Skip to content

Loading PrismWriting

Preparing your experience...

Stay informed. Stay independent.

Get research highlights, new articles, and community updates delivered to your inbox. No algorithms, no tracking—just quality content.

PrismWriting

Examining the systems that shape our world—history, politics, law, economics—while cultivating practical skills for self-sufficient living.

Ariel@prismwriting.com
Arizona, United States

Content

  • Blog
  • Content Library
  • Vlog
  • Knowledge Graph
  • Podcast
  • Education
  • Resources
  • Creators
  • Talent Board
  • Newsroom

Community

  • Round Table
  • Governance
  • Newsletter
  • Support Us
  • Contact

About & Legal

  • About
  • How It Works
  • Editorial Standards
  • Algorithm
  • Transparency
  • Corrections
  • Content License
  • Privacy Policy
  • Cookie Policy
  • Terms of Service
  • Accessibility

Independent, reader-supported research with no corporate backing.

Support Our Work
© 2026 Prism Writing LLC. All rights reserved.
Press?for shortcuts
Bot-friendly•Sitemap•Robots.txt
Made with ♥ and curiosity

Algorithm Transparency

How we discover, recommend, and rank content — and why our approach is different.

Last updated: February 7, 2026 · We publish all changes to our algorithm on this page.

Our Core Philosophy

Most platforms optimize for engagement — keeping you scrolling, clicking, and watching as long as possible. We optimize for understanding.

Our algorithm is designed with three commitments:

  1. Quality over virality — Verified, well-sourced content ranks higher than clickbait
  2. Small creator boost — New voices get algorithmic advantages, not penalties
  3. No dark patterns — No infinite scroll, no autoplay, no manipulation

How Content Discovery Works

Trending

Engagement velocity × Recency

Content with growing readership in the last 7 days. Newer content gets a recency boost that decays over 90 days.

trend_score = sessions × (1 + avg_engagement) × recency_boost

Popular

View count + Likes

All-time most-viewed content. Simple and transparent — the pieces that resonated most with readers.

rank = view_count × 1.0 + like_count × 0.5

Quality

Multi-factor score (0-100)

Content scored on 5 dimensions: engagement quality, verification status, creator reputation, freshness, and completeness.

quality = engagement×0.25 + verification×0.30 + creator×0.20 + freshness×0.10 + completeness×0.15

Rising

New content + Early engagement

Published in the last 14 days with strong early readership. Designed to surface promising new content.

rising = engagement_score (published < 14 days)

Recommendations

Category + Tags + Title similarity

Related articles shown at the end of each piece. Uses category match (40pts), shared tags (up to 30pts), same author (20pts), and keyword overlap (10pts).

score = category(40) + tags(30) + author(20) + keywords(10) + depth(10)

Search

TF-IDF fuzzy matching

Full-text fuzzy search across titles (2x weight), excerpts (1.5x), tags (1.3x), categories (1.2x), and body content (0.5x).

Fuse.js with threshold=0.4, ignoreLocation=true

Small Creator Boost

We believe great journalism shouldn't require a large following. Creators with fewer than 100 followers receive a 1.5× discovery boost in quality rankings.

Follower CountDiscovery BoostEffect
0 – 991.5×Quality score multiplied by 1.5 in discovery rankings
100 – 9991.2×Moderate boost for growing creators
1,000+1.0×No boost — established creators don't need algorithmic help

Quality Score Breakdown

Verification Status

30%

Has the content been fact-checked? Verified content with cited sources scores highest.

  • VERIFIED (100)
  • FACT_CHECKED (80)
  • NEEDS_REVIEW (40)
  • UNVERIFIED (20)

Engagement Quality

25%

Not just clicks — we measure completion rate, scroll depth, and time spent reading.

  • High completion + deep scroll (90-100)
  • Moderate engagement (50-70)
  • Low engagement / bounced (10-30)

Creator Reputation

20%

Based on the creator's track record of quality content, verification rates, and community feedback.

  • Consistently verified content (80-100)
  • Mixed track record (40-60)
  • New creator (default 50)

Completeness

15%

Does the article have an excerpt, featured image, proper word count, cited sources?

  • Full metadata + sources (100)
  • Partial (50-80)
  • Minimal (20-40)

Freshness

10%

Recently updated or published content gets a small boost. This prevents evergreen content from being permanently buried.

  • Published this week (100)
  • This month (70)
  • This year (40)
  • Older (20)

What We Don't Do

✕ No personalization bubbles

Everyone sees the same content rankings. No filter bubbles, no echo chambers.

✕ No user tracking

We don't track individual reading behavior for recommendations. No cookies for personalization.

✕ No pay-to-rank

No content gets boosted because someone paid. No sponsored placements in discovery feeds.

✕ No engagement bait

Click-through rate doesn't influence rankings. Clickbait titles don't get rewarded.

✕ No infinite scroll

Content is paginated. We don't trap you in an endless feed.

✕ No autoplay

Videos never autoplay. You choose what to watch.

Algorithm Changelog

February 7, 2026v2.0
  • +Added engagement-metric-based trending (sessions × engagement × recency)
  • +Added "Rising" section for new content with early momentum
  • +Added small creator boost (1.5× for <100 followers) to quality rankings
  • +Added category filtering to all discovery endpoints
  • +Published algorithm transparency page
February 4, 2026v1.0
  • +Initial quality scoring algorithm (5 weighted factors)
  • +Smart article recommendations (category + tags + author + keywords)
  • +Trending by engagement score
  • +Popular by view count
  • +Full-text fuzzy search with Fuse.js

Open Questions

We're actively discussing these topics with our community. If you have thoughts, use the feedback widget or visit our governance page.

  • Should we add opt-in personalization? (e.g., "show me more History articles")
  • How should we handle content in multiple languages?
  • Should community votes influence quality scores?