ACRA Insight · Manifesto · May 2026
Agentic Content
Marketing
The content function as software.
A 7-beat manifesto by Bret Kerr.
- Agentic Content Marketing
- A four-part discipline — Story Extraction, Content Architecture, Agentic Workflow, and Distribution Intelligence — that treats the content function as a software product. Operates via a Mixture of Expert Agents (MoEA) Loop governed by a human editorial membrane. Ships versioned content corpora optimized simultaneously for human readers, AI crawlers, and downstream distribution.
Day 01 · Flagship
The Content Function Is Becoming Curated Software.
Day 01 · Flagship
The Content Function Is Becoming Curated Software.
And the marketers who survive 2026 are the ones who can ship it.
Agentic Content Marketing treats the content function as a software product with a build pipeline, versioned corpus, and four primitives: Story Extraction, Content Architecture, Agentic Workflow, Distribution Intelligence. One operator plus a structured agent stack now produces what previously required a five-person team — at a fraction of the cost, with full architectural transparency.
For thirty years, content marketing has been a craft operation pretending to be an industrial one. We talked about “content engines” and “editorial calendars” the way carpenters talk about workshops. The metaphor has finally collapsed. The content function is not like software anymore. It is software. That sentence is the entire thesis.
Anthropic released the Model Context Protocol in November 2024 and donated it to the Linux Foundation in December 2025 alongside OpenAI and Block. Anthropic published the Agent Skills specification as an open standard on December 18, 2025, and within ninety days it was adopted by 32 tools including OpenAI Codex, Microsoft VS Code, Google Gemini CLI, JetBrains Junie, AWS Kiro, and Block Goose. Generative engines now ingest, decompose, and re-emit content at machine speed. A page is no longer a destination. It is a source corpus.
I have spent thirteen months proving this in production. I built an agentically orchestrated interactive book site on Anthropic’s first five years from 727 deep-research artifacts totalling ~1.4 million words. The retrieval layer runs locally on SQLite with sqlite-vec, FTS5, and reciprocal rank fusion at k=60. Embeddings are voyage-3-large. Every quoted passage passes through a substring-match verbatim verifier. Recurring inference cost: $0. The site is at contextjamming.com/book. That is not a content marketing project. That is a software product whose feature surface happens to be editorial.
The four primitives of Agentic Content Marketing
- 01Story Extraction. Deep research is a software primitive, not a meeting.
- 02Content Architecture. Pages decomposed into reusable, extractable claims.
- 03Agentic Workflow. A mixture of expert agents replaces the linear writer/editor/SEO chain.
- 04Distribution Intelligence. Content shaped at write-time to be extracted by ChatGPT, Perplexity, Claude, and Google AI Overviews.
Day 02 · GEO / AI-Discoverability
Your Content Has Two Audiences Now. One of Them Is a Crawler.
Day 02 · GEO / AI-Discoverability
Your Content Has Two Audiences Now. One of Them Is a Crawler.
What the Princeton GEO paper actually says, and what it means for your roadmap.
The Princeton GEO paper (Aggarwal et al., KDD 2024) provides the only peer-reviewed academic foundation under the GEO industry. It shows up to 40% visibility lift from citation-dense content, 30–40% improvement on Position-Adjusted Word Count, and 115.1% lift for rank-5 pages. Meanwhile ChatGPT referrals can convert at 15.9% vs. 1.76% for Google Organic.
In November 2023, six researchers from Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi released “GEO: Generative Engine Optimization” (Aggarwal et al., ACM SIGKDD 2024). It is the only peer-reviewed academic foundation under the entire GEO industry. Three findings should be tattooed on every content director’s eyelids:
“GEO can boost visibility by up to 40% in generative engine responses.”
— Aggarwal et al., Abstract, KDD 2024
“115.1% increase in visibility for websites ranked fifth in SERP.”
— Aggarwal et al., Section 5.2, KDD 2024; rank-1 sources decreased by 30.3% in the same table.
