The AI Briefing

The daily top 5 is served as JSON at /api/ai-news.json — open (CORS *), GET, cached 30 min, refreshed daily ~06:30 UTC. Reuse with attribution to solvX and the original source.

Response shape

{
  "service": "solvX AI Briefing",
  "date": "2026-06-29",            // YYYY-MM-DD (UTC)
  "generatedAt": "2026-06-29T06:31:00Z",
  "count": 5,
  "stories": [
    {
      "rank": 1,                   // number, 1..5 ascending (rank 1 = top story)
      "title": "string",           // full original headline
      "url": "string",             // canonical http(s) link
      "source": "string",          // e.g. "OpenAI News"
      "category": "string",        // per-story chip, e.g. "SECURITY", "HARDWARE", "BUSINESS"
      "tags": ["string"],          // array of strings
      "summary": "string",         // plain text, 1-3 sentences
      "publishedAt": "string",     // original publish timestamp

      // Video card fields. ON-SCREEN text (read) and VOICEOVER (heard) are
      // DIFFERENT content by design — the voiceover never recites the on-screen lines:
      "headline": "string",        // on-screen card title, <= 40 chars
      "beats": ["string","string"],            // on-screen: EXACTLY 2 facts, each <= 42 chars
      "takeaway": "string",        // on-screen "why it matters", <= 70 chars
      "voiceover": "string",       // SPOKEN narration, 2-3 sentences (~35-55 words), distinct from on-screen

      // v4 — depth + background sources:
      "explainer": "string",       // on-screen CONTEXT the viewer reads, 2-3 sentences (~35-75 words)
      "entities": ["string"],      // canonical company/product names for a REAL logo lookup; [] if none
      "image_keywords": ["string"],// 2-4 concrete visual nouns for a fallback image (no brands/text); [] if none

      // v5 — brand-arrival stinger (used when the story has no real logo):
      "bg_word": "string",         // short subject to show, 1-3 words (~18 chars), Title Case, non-empty
      "bg_tone": "string",         // one of: "launch" | "hot" | "money" | "shock" | "bad" | "neutral"

      // v6 — what/why split for the video:
      "so_what": "string",         // ONE-sentence implication (<=25 words), the "so what" — does NOT restate the event
      "detail": "string"           // one concrete anchor fact (<=24 chars): a number, version, price or proper name
    }
    // ... 5 stories total
  ]
}

Fetch example

const res = await fetch("https://solvx.uk/api/ai-news.json");
const { date, stories } = await res.json();
// stories[] is rank-ordered 1 to 5; map each onto a countdown card with no transform

Field guarantees

On every story: title, url, rank, headline, takeaway, voiceover and explainer are non-empty; beats is an array of EXACTLY 2 non-empty items; entities and image_keywords are always arrays (possibly empty []); all five stories share an identical key set. headline, beats and takeaway are the SHORT on-screen text; explainer is the longer on-screen CONTEXT the viewer reads to understand the story (distinct from the beats); and voiceover is flowing SPOKEN narration written for a different channel — it adds context and does NOT duplicate the on-screen lines. entities holds canonical company/product names for a logo lookup; image_keywords holds concrete visual nouns (never brand names or text) for a fallback background image. bg_word (non-empty, 1-3 words) and bg_tone drive the per-story "brand arrival" stinger shown when there is no real logo — bg_tone is always exactly one of launch, hot, money, shock, bad or neutral. so_what is a non-empty ONE-sentence implication (≤ 25 words) — the "why it matters", written to NOT restate the headline/beats/takeaway/explainer/summary (an overlap check swaps in a consequence line if it would); detail is one concrete anchor fact (≤ 24 chars — a number, version, price or proper name) taken from the story. The card fields and summary are plain text — HTML entities decoded, no markdown, no newlines — and length-capped (truncated at a word boundary, never overflowed) so they drop straight into a fixed-size card.