Inbound Is Dead: 91% Fraud, 800k Recruiter Hours, 214 Senior Sourcers

Greenhouse's 2026 AI Hiring Report landed with a number that should end an internal debate at every talent org: 91% of recruiters and hiring managers have spotted or suspected candidate deception. The Pragmatic Engineer's July 2026 hiring-market piece backs it up, quoting hiring managers whose inboxes are "full of AI slop, sometimes from bogus candidates." Some teams have stopped reading inbound applications entirely. The question isn't whether to filter harder. It's whether inbound is still a channel worth defending.

The headline number is 91%, but the operational number is 34%

91% of recruiters have caught candidate fraud in the last year, and 34% now spend up to half their week filtering spam and junk. That second figure is the one that actually breaks the funnel.

Do the math on the Greenhouse survey against the US recruiting workforce. In Refolk's index of professional profiles, there are roughly 108,810 people in the US carrying a "Recruiter" or "Technical Recruiter" title. If a third of them are burning half a week on spam, that's on the order of 800,000 recruiter-hours per week going straight into the AI-slop shredder. Fraud isn't the crisis. Recruiter-hours-lost is the crisis, and it flips the ROI math on outbound before you factor in a single deepfake.

800,000

The 91% figure is also a floor, not a ceiling. Fabric AI's platform tracked 19,368 live interviews between July 2025 and January 2026 and flagged 38.5% of candidates for AI-cheating behavior. That rate tripled in three months. Pindrop's 2025 Voice Intelligence Report clocked a 1,300% year-over-year jump in deepfake fraud attempts in hiring; Sumsub measured a separate 1,100% surge in North America alone in early 2025. Whatever your inbound funnel felt like in 2024, it is a different animal now.

What "candidate fraud" actually looks like in 2026

Candidate fraud in 2026 is mostly boring resume inflation at massive volume, not Mission Impossible face-swaps. Deepfakes get the headlines; exaggeration and fake references do the damage.

Greenhouse's breakdown of the most common deception tactics:

• Resume exaggeration: 63% of flagged deception incidents
• Fake references: 48%
• AI-assisted interview responses: 35%
• Fake voices or backgrounds (US): 32%
• AI scripts (US): 32%
• Deepfakes (US): 18%

The mechanism is volume, not sophistication. AI collapsed the cost of applying to a job to effectively zero. A candidate can now generate a tailored resume, a matching cover letter, and a plausible reference persona in under a minute. Multiply that by every open role at every company with a public careers page and you get the inbox everyone is describing. This is why "smarter ATS filters" keep losing: the attacker's cost curve dropped faster than the defender's.

The DPRK subplot

The one non-boring category is state-sponsored fake applicants, and it is not a rounding error. KnowBe4's Stu Sjouwerman went on record: "We get North Korean fake employees applying for our remote programmer/developer jobs all the time. Sometimes, they are the bulk of the applicants we receive." KnowBe4 famously hired one in July 2024. The Christina Chapman laptop farm case, prosecuted in the US, confirmed this is an industrialized supply chain with domestic co-conspirators, not a curiosity.

Google's Threat Intelligence Group now reports DPRK IT workers pivoting to European targets as US enforcement tightens. If you run hiring in the UK or EU and felt insulated in 2024, that insulation is gone in 2026. The Pragmatic Engineer's H1 2026 reporting matches this: ghosting is more commonplace than in the US, and fake applicants are a bigger issue.

Why "just screen harder" is the wrong answer

Screening harder loses on both economics and candidate experience: it costs more per applicant than sourcing costs per candidate, and it actively repels the real hires you want. AI resume screening is fighting the last war.

Two data points make this concrete. First, 70% of US candidates who experienced an AI interview say the AI evaluation wasn't clearly disclosed to them. Second, 38% of candidates have walked away from a hiring process because it included an AI interviewer. Companies leaning on AI screening to fight AI fraud are quietly losing the real candidates they're trying to protect. The arsonist and the fire brigade are the same vendor.

The attacker's cost curve dropped faster than the defender's. Every extra filter you bolt on is one more reason your best candidate closes the tab.

Vendors are of course selling the defense. Greenhouse's own "Real Talent" product combines identity verification with anti-fraud and anti-spam controls including IP and email domain blocking. It will help at the margin. It will not resurrect inbound as a primary channel, because the underlying economics have not changed: applying is free, sourcing is not, and only one of those two sides has a natural quality filter (the recruiter's own targeting).

The outbound pivot everyone is recommending, and nobody is staffed for

The consensus fix (go outbound) collides with a structural problem: the US recruiting industry is not staffed to source. There are 85 recruiters for every dedicated sourcer, and only 214 senior sourcing operators in the entire country.