Translation: if you are not the dominant brand in your category, GEO is a structural opportunity to leapfrog the dominant brand in the only ranking surface that matters in 2026.
ChatGPT referral traffic: 15.9% conversion rate — Seer Interactive B2B case study, 2024–2025, 2025
GPTBot share rose from 5% → 30% of AI crawler traffic, May 2024–May 2025 — João Tomé, Cloudflare Blog, July 1, 2025
~65% of AI Mode-cited pages and 71% of ChatGPT-cited pages include structured data — SE Ranking structured-data dataset, 2026
The five tactics that move the needle (Princeton-validated)
- 01Statistics density with named primary sources. "76% of B2B SaaS teams cite onboarding as their top churn driver (ChartMogul 2025)" beats "most teams struggle with onboarding."
- 02Quotation Addition. Direct quotes from named experts, attributed inline.
- 03Cite Sources. Inline links to credible third-party sources, not your own product pages.
- 04Entity clarity. Organization, Person, Article, FAQPage, and BreadcrumbList JSON-LD. ~71% of pages cited by ChatGPT include structured data.
- 05Extractable answer capsules. 40–60 words under each H2, written to make complete sense in isolation — the "information island" test.
Infrastructure move: serve a clean llms.txt at root and per-page .md mirrors. Per SE Ranking’s November 2025 study of nearly 300,000 domains, only 10.13% had an llms.txt file in place. Anthropic, Stripe, Cloudflare, Vercel, and Perplexity all serve it. So does contextjamming.com.
Day 03 · Agentic Content Infrastructure
Content as Code: The Build Pipeline Replaces the Editorial Calendar.
Day 03 · Agentic Content Infrastructure
Content as Code: The Build Pipeline Replaces the Editorial Calendar.
Why the next ten years of content will be written in TypeScript.
The editorial calendar is dead. It has been replaced by a deterministic build pipeline: ingest sources → embed and index → retrieve and synthesize → generate multi-surface outputs → verify → deploy → observe. Content as code has four primitives: corpus, schema, pipeline, observability layer.
For a decade, content ops meant Notion + CMS + Friday standup. Linear: brief → draft → edit → publish → promote. In 2026, that chain is replaced by a pipeline that looks like software engineering. The vocabulary has changed: “editorial calendar” → “content model”; CMS → content operating system; writer → build orchestrator.
The contextjamming.com book stack
| Layer | Technology | Detail |
|---|---|---|
| Embeddings | voyage-3-large | 1024-dim + rerank-2 |
| Vector DB | SQLite + sqlite-vec | KNN vec0 + FTS5 BM25 |
| Retrieval | RRF k=60 | Reciprocal rank fusion |
| Synthesis | Claude Sonnet 4.6 | Agentic tool-use loop |
| Verifier | Substring-match | Verbatim quote guard |
| Build | TypeScript / tsx | unified · remark · rehype |
| Audio | Kokoro TTS | Apache 2.0 · 82M params · $0 |
| Deploy | Vercel + Next.js 15 | ~190MB CDN edge |
Final tally: 12 chapters, 1h 41m of narration, ~190MB deployed, $0 recurring inference cost. That is not content marketing. That is content as code.
Four primitives of an agentic content infrastructure
- 01A corpus, not a queue. The unit of production is a versioned corpus of indexed claims, not a publication queue.
- 02A schema, not a template. Content modeled as structured entities — Article, FAQPage, Person, HowTo — not fillable templates.
- 03A pipeline, not a workflow. Deterministic, versioned, observable. Runs on a trigger, not a calendar.
- 04An observability layer, not a quarterly report. AI Citation Rate, Response Inclusion Rate, AI-referred conversion — measured weekly, not annually.
Day 04 · MoEA Loop / Triple Transformation
The MoEA Loop: A Mixture of Expert Agents Is the New Editorial Desk.
Day 04 · MoEA Loop / Triple Transformation
The MoEA Loop: A Mixture of Expert Agents Is the New Editorial Desk.
How four roles — orchestrator, auditor, adversary, editorial membrane — replace the writer/editor/SEO chain.
