Turn Dageno prompt opportunities into a fanout backlog, then turn one selected backlog row into an editorial brief, a draft contract, a review contract, and publishable GEO content.

Positioning

GEO Content Writer is a backlog-row-first content production system for teams that want more than one-shot article generation.

It is designed for a practical workflow:

• find real prompt opportunities in Dageno
• extract real fanout instead of guessing article ideas
• organize those ideas into a reusable backlog
• choose one row to write next
• generate an editorial package that external agents can actually execute

This project is built to answer a practical growth question:

If I already have Dageno data, how do I turn it into a repeatable article-production workflow without producing thin, repetitive, template-like content?

Outcome

Instead of jumping from prompt data straight into article text, this project creates a production layer in between:

• backlog row
• editorial brief
• draft package
• review package
• final gate before publish

That makes it better suited for:

• AI agents
• human editors
• multi-article content calendars
• teams that care about repeatability, differentiation, and QA

Best For

• GEO and SEO teams turning Dageno data into real content workflows
• agencies that need a repeatable content operating system, not isolated prompts
• SaaS, ecommerce, industrial, and B2B brands building article pipelines from AI-answer opportunity data
• operators who want to generate briefs and review contracts before writing
• teams that want external agents to write section by section instead of improvising from one giant prompt

Why It Feels Different

Most AI content workflows start too late.

They jump from:

• keyword or prompt
• into article generation

That usually creates:

• topic-label articles that do not sound like real buyer language
• repeated listicles and comparisons
• prompt-shaped content instead of editorially chosen angles
• weak QA because the system never made the writing object explicit

This project starts earlier and gets more specific before writing begins:

• earlier with Dageno opportunity discovery
• more specific with real fanout extraction
• more controlled with backlog rows and cluster roles
• more operational with draft contracts, review contracts, and final gates

What You Get

• a reusable fanout backlog built from real Dageno data
• cluster-role planning before article generation
• an editorial_brief with audience, angle, differentiation targets, and E-E-A-T guidance
• a draft_package with section-by-section writing contracts
• a review_package with section review, assembly review, and final-gate checks
• publish-ready markdown or WordPress handoff

Output Quality Contract

To keep output quality stable at decision-grade level (not skeleton SEO content), publish-ready drafts should always include:

• explicit exclusion boundaries per major option (not ideal when ...)
• a forced default ranking fallback
• head-to-head competitive calls (same scenario, same inputs)
• an If X -> Choose Y decision engine
• a one-line convergence summary (If You Only Remember One Thing)
• at least 5 references including both editorial and official support/policy sources

Start With These Commands

PYTHONPATH=src python -m geo_content_writer.cli build-fanout-backlog --days 7 --max-prompts 10

Optional low-inventory fallback mode:

PYTHONPATH=src python -m geo_content_writer.cli build-fanout-backlog \
  --days 7 \
  --max-prompts 100 \
  --allow-exploratory-fallback \
  --exploratory-min-write-now 8 \
  --exploratory-max-items 30

PYTHONPATH=src python -m geo_content_writer.cli select-backlog-items --top-n 10

PYTHONPATH=src python -m geo_content_writer.cli publish-ready-article --backlog-id <row-id>

PYTHONPATH=src python -m geo_content_writer.cli draft-article-from-payload examples/publish-ready-payload.json

External Access And Minimum Credentials

This project can use several external layers:

• Dageno API for prompts, fanout, citations, and opportunity discovery
• optional citation crawling for structure learning
• optional web research for E-E-A-T evidence and comparison checks
• optional WordPress publishing

Recommended minimum setup:

• DAGENO_API_KEY: required
• local brand knowledge base: strongly recommended
• citation crawling access: optional
• WordPress credentials: optional

Access policy:

• real Dageno fanout is required for the main workflow
• guessed fanout should not be used as the production seed
• citation crawling is helpful but not required
• live external research can be handled by a downstream agent using the payload’s external_research_tasks
• when write-now inventory is low, exploratory fallback can be enabled; fallback rows are tagged status=exploratory and are not publish-ready by default

