Trending Products Radar — AI-Scored Amazon Trends

Raw bestseller scrapers tell you WHAT is moving. This Actor tells you whether it's worth acting on.

Every run pulls trending product candidates from Amazon's ranking charts (New Releases, Best Sellers, Most Wished For — US and Canada), then scores each one 1–10 with Claude using a hindsight-validated rubric: demand trajectory, competition saturation, real buyer-intent search terms, and — critically — event-window timing (is this product riding a dated event you can still get ahead of?). You get a ranked, decision-ready list instead of 200 undifferentiated product names.

Why scoring beats raw rank (a validated result)

In an April 2026 backtest we scored 9 trending products and checked their outcomes 3 months later:

• The model's top pick was a World Cup match ball — flagged with an explicit event window 3 months before the final, exactly when acting on it still paid.
• The model's lowest pick (a saturated drink-bottle accessory) died, as predicted.
• The score rank-ordered winners and losers almost perfectly — while the raw trend rank had negative signal (top-of-chart items were often already peaked).

This is a small, honest sample (n=9, one cohort), not a guarantee — but it's exactly the judgment layer raw scrapers don't have: candidate generation is a commodity, the filter is the value.

Use cases

• Dropship / FBA sellers — triage 200 candidates down to the 5 worth testing, with buyer-intent keywords for listings and ads, and event windows so you don't stock last month's trend.
• Content & newsletter operators — a ranked "what to cover this week" feed with ready-made search terms for titles and SEO.
• Affiliate / commerce media — competition notes tell you where content is already saturated before you write.
• AI agents — a clean, structured "what's trending and is it any good" tool: one call, ranked JSON out.

Input

Field	Type	Default	Description
marketplace	select	amazon.com	amazon.com or amazon.ca
categories	multi-select	all 8	home-garden, kitchen, sporting-goods, pet-supplies, beauty, electronics, office-products, toys-and-games
trendSource	select	new-releases	Trend signal: New Releases (products breaking out before the main charts), bestsellers, or most-wished-for
maxProductsPerCategory	integer	25	1–100 candidates per category
minScore	integer	0	Only output products scoring ≥ this (scored mode only)
anthropicApiKey	secret	—	Required for scoring. Your Anthropic API key — used only for scoring, ~$0.01–0.02 of your Anthropic spend per 10 products. Without it you get UNSCORED raw candidates.

{
    "marketplace": "amazon.com",
    "categories": ["kitchen", "pet-supplies", "toys-and-games"],
    "trendSource": "new-releases",
    "maxProductsPerCategory": 25,
    "minScore": 7,
    "anthropicApiKey": "sk-ant-..."
}

Output

One dataset item per product, ranked best-first:

{
    "rank": 1,
    "name": "Adidas Trionda Official Match Ball 2026",
    "price": 34.99,
    "category": "sporting-goods",
    "url": "https://www.amazon.com/dp/B0EXAMPLE",
    "image": "https://m.media-amazon.com/images/I/example.jpg",
    "trendSignal": {
        "source": "new-releases",
        "rank": 7,
        "asin": "B0EXAMPLE",
        "rating": 4.8,
        "reviewsCount": 312,
        "currency": "USD",
        "marketplace": "US",
        "categoryName": "Sports & Outdoors"
    },
    "score": 9,
    "verdict": "strong_buy",
    "demand_notes": "Demand accelerating into the tournament; official-license products historically spike 2-3 months ahead of the final.",
    "competition_notes": "Single official supplier keeps listing competition low; content competition still thin.",
    "buyer_intent_terms": ["world cup 2026 ball", "trionda official match ball", "adidas trionda review", "best soccer ball 2026", "trionda vs al rihla"],
    "event_window": "FIFA World Cup Jun-Jul 2026",
    "scoredAt": "2026-07-20T12:00:00.000Z"
}

A RUN_SUMMARY object (mode, counts, upstream run id) is stored in the run's key-value store.

Cost transparency

Two separate costs, both small and both visible:

1. This Actor (your Apify account): platform compute plus the upstream chart scrape at ~$0.005/product. A typical 4-category × 25-product run ≈ $0.50–0.60.
2. Scoring (your Anthropic account, via your anthropicApiKey): batched calls to claude-sonnet-4-6, roughly $0.01–0.02 per 10 products. A 100-product run costs you ~$0.10–0.20 in Anthropic credit. No web search, no hidden calls — your key is used only for the scoring requests and never stored.

No key? The run still completes in unscored mode: raw ranked candidates with "verdict": "unscored" and a clear notice in the log.

Coverage & data quality

• Marketplaces: amazon.com (US), amazon.ca (Canada).
• Categories: the 8 highest-signal consumer departments (see input table). More on request.
• Trend charts: New Releases / Best Sellers / Most Wished For. Amazon retired its Movers & Shakers pages as a scrapeable surface in mid-2026 (they now render as empty client-side shells) — New Releases is the closest live "breaking out" signal, and it's what we default to.
• Products are deduplicated per chart, ranked by score (then chart rank). Failed scoring batches degrade to unscored rows rather than failing the run.

Fair use

Scores are model-generated judgments to prioritize your own research — not financial advice. Product data comes from public Amazon ranking pages via a maintained upstream scraper; no personal data is collected.