Your journalism, structured for AI search, licensing, and agents.
GrafStack is a methodology and reference architecture for newsrooms. We help you organize stories into paragraph-level units with sentence-level precision, so AI systems cite, license, and correct your work.
The problem
AI answer engines use your reporting and send fewer readers back. Publishers face three pressures at once.
- Traffic loss. Zero-click AI answers replace search referrals.
- Attribution loss. AI systems strip bylines, sources, and dates from your facts.
- Trust loss. Readers struggle to tell verified reporting from synthetic text.
Most newsroom content lives in HTML pages built for browsers. AI systems read those pages as flat text. Your sourcing, fact checks, and corrections disappear along the way.
How GrafStack works
Keep your CMS.
GrafStack sits on top of the platform you run today: Sanity, Contentstack, Arc XP, Brightspot, WordPress VIP, and others.
Grafs as the working unit.
Editors write and edit in paragraphs. GrafStack gives each graf an identity, a role (reported fact, quote, analysis, context), and rights metadata.
Sentences as the precision layer.
Quotes, statistics, and claims about named people get sentence-level attribution, entity links, and correction history.
Media as first-class units.
Images, charts, datasets, video segments, and embeds each carry credit, rights, and provenance. Captions and cited data points link to the sentences using them.
AI proposes, editors decide.
An LLM proposes tags. Editors review the high-risk ones: quotes, numbers, attributions, and claims about people. Low-risk tag types move to auto-accept once editors approve them consistently.
Corrections at the right scale.
| Change | What GrafStack does |
|---|---|
| Typo or style fix | Edit in place and log the change. |
| Fact change in one sentence | New sentence version. |
| Change across sentences or in framing | New graf version. |
| Change to headline, lede, or central claim | New story version plus a public correction note. |
Every fact correction triggers a recheck of related sentences, captions, charts, and stories. Licensees and AI partners holding the old version receive notice through a correction feed.
The seven-layer stack
GrafStack maps your newsroom across seven layers, from authoring to agent access.
Content structure. How your CMS models and stores stories.
Provenance and attribution. Content credentials, watermarks, and citation tracking.
Rights expression. robots.txt, RSL, and machine-readable license terms.
Transport and monetization. Pay-per-crawl, marketplaces, and content protocols.
Direct deals. Bilateral licensing with AI companies.
Legal and collective action. Litigation, coalitions, and regulation.
Discovery and agent access. MCP servers, answer engine visibility, and agent protocols.
Most newsrooms work on one or two layers. GrafStack shows how the layers connect and where your gaps cost you.
The AI Editorial Workflow Audit
Four weeks. Fixed fee. A clear roadmap at the end.
Price$22,000 to $28,000based on scope.
- Crawl and score of your public content against the GrafStack rubric
- Review of your CMS content model and editorial workflow
- Assessment of AI bot traffic, robots.txt rules, and rights declarations
- Gap map across all seven layers
- Prioritized roadmap with effort estimates
Who runs GrafStack
GrafStack comes from Bill Thornton, founder of Bite Size Elephant. Bill brings 20+ years in digital media and content technology, including VP-level product leadership at Dow Jones. His work focuses on editorial platform strategy, CMS modernization, and AI workflow integration.
Related work
Sannuta Raghu's Directory of Liquid Content (2025) and News Atom (Reuters Institute for the Study of Journalism, October 2025) define a vocabulary and a sentence-level metadata schema for modular journalism. GrafStack is an independent methodology for building structured content into production CMS platforms and connecting the result to licensing and AI access. The efforts share goals and differ in scope.