What This Project Produces

For one selected backlog row, the system can produce:

• a structured writing seed
• a differentiated editorial brief
• section drafting instructions
• section review instructions
• assembly review guidance
• final-gate checks before publishing
• markdown suitable for publishing or editor handoff

Workflow

flowchart LR
    A["[Dageno](https://dageno.ai/?utm_source=github&utm_medium=social&utm_campaign=official) Prompt Opportunities"] --> B["Real Fanout Extraction"]
    B --> C["Fanout Backlog"]
    C --> D["Cluster Role Planning"]
    D --> E["Editorial Brief"]
    E --> F["Draft Package"]
    F --> G["Review Package"]
    G --> H["Final Gate"]
    H --> I["Publish / Handoff"]

Skill Logic (Input -> Output)

flowchart TD
    IN["Inputs: DAGENO_API_KEY + date window + optional backlog_id + optional brand KB"] --> OPP["Opportunity Discovery"]
    OPP --> FAN["Real Fanout Extraction"]
    FAN --> BL["Prioritized Backlog"]
    BL --> SEL["Select One Backlog Row"]
    SEL --> BRIEF["Editorial Brief + Draft/Review Contracts"]
    BRIEF --> DRAFT["Draft Article Generation"]
    DRAFT --> GATE["Quality Contract + Final Gate"]
    GATE --> OUT["Outputs: payload JSON + publish-ready markdown + optional WordPress draft"]

Input Surface

Output Surface

Command To Output Mapping

Core Workflow

A. Opportunity Layer

1. discover high-value prompts
2. extract real fanout for each prompt
3. save all fanout into one backlog

B. Backlog Layer

4. mark overlap / merge / duplicate items
5. keep one prioritized backlog with statuses
6. assign a cluster role to each row
7. choose which backlog row to write next

C. Writing Layer

8. crawl top citation pages for the selected fanout
9. analyze citation patterns
10. build one editorial brief from one selected backlog row
11. generate section-by-section draft instructions
12. generate section-by-section review instructions
13. assemble one publish-ready article

D. Distribution Layer

14. publish to WordPress draft or publish status

What Makes The Production Object Different

The main production object is no longer a loose prompt.

It is a machine-readable payload designed for agent execution:

• backlog_row
• selected_fanout
• editorial_brief
• draft_package
• review_package
• writer_prompt

That gives agents and editors a clearer workflow:

• what to write
• why this angle exists
• what nearby articles to avoid overlapping with
• what evidence is still needed
• what QA checks need to pass before publishing

Official Path

The recommended production path is:

1. build-fanout-backlog
2. select-backlog-items
3. publish-ready-article --backlog-id <row-id>
4. draft-article-from-payload
5. run section reviews from review_package.section_review_contract
6. run assembly review from review_package.assembly_review_prompt
7. clear the final gate in review_package.final_gate
8. publish-wordpress

Commands still present for compatibility but no longer recommended as the main entrypoint:

• legacy-publish-ready-article
• content-pack
• first-asset-draft

Why Teams Use It

Typical AI Content Workflow

• prompt chosen ad hoc
• article generated too early
• little differentiation between related pages
• weak evidence and no scoped QA

With GEO Content Writer

• real fanout becomes backlog
• backlog rows become the article production unit
• cluster roles reduce content collisions
• external agents get a structured brief instead of one vague prompt
• section review and final-gate checks improve consistency before publishing

Citation Learning Policy

• prefer article-like pages first
• ignore app-store, forum, and similar non-article pages for primary structure learning
• if article-like pages are fewer than 3, switch to article_first_fallback
• in fallback mode, keep article pages as the primary learning source and use support pages only as secondary context

Non-Negotiable Rules

• only use real Dageno fanout
• do not generate guessed fanout for production writing
• exploratory fallback candidates are allowed only as status=exploratory and must be validated against fresh GEO data before promotion
• do not write directly from Dageno topic labels
• do not publish from prompt alone
• one selected fanout should map to one article
• if brand knowledge base and Dageno brand snapshot do not match, block publish-ready generation
  • You can explicitly pass --allow-brand-mismatch to override, but it will emit a warning and is not recommended as the default path.