Here's the shape of the workforce, from Refolk's index against the public Greenhouse and Fabric AI figures:

Read those top three rows again. If every US company decided tomorrow to run a sourcing-first funnel, there are 1,275 people in the country whose actual job title says they know how to do it, and 214 who could lead the function. That's a bidding war waiting to happen. Expect sourcer comp to move first and fastest in 2026.

Where the sourcers already are

The top employers of dedicated technical sourcers in Refolk's index skew heavily AI-native and high-growth: Anthropic, Rippling, MongoDB, Verkada, Zoox, Zipline, EvolutionIQ, Fetch Rewards. This is not a coincidence. These are the companies that figured out first that inbound to a hot AI brand is now indistinguishable from a DDOS, and reallocated headcount from screening to sourcing before the Greenhouse report gave everyone else permission.

What a sourcing-first funnel actually looks like

A sourcing-first funnel replaces "screen 10,000 applicants down to 20" with "identify 200 qualified candidates and outreach 40." The trust model inverts: you verify before you invite, not after they apply.

The mechanical shift, in five moves:

1. Turn off the public "Apply" flood. Keep a careers page for signal, but stop treating the inbox as a pipeline. Redirect the recruiter-hours that used to filter spam into targeted outreach.
2. Define the role as a search, not a JD. "Staff backend engineer with distributed RL exposure and prior Series B tenure" is a query. "Passionate about our mission" is not.
3. Run the search against verified profiles. GitHub commit history, LinkedIn tenure, conference talks, published papers. Anything an AI can generate in a minute is not signal. Anything an AI can't fake without years of history is.
4. Verify identity before the first live interview. Video call from a known number, live coding on a shared screen, or a paid onsite trial. Do not let identity verification be the interviewer's problem.
5. Measure recruiters on candidates-sourced, not applications-processed. The old metric rewards volume, which is exactly what the attackers have infinite supply of.

This is the exact gap Refolk closes for teams making the switch: you describe the person in plain English, and get a ranked shortlist pulled from GitHub, LinkedIn and the open web, with the verifiable signals a fake applicant can't manufacture. When your funnel starts with a real person you found, not a stranger who found you, the fraud problem stops being your problem.

The 2026 hiring stack, in one sentence

The 2026 hiring stack looks like this: fewer applications read, more profiles searched, identity verified before interview, and recruiter comp tied to sourced hires. Every team that has already made this switch is a team you've heard of. Every team still trying to filter the inbox is a team that is quietly losing a workday a week per recruiter.

214

The uncomfortable version of this article is that inbound recruiting didn't die because AI got scary. It died because AI made the cost of applying zero, which broke the only economic assumption inbound was ever built on: that submitting an application was mildly costly, and therefore mild signal. Once that assumption goes, the inbox is noise, and noise doesn't get better with a smarter filter. It gets better when you stop listening to it and start calling the specific people you want. Refolk is built for exactly that call.

FAQ

Is inbound recruiting actually dead, or just harder?

For roles with any AI-adjacent brand pull (AI, infra, security, developer tools), inbound as a primary channel is functionally dead in 2026. The Greenhouse 91% deception figure and Fabric AI's 38.5% interview-cheating flag rate mean the base rate of noise is now higher than the base rate of signal. For lower-volume roles (senior specialist, niche domain, in-person), a well-defended careers page can still contribute, but it should not be more than 20 to 30% of hires. The rest has to be sourced.

How do I actually verify a candidate isn't a deepfake or DPRK operator?

Combine four checks: live video from a device tied to a verifiable phone number, live pair-coding on a shared screen (not a take-home), a paid onsite or hybrid workday before offer, and reference calls placed to numbers you found yourself rather than numbers the candidate provided. The Christina Chapman laptop farm case showed the supply chain has US-based co-conspirators, so IP geolocation alone is not enough. Identity verification tools like Greenhouse Real Talent help at the top of the funnel, but the last-mile verification is a human process.

If there are only 214 senior sourcers in the US, how do I actually hire one?

You poach from Anthropic, Rippling, MongoDB, Verkada, Zoox, Zipline, EvolutionIQ or Fetch Rewards, or you promote internally from your recruiter bench. Refolk's index shows these eight companies employ a disproportionate share of the senior sourcing talent, and their tenure patterns suggest a two-to-three-year sweet spot for outbound approaches. Given the 85:1 recruiter-to-sourcer ratio, the faster path for most teams is training two or three of your best recruiters into sourcing operators, then hiring one senior leader to run them.

What's the single first move if my inbound is drowning right now?

Turn off automatic resume forwarding to hiring managers for one week, and reassign the recruiter-hours you save to sourcing 20 named candidates per open role. Measure: interviews-per-hire and offer-accept rate. In almost every case we've seen the outbound cohort converts 3x to 5x better than the inbound cohort, which gives you the internal proof to make the reallocation permanent. Start with your two hardest-to-fill roles, not your easiest, because that's where the inbound noise is worst and the outbound math is most favorable.