Spotting a fake resume isn't about detecting whether the candidate used AI to write it. That's a losing battle — more than half of all job seekers now use AI to polish their resumes, and most of them are being honest. What you're actually looking for is inflation: resumes that claim experience the candidate doesn't have, accomplishments without context, or backgrounds that pattern-match too cleanly to your job description.

Five specific patterns reveal inflation:

1. Overfit to the job description — every requirement appears verbatim, in the same order, with the same phrasing.
2. Suspicious specificity without underlying depth — precise claims at the surface level that collapse the moment you ask a single probing question.
3. Inflated accomplishments without context — numbers with no baseline, team size, timeframe, or market conditions.
4. Tenure and timeline anomalies — roles that don't match the stated accomplishments, suspiciously fast promotions, glossed-over gaps.
5. Language texture — the weakest signal, and the most commonly misused. Use with caution.

Humans can't catch these patterns reliably at 250 resumes per posting. AI agents can — not by reading for AI writing, but by cross-checking each claim against role tenure, company size, and what a plausible career in that space actually looks like. The rest of this piece walks through each pattern in detail, why this matters now, and what to do about it.

The Real Problem Isn't AI. It's Inflation.

The headline panic you keep reading — "50% of candidates use AI to write resumes" — is aimed at the wrong target. Using AI to write a resume is not fraud. It's the same kind of assistance as hiring a resume coach, using a template, or running your draft through Grammarly. Candidates are responding to a system that rewards polished presentation, and AI is the cheapest polish available.

That isn't what's broken.

What's broken is that AI made something else trivially easy: inflating what's on the resume. Claiming expertise you don't have. Scaling a one-person project into a team effort. Reverse-engineering the whole resume to match the job description so perfectly that every ATS filter waves it through. These things used to take effort. Now they take ninety seconds of prompt engineering.

The distinction matters, because if you focus on catching "AI writing," you'll waste your time and penalize the wrong candidates. You'll flag honest applicants who used ChatGPT to tighten their bullet points. You'll miss dishonest ones whose writing happens to read as natural.

The right target is the claim, not the prose. The question isn't "did AI write this?" It's "is what this resume claims actually plausible?"

Why Keyword Matching Made This Worse?

The Applicant Tracking System was designed in the late 1990s for a world where a posting got thirty applications and candidates wrote their resumes by hand. Its core logic — parse the resume, extract keywords, rank by density — was a reasonable proxy for effort back then. A candidate who took the time to include the right terms had probably read the job description carefully and was probably serious about the role.

None of that logic holds now. A candidate using AI can produce a resume perfectly matched to your keywords in the time it takes to read the posting. The signal the keyword match was designed to capture — candidate effort and attention — has been automated away. What's left is noise.

This is what "ATS bloating" looks like in practice: resumes padded with keywords pulled directly from the JD, often in blocks of language that mirror the job description's own phrasing. The resume looks like a perfect fit on the keyword match. It passes your filter. It lands in your shortlist. And the keyword match tells you nothing about whether the candidate can actually do the work — because the resume was specifically engineered to beat the filter, not to represent the candidate.

The ATS didn't fail. The ATS is doing exactly what it was built to do. The problem is that what it was built to do is no longer useful.

The Five Patterns of an Inflated Resume

Here's what to actually look for when you're reading a resume for plausibility rather than for keyword match.

Pattern 1: Overfit to the JD

Real careers are messy. They include sidesteps, stretches, pivots, tools the person learned for a specific project and never used again. A resume of a real career reads as a record of real work — uneven, specific, sometimes including things that have nothing to do with the role you're hiring for.

An inflated resume reads as a mirror of your job description. Every required tool appears. Every preferred qualification is met. The bullet points track the JD's responsibilities almost in order. The language echoes your language — sometimes verbatim.

When a resume tracks your JD too cleanly, it's worth asking: "Has this person actually done all of these things, or did they write a resume to match this specific posting?" The answer usually becomes clear in the first probing question of a phone screen.

Pattern 2: Suspicious specificity without underlying depth

Inflation loves specificity. Vague claims get flagged. Specific claims feel authoritative. So inflated resumes often include detailed-sounding accomplishments: "led migration of 40TB of data to Snowflake across 14 production systems", or "scaled customer support operations from 3 to 27 agents across 4 time zones"

The test is simple: ask one probing follow-up. "What was the hardest decision you had to make during that migration?", "Which of those hires turned out to be the most difficult to onboard, and why?" Real experience produces a confident, textured answer. Inflated claims produce generalities, deflection, or pivots back to the surface-level numbers.

The pattern doesn't prove fraud on the page. It flags claims that need verification.

Pattern 3: Inflated accomplishments without context

A number without context is theatre. "Increased revenue by 340%." From what baseline? Over what time? With what team? In what market? A candidate who grew revenue from $100K to $440K in a tiny niche during a category boom has done a very different thing than one who grew $50M to $220M against headwinds.

Real accomplishments come with context because the person doing the explaining naturally reaches for it — they remember the scale, the team, the conditions. Inflated accomplishments come stripped of context because context would weaken the claim.

Look for numbers that sit naked on the page. The absence of context is the signal.