Quick Start

1. Discover prompt candidates

PYTHONPATH=src python -m geo_content_writer.cli discover-prompts --days 7 --max-prompts 20

2. Build the fanout backlog

PYTHONPATH=src python -m geo_content_writer.cli build-fanout-backlog --days 7 --max-prompts 20

Default backlog file:

knowledge/backlog/fanout-backlog.json

Suggested backlog statuses:

• write_now
• needs_merge
• needs_cleanup
• skip

3. Select the next backlog items

PYTHONPATH=src python -m geo_content_writer.cli select-backlog-items --top-n 10

4. Generate one backlog-row-first article payload

PYTHONPATH=src python -m geo_content_writer.cli publish-ready-article \
  --backlog-file knowledge/backlog/fanout-backlog.json \
  --backlog-id your-backlog-row-id \
  --output-file examples/publish-ready-payload.json

This outputs a structured payload with:

• editorial_brief
• draft_package
• review_package
• writer_prompt

If you do not pass --backlog-id, the CLI will fall back to the top write_now row.

5. Draft an article from the payload

PYTHONPATH=src python -m geo_content_writer.cli draft-article-from-payload \
  examples/publish-ready-payload.json \
  --output-file examples/publish-ready-article.md

6. Publish to WordPress

export WORDPRESS_SITE_URL="https://your-site.com"
export WORDPRESS_USERNAME="your-username"
export WORDPRESS_APP_PASSWORD="your-application-password"
PYTHONPATH=src python -m geo_content_writer.cli publish-wordpress examples/publish-ready-article.md --status draft

For wordpress.com hosted sites, also set:

export WORDPRESS_CLIENT_ID="your-client-id"
export WORDPRESS_CLIENT_SECRET="your-client-secret"

7. Run One-Click Article Quality Gate

PYTHONPATH=src python -m geo_content_writer.cli check-article-quality \
  examples/publish-ready-article.md \
  --min-words 1200

Optional JSON report for CI:

PYTHONPATH=src python -m geo_content_writer.cli check-article-quality \
  examples/publish-ready-article.md \
  --min-words 1200 \
  --json

Payload Shape

The primary production payload is a machine-readable object for external agents:

• backlog_row: the selected production unit
• selected_fanout: normalized writing seed
• editorial_brief: audience, angle, differentiation targets, adjacent rows to avoid, evidence guidance, and E-E-A-T layer
• draft_package: target word counts, draft_sections, and assembly notes
• review_package: final review prompts plus section_review_contract
• writer_prompt: a convenience prompt derived from the structured payload

See:

• schemas/article_generation_payload_schema.json
• examples/publish-ready-payload-trip.json

Cluster Roles

Each backlog row now carries a cluster_role to make the content calendar more deliberate before writing begins. Examples:

• category_article
• buyer_shortlist_article
• decision_stage_comparison_article
• workflow_guidance_article
• fit_assessment_article

The goal is to stop adjacent rows from becoming near-duplicates that differ only in title wording.

Benchmarks

A lightweight benchmark suite now lives in:

• examples/benchmarks/README.md
• examples/benchmarks/benchmark_manifest.json

It uses real examples across multiple content roles to evaluate:

• distinctness
• naturalness
• decision support
• brand fit
• cluster role clarity

Repo Structure

geo-content-writer/
├── README.md
├── LICENSE
├── manifest.json
├── agents/
│   └── openai.yaml
├── skills/
│   └── content-writer.md
├── knowledge/
│   ├── brand/
│   │   └── brand-knowledge-base.json
│   └── backlog/
├── schemas/
├── references/
├── examples/
└── src/

Technical Notes

• Dageno remains the opportunity discovery layer
• the backlog row is the core production object
• fanout remains the writing seed, but only after backlog selection
• citation crawl is still lightweight and not yet a full browser-rendered implementation
• publish-ready-article is the main backlog-row-first payload builder
• the main writing interface is designed for external agents that can draft and review section by section
• WordPress publishing is a lightweight distribution example, not the center of the system

License

MIT