above network average (sounds good, usually isn't)
• Cluster of call durations near payout thresholds
• High qualification rate but low downstream conversion for the buyer

At the call level:

• Unnatural pauses before answers (caller being coached in real-time)
• Identical phrasing across multiple calls from the same source
• Caller details that don't match (says one thing, area code suggests another)
• Background noise consistent with a call center, not a consumer environment

At the pattern level:

• Same caller voices appearing across different publisher IDs
• Consistent call scripts (verbatim phrases repeated across callers)
• Temporal patterns (all calls come in during narrow windows)

Scaling QC With AI

The core problem is volume versus attention. You need to review every call to catch systematic fraud, but humans can't economically review every call. This is where AI-driven call analysis changes the equation.

ConvoQC transcribes and analyzes every call automatically — not a sample, every single one. Each call is flagged individually for coached call indicators, compliance issues, and other red flags. The dashboard then aggregates flag rates and conversion rates by publisher, so patterns that would be invisible in a sample become obvious when you can see every call from every source.

Fraud in pay-per-call isn't going away. But the brokers who treat QC as a data problem — not a headcount problem — are the ones who stop losing money to it.