The premise: continuous observation, not sampling
A retrospective trend tool is essentially a search engine of yesterday's data. It looks back at the past week, month or quarter, summarises what was popular, and presents it neatly. That works for evergreen planning, but it answers a question editors had yesterday. The question they need answered today is different: what is emerging right now, and is it worth covering before the wave breaks?
EarlyForge is built around that second question. The engine never sleeps; it never samples. It listens continuously, on every signal that matters, in 50 countries and 28 languages — and turns the raw flood of mentions into something an editor can act on within minutes.
This is how it works, stage by stage.
Stage 1 — Continuous listening across thousands of sources
The detection layer maintains active connectors to thousands of data sources, refreshed every few minutes. The sources fall into seven families:
- News and editorial feeds — wire services, breaking-news aggregators, regional newsrooms
- Social platforms — short-form public discussion across the major networks
- Search-pattern signals — query-pattern indicators that surface intent before content exists
- Scientific and institutional sources — pre-print servers, regulatory filings, official press releases
- Financial and market feeds — listed-company news, commodity moves, crypto activity
- Niche communities — long-form forums, professional networks, specialist publications
- Cultural and entertainment feeds — music, gaming, video, books, lifestyle
Each new mention is normalised into a canonical signal record: source, timestamp, language, country, raw text, extracted entities. This layer is intentionally aggressive — the engine collects far more than it surfaces, because the value of pre-viral detection comes from seeing signals that do not yet look important.
Stage 2 — Recognising: cross-lingual root events
A story rarely breaks in one language. It breaks in one place and reaches the rest of the world by translation, by retelling, by parallel reporting. The detection engine clusters these into a single canonical entity called a root event.
When the same underlying story appears as "Bitcoin crashes" in English, "BTC s'effondre" in French, and "ビットコイン暴落" in Japanese, the root-event engine recognises that these are facets of one signal, not three independent ones. It merges them into a single trend record while preserving the per-language breakdown for editors who want to see how each market is framing the story.
This clustering also disambiguates homonyms — Apple the fruit and Apple the company never get fused into one signal — and applies category-aware time windows: short ones for breaking news, much longer ones for slow-moving topics like science or investigations.
Stage 3 — Scoring: five factors, continuously updated
A signal that has been observed and clustered is not yet actionable. It needs to be scored against editorial intent. The scoring engine combines five factors, refreshed continuously:
- Velocity — how fast the volume is rising
- Diversity — how many independent sources are picking up the topic
- Recency — how fresh the underlying event is
- Cohesion — how tightly the mentions cluster around one canonical story
- SEO potential — predicted search demand based on query patterns
The weights are tunable per team. A breaking-news desk tilts velocity and recency high; an evergreen blog tilts SEO potential and cohesion high. The same engine surfaces different signals to different teams without any retraining.
Stage 4 — Lifecycle: emerging → peaking → declining
Every scored signal is classified into one of three lifecycle stages. The classifier looks at the trajectory of the five factors over time, not at their absolute value:
- Emerging — score rising, volume below mainstream coverage, window of opportunity open
- Peaking — score plateauing, volume at saturation, last call for relevant coverage
- Declining — score falling, attention shifting elsewhere, editorial value diminishing
This classification is what makes "pre-viral" actionable. A topic in the emerging stage is, by definition, ahead of the wave. EarlyForge surfaces these first, in their own dashboard panel, separated from peaking topics editors might assume are still fresh but are in fact already crowded.
The lifecycle isn't just a label — it drives downstream behaviour. Alerts can be wired specifically to lifecycle transitions: notify the team when topic X flips from emerging to peaking, or when topic Y, previously declining, re-emerges. These transitions are often the editorial signal that matters more than the absolute score.
Stage 5 — Daily Pulse: a ranked daily snapshot
Every day, the engine takes a snapshot of the top 10 emerging trends globally, then per region and per category. This Pulse is the daily ranked view editors check at the start of their day.
Each Pulse snapshot is immutable once written. Yesterday's top 10 stays available; last week's stays available; last month's stays available. Over time this builds a public archive of editorial signals — a record of what was emerging, when, and how it evolved. The archive is searchable by date, by region, by category.
