How YLOAD News works

1. Source selection

We hand-curate a list of 40+ RSS feeds from established freight, maritime, air cargo, road and rail publishers — outlets like The Loadstar, FreightWaves, Splash247, Heavy Lift PFI, Air Cargo News, RailFreight.com and the trade-press industry standards. Sources are selected for editorial track record, reporting depth, and the relevance of their coverage to international freight forwarders. The full active source list is visible at /sources (when available), and every individual article on this site links back to its original publisher.

We do not scrape paywalled content. We only consume publicly available RSS / Atom feeds plus a small number of corporate press feeds where the publisher has chosen to make them public (e.g. AAL Shipping, Jumbo Maritime, dship Carriers, Kuehne+Nagel Newsroom).

2. Ingestion

A scheduled job (cron) polls every active feed every 15 minutes. New items are fingerprinted using SimHash so cross-publisher echoes of the same story (e.g. two outlets covering the same Maersk press release) are detected and grouped rather than duplicated. Each candidate item is fetched, the article body extracted server-side, and the cleaned text passed to the next stage.

3. AI rewriting & classification

Cleaned article text is sent to Google's Gemini 2.5 Flash model with a fixed system prompt. The model produces, for every article:

• A rewritten title and short summary in clear operational English (we never reuse more than 8 consecutive words from the source).
• A longer body summary (500–800 words) with a freight-forwarder lens — what does this mean for ops, rates, capacity, routing.
• A severity tag (info / medium / high / critical), a 1–5 relevance score, and a boolean is_news flag (opinion / sponsored / evergreen guides are filtered out).
• Extracted entities — ports (resolved to UN/LOCODE), carriers, regulations, transport modes, and editorial topic slugs from a canonical list of ~100 tags.
• SEO meta title and meta description.

The model is instructed to never invent facts — if the source doesn't state a figure or a port code, the model is required to omit rather than guess. The full system prompt is auditable in our repository.

4. Quality gates & publishing

Before an article goes live, automated gates check: (a) the rewrite hasn't plagiarised the source above an 8-word window, (b) the hero image meets a minimum resolution, (c) the relevance score is ≥ 4, and (d) the SimHash signature doesn't match an article already published in the last 7 days. Items that fail any gate are queued for editorial review rather than published.

Every published article keeps the original publisher name, byline (if provided in the feed), and original publication timestamp, plus a prominent "Read at the original publisher" link to the source URL.

5. Editorial oversight & corrections

The YLOAD News editorial team monitors the admin queue daily and reviews any item that the automated gates flagged. Articles can be edited, depublished, or marked as featured at any time. If a publisher or reader believes an article misrepresents the source, we will correct or remove it within 24 hours of a request to hi@getworkload.com.

6. AI disclosure

Every article on YLOAD News is rewritten by Gemini 2.5 Flash under the supervision of the YLOAD News editorial team. We disclose this prominently here, in the footer of every article, and on the About page. We do not publish AI-generated content that pretends to be a human-authored scoop — every story is a transparent summary of a real piece of reporting by a named publisher, with a one-click path back to the original.

7. What we don't do

• We don't generate news from prompts ("write me an article about X").
• We don't publish opinion or speculative AI-imagined market scenarios.
• We don't run native ads or sponsored content.
• We don't sell visitor data — see the Privacy Policy.