The MoEA Loop assigns four distinct epistemic roles: Claude Sonnet 4.6 as build orchestrator; Claude Opus 4.7 as auditor in a separate context window; Gemini Deep Research / 3.1Pro as adversarial agent; and the human as editorial membrane — the only role that can authorize a ship. The value is not in the agents. It is in the loop.
Every senior content JD I’ve seen in 2025–2026 has the same hidden problem: hiring managers want the output of a five-person team for one person plus AI tools. The honest answer is not “use ChatGPT to write faster.” It is a production methodology that turns one operator into a four-agent newsroom.
The four roles
Build Orchestrator
Claude Sonnet 4.6 / Opus 4.7
Runs the pipeline. Reads sources. Generates drafts. Calls tools.
Auditor
Claude Opus 4.7
Separate context window. Reads cold. Hunts hallucinations, weak claims, broken JSON-LD.
Adversarial Agent
Gemini Deep Research / 3.1Pro
Stress-tests the argument. Runs adjacent searches. Surfaces counter-evidence.
Editorial Membrane
The Human
Owns voice, judgment, taste, and the final verbatim-verifier pass. Nothing ships without crossing the membrane.
The Triple Transformation pipeline
The MoEA Loop applied to one task: turning a research artifact into multi-surface published content in one pass.
- 01Stage 1 — Gemini Deep Research / 3.1Pro. gemini-deep-research-xml SKILL generates the XML brief; Gemini returns structured output with STT-ANCHOR blocks.
- 02Stage 2 — Claude Synthesis. semantic-triple-transformation SKILL emits three artifacts in one pass: Substack longform, Infographic JSON, X social thread.
- 03Stage 3 — Verbatim Verifier + HTML Production. responsive-html SKILL — every quoted passage normalized and substring-matched before publication.
All three SKILL.md files are open-sourced at contextjamming.com/book/skills under Apache 2.0, conformant to the Anthropic Agent Skills open standard (December 18, 2025), adopted by 32 tools as of March 2026.
Day 05 · Proof / Portfolio
727 Documents, 1.4 Million Words, One Agent: A Working Demo.
Day 05 · Proof / Portfolio
727 Documents, 1.4 Million Words, One Agent: A Working Demo.
The Anthropic book is not a marketing project. It is an industrial-scale demo.
Over thirteen months, the contextjamming.com book project ingested 727 deep-research documents, ~1.4 million words, indexed into 21,222 retrieval chunks, shipped as 12 chapters and 1h 41m of locally generated narration — at $0 recurring inference cost. Every quoted passage passes through a substring-match verbatim verifier before publication.
727 source documents · ~1.4M words · 21,222 retrieval chunks · $0 recurring inference cost — contextjamming.com/book/stack, 2025–2026
Hallucination is not a UX problem — it is an existential risk to an artifact whose subject is the company whose model is doing the writing. The verbatim verifier is not optional. It is the reason the project can exist at all.
The proof stack
- 01The Book. contextjamming.com/book — 727 sources, 1.4M words, 12 chapters, 21.2k retrieval chunks.
- 02The Skills. contextjamming.com/book/skills — Three open-sourced SKILL.md files under Apache 2.0.
- 03The Stack. contextjamming.com/book/stack — Full architectural disclosure.
- 04The Site. contextjamming.com — Daily dispatches. AI Wire ticker on 30-min ISR.
- 05The Substack. substack.com/@contextjamming26 — Field notes from inside the context window.
Before this, I spent approximately ten years at Mimecast as Video Production Director — direct creative partner to founder/CEO Peter Bauer, through their NASDAQ IPO and Permira acquisition. The craft was always there. The agentic infrastructure is what turned the craft into a content function that ships at software speed.
Day 06 · Counter-Positioning
What Agentic Content Marketing Is Not.
Day 06 · Counter-Positioning
What Agentic Content Marketing Is Not.
The five things people mean when they say “AI content,” and why none of them are this.
