A vertical's publisher universe runs to tens of thousands of live sites — and it drifts every month. New publishers launch, old ones go dark, and plenty quietly change what they are. Scheduled re-runs over 120M+ classified domains re-screen the whole category against your locked ICP and hand you a clean quarterly delta: what's new, what died, what changed.
The first full run gives you a deduplicated universe of qualified publishers. But affiliate ecosystems are living things — and a database that isn't refreshed silently loses value in three directions at once.
Sites that didn't exist — or weren't yet on-topic — at the last run. These are the uncontested names your competitors haven't reached either, and they're only discoverable by re-scanning the full web, not by revisiting a fixed list.
Publishers that expired, went to a parking page, or abandoned the niche. Outreach to a dead domain wastes a rep's day and dents sender reputation. The refresh flags them dead/parked so they drop out of sequences.
A blog that pivoted, added its own store (a disqualifier), dropped affiliate links, or shifted language/geo. The verdict that was true last quarter may be false now — re-screening against the same ICP catches the flip.
Every cycle repeats the exact pipeline of your first run, with the ICP definition locked so the comparison is apples-to-apples. You review only what moved — not the whole universe again.
Your last delivered universe becomes the reference snapshot — every canonical domain, verdict, and confidence score, frozen as the state we diff against.
The whole vertical is pulled again from 120M+ classified domains — including domains newly classified into the category since the last cycle. Nothing is limited to the old list.
Every domain in scope is read by the LLM against the same required / preferred / disqualifying signal set, so a verdict change reflects the site changing — not the rules.
Each known domain is re-fetched. Expired, parked, redirect-looped, or hard-404 sites are marked dead/parked with the date first observed unreachable.
The diff engine sorts every domain into added, removed, or changed, and for changes records which signals flipped and how confidence moved.
You get a delta CSV of only what moved, plus a refreshed master universe in the same schema — ready to sync straight into your partner CRM.
The worked example below is drawn from real refresh runs. Because the publishers are independent sites, we follow standard confidentiality practice on public pages: domains are withheld and descriptions generalized, so the figures and profiles shown here are intentionally not traceable to specific sites — including through a web search. That protects the publishers without changing the underlying data. Your pilot and refresh deliverables carry the actual domains and every signal, so everything can be verified directly.
An illustrative quarterly delta for a program whose last run left them with roughly 41,000 qualified publishers. Every figure below is a labeled estimate extrapolated from calibration slices — your own deltas come from your real baseline and locked ICP.
Freshly launched or newly on-topic sites clearing the ICP for the first time — names not in any prior export.
Domains now expired, parked, or abandoned — pulled out of outreach before a rep ever hits a bounce.
Verdict or key signals flipped — e.g. a publisher that added its own store, dropped affiliate links, or changed geo focus.
The same full-web pipeline that built your universe re-runs on a cadence you set — monthly, quarterly, or annually — so the database stays a live asset instead of aging into a stale spreadsheet.
Reference engagement (a large-scale production run on our own classification platform): a 30M dataset → 1.73M travel-related domains → 1.2M ICP-screened → ~96% precision on a reviewed sample. Refresh figures on this page are estimates shown for illustration.
Same one-row-per-canonical-domain schema as your master universe, with change-tracking columns added so your CRM can act on each row automatically.
| Column | Example value | What it tells your team |
|---|---|---|
change_type | added / removed / changed | How to route the row: enroll, suppress, or re-review |
status | alive / dead / parked | Reachability at this cycle — suppress the unreachable |
first_seen / last_seen | 2026-02-14 / 2026-08-12 | When the domain entered and was last confirmed in the universe |
is_icp (prev → new) | fit → not-fit | Verdict flips, so pivots and disqualifications are explicit |
signals_changed | +own_store, −affiliate_links | Exactly which signals moved and in which direction |
confidence (prev → new) | 0.78 → 0.91 | Slice re-review by how much certainty shifted |
Start with a free pilot to see the full pipeline on your vertical, then keep the universe alive with scheduled deltas. First 20 qualified publishers and a full-run projection, free.
Request a Free Pilot