And it is genuinely public: no account, no wall. The daily top 10, the per-region and per-category views and the full dated archive are open to everyone — human readers and AI assistants alike. The Pulse serves two roles: a daily intelligence brief for editors, and a permanent, freely citable reference for how trend patterns have moved across markets. The real-time layer — live score curves, fresh sources as they land, alerts — is what an account adds on top.
The Pulse is also the most visible answer to the question "is EarlyForge actually seeing things earlier?". A reader can pick any day in the archive, look at the top 10, and compare to what mainstream coverage produced 6 to 48 hours later. The lead-time is measurable and recorded.
Stage 6 — From signal to sourced article
When an editor green-lights a topic, the editorial AI takes over. The brief — angle, length, sources, target keywords — is generated from the trend record. Briefs can also run on a recurring schedule: each new trend matching the brief's scope becomes one article, and the same trend never generates twice. Each brief can pin its own AI model — the managed default picks automatically, and teams that bring their own API key unlock the full model catalogue with no word quota.
The article is drafted with target keywords embedded from the start (not appended), with optimised meta description, slug, headings, citation policy and readability score.
The human review that follows is fast because the draft is already aligned with the editorial line, the SEO target and the source citations — and because the AI editor has full control of the document: ask it to restructure, rewrite one section or insert a sourced paragraph and it edits exactly the passage you point at, with every change tracked in a clickable version history. A round-trip from emerging signal to published article fits comfortably inside a one-hour editorial window — which is the window that captures the asymmetric SEO and social upside of being first.
Stage 7 — Publishing and broadcasting
The article publishes into the configured destination — WordPress, Ghost, Medium, LinkedIn, Blogger, Shopify, Wix, Webflow and more, with optional cross-syndication that preserves canonical authority on the apex domain, and a Zapier bridge for any external automation a team has wired up. Alerts fire in parallel: trends matching a team's configured categories and regions land in the right inbox the moment they appear.
The full cycle — listen, recognise, score, classify, brief, write, publish, alert — runs every minute of every day. Editors are not chasing the engine; the engine is queueing decisions for them.
Tracking how trends evolve
A single day of detection is useful. The compounding value of EarlyForge is in observing how trends evolve across days, regions, languages and categories.
Every signal carries a history, and every trend opens into a live view: the score curve as it moves, the geographic propagation map, the source list, and per-country context that explains how each market is framing the story. The same topic that surfaces on Tuesday in Japan can be tracked across its propagation to Europe on Wednesday and the Americas on Thursday. The five scoring factors update continuously, so a topic's trajectory is recorded in fine resolution. Editors can replay the rise — was velocity what mattered, or was it diversity? Did SEO potential predict the search wave that arrived two days later? Did the topic emerge in one region and propagate, or did it appear simultaneously in three?
This evolution view is what makes EarlyForge more than a daily list. The same engine that surfaces today's emerging signals is the engine that, over a quarter, lets a team understand which categories produce reliable lead-time, which regions act as predictors for others, and which signal patterns translate into editorial outcomes.
Why pre-viral detection compounds
Traffic distribution on emerging topics follows a power law. The first three articles on a rising story capture a disproportionate share of links, search clicks and social shares. The fourth article gets a fraction; the tenth gets noise. Detecting a signal six hours after it has crested is detecting it too late.
Pre-viral detection — surfacing topics while they are still emerging, with enough lead time for editorial review and publication — is what separates a content team that surfs the wave from a content team that arrives at the beach.
A single day of pre-viral detection is useful. A year of it changes a publisher's positioning. Topical authority compounds: every article published on time, on a relevant signal, with proper canonical integrity, reinforces the publisher's standing in search-engine and answer-engine knowledge graphs. Six months in, the publisher is not ranking individual articles — it is recognised as the authoritative entity on its categories.
EarlyForge is the architecture that makes that compounding sustainable. Detection runs continuously, scoring runs continuously, articles ship within the window, archives persist, evolution is recorded. The system gets stronger with every day it operates — and so does the publisher built on top of it.