Agentic Content Marketing is not prompt engineering, ChatGPT for marketers, Agentic SEO, content automation, or human replacement. It sits one architectural layer above Agentic SEO — the entire content function rebuilt as software, with SEO as a downstream consumer, governed by a human editorial membrane.
NOT prompt engineering with extra steps.
Prompt engineering is single-shot. Agentic Content Marketing is a multi-agent, multi-context, version-controlled production pipeline. The Anthropic Agent Skills spec (December 18, 2025) made the distinction concrete: a skill is procedural knowledge that loads on demand from a structured folder. It is not a prompt. It is a deployable artifact.
NOT "ChatGPT for marketers."
The 2025 wave of AI writing tools (Jasper, Copy.ai, Writesonic, Surfer SEO) sat on top of generative APIs. That market has plateaued. The next wave is unified AI content operations platforms with structured content models, native MCP integration, and agent orchestration. Agentic Content Marketing is the methodology, not the toolset.
NOT Agentic SEO.
Kevin Indig has done excellent work positioning Agentic SEO — agents that write to the web, not just read it. In his Growth Memo essay "10 SEO, marketing, and tech predictions for 2026," Indig argues 2026 marks the beginning of the agentic era. He is right. But Agentic SEO stops at the SEO output. Agentic Content Marketing is one architectural layer higher: the entire content function rebuilt as a software system. SEO is a downstream consumer, not the system itself.
NOT content automation.
Automation is the workflow layer — Zapier, Make, Notion automations. That is operational hygiene. Agentic Content Marketing is the orchestration layer above the workflow — agents reasoning about what should be written, retrieved, synthesized, shipped. Workflow tools execute. Agents decide.
NOT human replacement.
The MoEA Loop has four roles; one is always human. The quality bar is perspective and provocation, not volume. Agents scale production. Humans hold taste. Remove the membrane, ship slop. Remove the agents, ship slow.
Nearly 2,000 global professionals surveyed; more than 40% of B2B deals stall from buying-group misalignment — 2025 Edelman–LinkedIn B2B Thought Leadership Impact Report, 2025
Day 07 · Conversion CTA
If You're Hiring a Content Engine, Hire the Engineer Who Already Built One.
Day 07 · Conversion CTA
If You're Hiring a Content Engine, Hire the Engineer Who Already Built One.
A direct note to heads of marketing at Apollo, Tulip, Formlabs — and everyone hiring against the same JD.
If your content JD uses phrases like “AI-powered GTM content operations,” “design content systems that scale,” or “content quarterback,” this 7-day manifesto and its proof artifacts are the long-form version of an application — from someone who has already shipped the demo.
What I’ve shipped at ACRA Insight (April 2025 – May 2026)
- →Industrial-scale agentic content infrastructure: 727 source documents, ~1.4M words, 21,222 retrieval chunks, $0 recurring inference cost. Live at contextjamming.com/book.
- →Three open-sourced SKILL.md files conformant to Anthropic Agent Skills, Apache 2.0. contextjamming.com/book/skills.
- →Documented MoEA Loop methodology and Triple Transformation pipeline.
- →Daily-cadence Substack and field-notes site at contextjamming.com.
Before that (Mimecast, ~2015–2025)
- →Video Production Director and creative executive for approximately a decade. Direct creative partner to founder/CEO Peter Bauer.
- →Built Mimecast's North American video production capability from scratch through NASDAQ IPO and Permira acquisition.
- →Translated highly technical cybersecurity narratives into customer-facing case studies, webinars, brand campaigns.
The thing nobody else in this hiring pool does: I have shipped the demo. The contextjamming.com book is not a portfolio piece. It is a working agentic content infrastructure. Your engineering team can clone the skills and audit the stack in an afternoon.
Proof Artifacts
The Infrastructure, Audited and Open
The Anthropic Book →
727 docs · ~1.4M words · 21,222 chunks · $0/mo
An industrial-scale interactive book on Anthropic's first five years, built from deep-research artifacts.
The Stack →
voyage-3-large · SQLite + sqlite-vec + FTS5 · RRF k=60 · Kokoro TTS
Full architectural disclosure. Clone and audit the retrieval layer, verifier, and build pipeline.
