Use case — fraud & quality pre-screening

Pre-screen every publisher for quality before you approve it

A recruitment list is only as good as the sites that survive the first look. We read each candidate against 120M+ classified domains and return an auditable quality verdict — alive, fresh, real content, own inventory or not — using reproducible proxies, never invented traffic numbers.

hundreds of thousands

of candidate domains a single pre-screen run can process (estimate)

10+

quality & risk signals checked per domain

invented visitor numbers — proxies only

Request a Free Pilot See Free Samples

The job to be done

Manual approval doesn't scale, and rubber-stamping is worse

Every publisher you approve costs review time up front and payout risk later. Doing it by hand throttles growth; skipping it lets dead sites, thin content, and competitor storefronts into the program. Pre-screening puts a consistent, auditable bar between the application and the approval.

Dead & parked sites

Domains that no longer resolve, redirect to a parking page, or have not published in years still sit in applicant queues. They should never reach a human reviewer.

Thin & spun content

Templated, scraped, or purely generative pages wrapped around affiliate links look plausible in a list and convert nothing. They need a real-content check, not a keyword match.

Own-inventory competitors

A site that sells the same product you do is a competitor, not a publisher. Keyword filters wave it through; an own-inventory check removes the whole class.

How the pipeline pre-screens

From a raw applicant list to a graded verdict — five steps

Send the domains awaiting approval; get back a pass / hold / reject verdict per site with the evidence behind it, so a reviewer only ever looks at the borderline cases.

Resolve & canonicalize

Each applicant domain is resolved and canonicalized. Dead, parked, and redirect-only domains are caught here, before any deeper check spends effort on them.

Alive & freshness check

We confirm the site is live and reachable, and read publication recency — is it maintained, or a shell that last posted years ago?

Real-content read

An LLM reads the pages for genuine editorial depth versus thin, templated, or AI-spun filler — and flags original vs scraped content.

Own-inventory & risk flags

We detect sites selling their own inventory, competitor storefronts, and other disqualifiers you define — plus reproducible traffic proxies for context.

Grade & route

Signals roll up into a pass / hold / reject verdict with a confidence score. Clear passes and clear rejects auto-route; only holds reach a human.

Re-screen on cadence

Approved publishers drift. Optional re-screens catch sites that went dark, got thin, or pivoted into inventory after approval.

The quality signals

What the pre-screen actually checks

Each signal is extracted from what the site actually publishes and shipped with a confidence score. Green signals build quality; red signals are disqualifiers you can tune.

Alive & reachable Content freshness / recency Real editorial depth Original vs scraped content Human authorship signals CrUX presence (proxy) Rank group (proxy) Individual vs company run Sells own inventory (disqualifier) Thin / AI-spun filler (disqualifier) Scraped / duplicate content (disqualifier) Parked / dead (disqualifier)

Proxies only, by design. Traffic context comes from CrUX presence and a public rank group — reproducible and auditable. We never attach a modeled visitor count, because a number a reviewer cannot verify is worse than none.

Worked example

Three applicants, three verdicts

Illustrative verdicts on three candidate publishers. Domains are withheld and values are illustrative; your deliverable carries the real domains and every signal.

PASS 0.93

Aliveyes

Freshnessposts < 2 mo

Real contentoriginal, in-depth

Own inventoryno

Traffic proxyCrUX · group 2

HOLD 0.61

Aliveyes

Freshnessposts 14 mo old

Real contentmixed — some thin

Own inventoryno

Traffic proxynot in CrUX

REJECT 0.97

Aliveparked page

Freshnessno dated content

Real contentAI-spun filler

Own inventoryyes — storefront

Traffic proxynot in CrUX

A note on confidentiality

The verdicts above are illustrative and drawn from real screening output, so we follow standard confidentiality practice on public pages: domains are withheld and descriptions generalized. The published examples are intentionally not traceable to any site — including through a web search — which protects the publishers without changing the underlying data. Your pre-screen deliverable contains the actual domains with every signal and confidence score, so each verdict can be verified directly. Start with a free pilot →

The deliverable

A graded queue you can route automatically

A deduplicated CSV — one row per canonical domain, each signal confidence-scored — that drops into your approval workflow. Clear passes and rejects route themselves; only holds need a human. Values below are illustrative.

canonical_domain	verdict	alive	fresh	real_content	own_inventory	traffic_proxy	conf
applicant-a.example	pass	yes	yes	original	no	crux_g2	0.93
applicant-b.example	hold	yes	stale	mixed	no	none	0.61
applicant-c.example	reject	parked	no	spun	yes	none	0.97
applicant-d.example	reject	dead	no	none	no	none	0.99

You set the thresholds: which signals are hard disqualifiers, what confidence auto-approves, and what routes to manual review are all tuned to your program's risk appetite during setup.

The measured basis

Screening proven at scale

The same reading engine that grades your applicants screened an entire vertical for a large-scale production run on our own classification platform — the reference for how it holds up across millions of domains.

domains screened in one run

precision on a reviewed sample

classified domains as the base

vertical domains identified

120M+

classified base

1.73M

vertical matched

1.2M

screened

~96%

precision confirmed

Reference: a large-scale production run on our own classification platform. Your pre-screen run is scoped to the applicants you send.

Questions

Fraud & quality pre-screening — FAQ

Is this fraud detection or quality screening?

It is content-and-site quality pre-screening: we assess whether a site is alive, real, maintained, and a genuine publisher rather than a competitor storefront or spun-content shell. We do not claim to detect click fraud or transaction fraud — those need your network's behavioral data. This screen keeps low-quality and mis-categorized sites out before approval.

Why proxies instead of real traffic estimates?

Because a reviewer has to be able to verify a screening decision. We report whether a domain appears in the Chrome UX Report and its public rank group — both reproducible. A modeled visitor count would look precise and be unauditable, so we don't ship one.

How do you tell real content from AI-spun filler?

The read looks for genuine editorial depth — specificity, original detail, human authorship signals — versus templated, scraped, or purely generative patterns. It is a judgment with a confidence score attached, so you can set how strict the threshold is and send borderline cases to manual review.

Can I automate approvals off the verdict?

Yes. You set the confidence thresholds that auto-pass and auto-reject, and everything else routes to a human as a hold. Most teams start conservative — auto-reject only the clear dead/spun cases — and widen automation as they trust the grading.

What does the free pilot show for pre-screening?

We run the pre-screen on a sample of your applicant or existing-publisher list and return graded verdicts with the evidence and confidence per signal — so you can see how the bar behaves against real sites before scoping the full run. No cost, no obligation.

Can I re-screen publishers I already approved?

Yes. Approved publishers drift — sites go dark, let content go stale, or pivot into selling their own inventory after they join. A scheduled re-screen re-runs the same signals over your active roster and flags any that have fallen below your bar, so the quality gate applies for the life of the partnership, not just at signup.