You want to launch localized editions — an EN program, a DE program, an FR, ES and PL program — and each needs its own list of native-language publishers, not English sites with a translated footer. We pull your vertical from 120M+ classified domains, detect the real content language on every domain, and LLM-screen each market separately against the same ICP.
Localized programs fail when the partner list is machine-filtered by TLD or by a single hreflang tag. We read what each site actually publishes, so a French program gets publishers writing fluently in French — wherever they are hosted. Six steps, one per language edition, run in parallel.
Your category maps to our 120M+ classified-domain database. We pull every active domain in scope worldwide — the multilingual long tail included, not just the .com layer.
Each domain's primary language is detected from its actual body copy — not the html lang attribute, which is wrong constantly. Consistency across pages is scored so half-translated shells are caught.
Domains sort into EN, DE, FR, ES, PL and any other market you name. A site with genuine parallel editions is placed in each language it truly serves, flagged as multilingual.
The LLM reads every domain in a language bucket against your required, preferred and disqualifying signals. Native fluency and geo focus are screened alongside your vertical's own signals.
Mirrors and country variants collapse to one canonical domain. A multilingual publisher appears once per relevant edition, never as four near-duplicate rows in your outreach queue.
You receive separate deduplicated CSVs — one per market — each with confidence scores, so a regional manager can work their list without wading through other languages.
Detecting a language is the easy part. Telling a native publisher apart from an auto-translated doorway page is where localized programs live or die. Every domain is read against these, each with its own confidence score, on top of your vertical's signal catalog.
Need a market we haven't named — Italian, Dutch, Portuguese-BR, Nordic languages? Any language is a bucket; the screen is calibrated per market on a reviewed sample the same way.
An illustrative full run for an advertiser taking a consumer-finance program into five European editions. Category and per-language yields are estimates extrapolated from our classification database; your pilot returns real domains and a projection tuned to your exact ICP and markets.
Reference engagement (a large-scale production run on our own classification platform): 30M dataset → 1.73M vertical domains → 1.2M ICP-screened → ~96% precision on a reviewed sample. Language-split figures above are estimates, shown for illustration.
The profile below is drawn from a real screening run. Because these are independent publishers, we follow standard confidentiality practice on public pages: domains are withheld and descriptions generalized, so the profile is intentionally not traceable to the site — including through a web search — which protects the publisher without changing the underlying data. Pilot and client deliverables carry the actual domains with every signal, so everything can be verified directly.
Each language CSV carries the ICP verdict, every signal value, confidence scores, and the detected language and geo focus. Here is a single generalized row as it would appear in the German edition.
A TLD filter would have put this site in a generic bucket and a machine-translation shell in the same list. Reading the body copy separates a native German publisher from an English site with a bolted-on translation, so your DE manager never wastes outreach on a site that can't actually publish in German.
Every field ships with a confidence score, so a regional manager can work the highest-certainty native publishers first and leave the borderline multilingual cases for a second pass. Language and native-fluency scores are independent, so you can insist on both being high before a domain enters a market's outreach queue.
The free pilot runs your vertical, splits it by detected language, and delivers your first 20 qualified publishers per market plus a full-run projection. No cost, no obligation.
Request a Free Pilot