The Skills →
Apache 2.0 · Anthropic Agent Skills compliant
Three SKILL.md files open-sourced under the dominant agent standard, adopted by 32 tools.
Context Jamming Substack ↗
Field notes from inside the context window
The dispatch cadence — editorial output of the MoEA Loop in production.
The Site →
Daily dispatches · AI Wire ticker · 30-min ISR
Agentically produced. 19+ routes, full editorial design system, GEO-optimized throughout.
ACRA Insight Consulting →
Audit · Blueprint · Build
Engaged for audits, architecture blueprints, and full agentic content builds.
Case Study
How a Mid-Market B2B SaaS Rebuilt Its Content Function in Six Weeks
An illustrative ACRA Insight engagement scenario based on the contextjamming.com architecture, applied to a hypothetical Apollo-shaped customer.
Scenario
40-person B2B SaaS in the GTM sales intelligence category. 200+ existing blog posts. Zero presence in ChatGPT, Perplexity, or Google AI Overviews. Hiring against a "build a content engine" JD with no internal candidate who can deliver it.
Inputs
Full blog corpus export. JTBD interview transcripts with 6 customers. ICP definition doc. Existing SEO keyword map. Current CMS schema. Competitor citation analysis from Perplexity and ChatGPT.
Pipeline
- →Week 1 — Story Extraction. Gemini Deep Research / 3.1Pro over the existing corpus + JTBD interviews. Surfaces 40+ core claims, 12 primary sources, 8 high-intent questions with zero current coverage.
- →Week 2 — Content Architecture. Schema design (Article, FAQPage, Person, HowTo). Entity graph: Company → Product → Use Case → Customer Outcome. Claim library indexed in SQLite.
- →Weeks 3–4 — Agentic Workflow. MoEA Loop installed. SKILL.md files customized for client voice and compliance review. Triple Transformation producing Substack + infographic + X copy per claim.
- →Week 5 — Distribution Intelligence. llms.txt + per-page .md mirrors deployed. JSON-LD rolled out across all 200+ posts. robots.txt updated to allow AI crawlers explicitly.
- →Week 6 — Observability. Baseline AI Citation Rate measured. Queries: product category, use case, competitor comparison. Response Inclusion Rate and AI-referred conversion baseline set in analytics.
Projected Impact (Illustrative)
Illustrative projections only — not guaranteed outcomes. Based on published academic and industry benchmarks.
AI citation visibility lift
Aggarwal et al., KDD 2024 (Princeton GEO paper)
Up to 40%
Visibility lift for previously rank-5 pages
Aggarwal et al., Section 5.2
Up to 115.1%
AI referral conversion rate vs. Google Organic
Seer Interactive B2B case study, 2025 (single client — directional)
~15.9% vs. ~1.76%
Domains with llms.txt competitive advantage
SE Ranking study of 300K domains, Nov 2025 — only 10.13% have llms.txt
~89.9%
For Hiring Managers
The JD You’re Hiring Against
If you are hiring against any of these roles, this manifesto and the proof artifacts behind it are the long-form version of an application from Bret Kerr.
Apollo.io
Senior Integrated Content Strategist
"Hands-on, individual contributor role for a senior builder who can set bold strategy and also create content that directly drives growth … pioneering an AI-powered model for modern GTM content operations." 8–10 years required.
How ACRA Insight Maps
- →contextjamming.com/book — the shipped infrastructure.
- →contextjamming.com/book/skills — open-sourced SKILL.md methodology.
Tulip Interfaces
Content Marketing Architect for AI-Driven Growth
"Design content systems that scale … automate tagging, QA, and refresh cycles … ensure Tulip's content is structured and discoverable across Google AI Overviews, generative engines, and traditional search engines." Somerville, MA hybrid. $100–130K.
How ACRA Insight Maps
- →GEO architecture (Day 2 above) — the proven structured-data and llms.txt playbook.
- →contextjamming.com/book/stack — the content model and observability layer.