Pattern 4: Tenure and timeline anomalies

This pattern is the hardest to fake and therefore one of the most useful to check.

Look for mismatches between stated accomplishments and the time available to produce them. "Led a 200-person engineering organization" after six months in the role is extremely unlikely. "Rebuilt the go-to-market motion" during a three-month tenure almost certainly means something weaker than "rebuilt." "Led the Series B fundraise" from a candidate whose role was Director of Marketing at the time merits a question.

Also watch for glossed gaps — unexplained 8-month breaks between roles, or roles listed without exact month ranges where every other role has them. These aren't automatically red flags. A lot of people have legitimate gaps. But a resume that hides its timeline is a resume that might have a reason to.

Pattern 5: Language texture (use with caution)

This is the pattern most often misused, and it deserves a careful warning.

AI-generated writing has certain tendencies: a particular smoothness, a taste for tricolon structures, a preference for certain transition words. A lot of articles about "spotting AI-written resumes" focus on these tells. Don't.

Smooth writing isn't proof of dishonesty. It disproportionately penalizes non-native English speakers, candidates who worked with editors, and frankly, good writers. Using language texture as a primary detection signal will cost you qualified candidates and won't consistently catch inflated ones — because the inflated resumes that actually matter are the ones whose substance is fabricated, not the ones with a particular sentence rhythm.

Treat language texture as a weak, late-stage signal, and only in combination with other patterns. If a resume also shows overfit to the JD, specificity without depth, and timeline anomalies — then yes, smooth AI-feeling language adds to the overall picture. On its own, it proves nothing.

Why Humans Can't Catch This at Scale?

A thoughtful human reader can catch most of these patterns on any single resume. The problem is that a thoughtful human reader can't read 250 resumes in one sitting without losing the thread.

By resume 83, your pattern-matching is fatigued. By resume 140, you've forgotten what resume 14 said. By resume 220, you're scanning for keywords just to get through the pile. That's not a failure of professionalism — it's a limit of human attention applied to the wrong kind of work.

Worse, some of these patterns only reveal themselves across the batch. If three candidates for the same role all describe their accomplishments using nearly identical phrasing, that's a strong signal something's off. But you can only see it if you're holding all three in your head at once, and at 250 applications, no human can.

This is the specific job that was never done well by the ATS and will never be done well by human effort alone. It's a scale problem, and it requires a tool built for the scale.

How AI Agents Verify What Resumes Claim?

The tool for this job isn't a fraud detector. It's a plausibility checker.

Reasoning-based AI agents — the kind built specifically for resume analysis — do four things that humans at scale cannot:

Cross-check claimed accomplishments against role context. A candidate claiming to have "led a company-wide transformation" while in a Senior Manager role at a 10,000-person company is making a claim the agent can flag as implausible based on the scope that role typically has.

Compare resumes within a batch. The agent notices when seventeen candidates for the same role describe their accomplishments with suspiciously similar phrasing, or when three resumes all claim unusually precise ownership of the same rare tool combination.

Generate targeted follow-up questions. The agent reads each claim and generates the probing question that would verify it — "Tell me about the hardest technical decision you made during the 40TB Snowflake migration." This produces a phone-screen question set specific to each candidate, not a generic template.

Flag pattern-matching to the JD that exceeds plausibility. When a resume's language overlaps with the JD's language beyond what any real career would, the agent notices. A human reader might sense something is off. The agent can say exactly which phrases map to which JD lines.

None of this replaces a human reviewer. It surfaces the resumes and the specific claims that need human judgment, at a scale no human could cover alone.

The Paper Tiger Problem

There's a name for the hire this whole system is designed to prevent.

A "Paper Tiger" is a candidate whose resume looks like an A Player and who performs like a B or C Player on the job. They pass every filter. They clear every phone screen. They handle the interview loop with rehearsed confidence. And then they start the role and the gap between the resume and the reality opens up — slowly at first, then unmistakably.

Paper Tigers are the most expensive hires a company can make, and it isn't close. The numbers run in multiples of the role's salary: the lost output, the team morale cost, the time the manager spends coaching instead of executing, the eventual performance management process, the gap when the role is empty again, the second recruitment effort. Industry research consistently finds that 74% of companies admit to suffering bad hires from poor fit or misaligned skills. A meaningful share of those bad hires are Paper Tigers.

Historically, catching a Paper Tiger required three rounds of interviews and often a reference check, because the resume and the first two screens didn't give you enough signal. The inflation was real but plausible enough to survive the early filters.

That's the specific problem reasoning-based screening was built to solve. AgentR's resume intelligence verifies what candidates claim at the scale of every application — not just the shortlist. It catches most Paper Tigers before the first phone call, which is where they're cheapest to catch. Companies using this approach report 59% fewer Paper Tiger hires. Not because the AI is harsh — because the AI doesn't get tired at resume 83.

The faking problem isn't going away. It's going to accelerate. By 2028, industry forecasts put 1 in 4 applications at materially fraudulent. The question isn't whether your screening needs to evolve. It's how quickly.

The five patterns above are the framework. The tools that verify them at scale are the answer.

See how AgentR handles this at the application layer: [agentr.global](https://agentr.global).

Next: Lesson 11 — How AI Reads a Resume — And How You Should?