Formlabs
Technical Content Marketing Manager
"Content 'quarterback' who can translate complex hardware innovations into compelling narratives that drive lead generation and brand authority."
How ACRA Insight Maps
- →Ten years at Mimecast translating technical cybersecurity to customer narrative.
- →Triple Transformation pipeline — complex → structured → multi-surface in one pass.
Forty-five minute mapping call. First hour is on me.
bret@acrainsight.comFrequently Asked
Common Questions
What is Agentic Content Marketing?↓
Agentic Content Marketing is a four-part discipline — Story Extraction, Content Architecture, Agentic Workflow, and Distribution Intelligence — that treats the content function as a software product. It uses a Mixture of Expert Agents (MoEA Loop) governed by a human editorial membrane to produce versioned content corpora optimized simultaneously for human readers, AI crawlers, and downstream distribution.
How is it different from Agentic SEO?↓
Agentic SEO (Kevin Indig's framing) focuses on shipping content outputs to the web — agents that write to search. Agentic Content Marketing sits one architectural layer higher: it rebuilds the entire content function as a software system, with SEO as a downstream consumer, not the system itself.
What is the MoEA Loop?↓
The MoEA Loop (Mixture of Expert Agents) assigns four distinct epistemic roles: Claude Sonnet 4.6 as build orchestrator; Claude Opus 4.7 as auditor in a separate context window; Gemini Deep Research / 3.1Pro as adversarial agent; and the human as editorial membrane. The value is not in the agents — it is in the loop. Four distinct contexts, each with a different epistemic role, governed by a human with authority to kill output.
What is the Triple Transformation pipeline?↓
The Triple Transformation pipeline applies the MoEA Loop to a single task: turning a research artifact into multi-surface published content in one pass. Stage 1: Gemini Deep Research / 3.1Pro via gemini-deep-research-xml SKILL. Stage 2: Claude synthesis via semantic-triple-transformation SKILL — emits Substack longform + infographic JSON + X social copy simultaneously. Stage 3: Verbatim verifier + HTML production via responsive-html SKILL.
What does the Princeton GEO paper actually say?↓
Aggarwal et al. "GEO: Generative Engine Optimization" (KDD 2024) found: (1) GEO tactics boost AI citation visibility by up to 40%. (2) Cite Sources, Quotation Addition, and Statistics Addition achieved 30–40% PAWC improvement and 15–30% Subjective Impression lift. (3) The Cite Sources method produced 115.1% visibility increase for rank-5 pages, while top-ranked pages lost 30.3% of citations. Full paper: arxiv.org/abs/2311.09735.
What is an llms.txt file and does my site need one?↓
llms.txt is a clean markdown file served at your domain root that gives AI crawlers a curated overview of your site — analogous to robots.txt but for LLM retrieval pipelines. Anthropic, Stripe, Cloudflare, Vercel, and Perplexity all serve it. Per SE Ranking's November 2025 study of nearly 300,000 domains, only 10.13% had one. If your content strategy depends on AI discoverability in 2026, you need it.
Who is Bret Kerr / ACRA Insight?↓
Bret Kerr is the founder of ACRA Insight LLC, an applied AI consulting practice operating contextjamming.com. He spent approximately ten years at Mimecast as Video Production Director, direct creative partner to founder/CEO Peter Bauer through their NASDAQ IPO and Permira acquisition. At ACRA Insight, he built an agentically orchestrated book site from 727 deep-research artifacts (~1.4M words) at $0 recurring inference cost, open-sourced three SKILL.md files under Apache 2.0 conformant to the Anthropic Agent Skills standard.
How do hiring managers evaluate an agentic content marketer?↓
The signal is shipped proof, not credentials or credentials-adjacent AI tool lists. The right questions: Does the candidate have a live agentic content infrastructure that can be audited? Can they explain the retrieval layer, the verifier, the build pipeline? Do they have an open-sourced methodology? Do they measure AI Citation Rate and Response Inclusion Rate — or just organic impressions? The architecture is the interview.