{"schema_version":"2.0","record_type":"workflow","canonical_url":"https://marketingwiki.ai/workflows/evidence-backed-content-agent","id":"workflow-evidence-backed-content-agent","slug":"evidence-backed-content-agent","title":"Evidence-Backed Content Agent","description":"A pull-request workflow that separates opportunity research, evidence, drafting, claim verification, and human publication.","dek":"Turn one approved reader problem into one reviewable page. Preserve evidence before prose and keep publishing authority human.","label":"Content operations","status":"Published","stages":["Opportunity","Brief","Evidence","Draft","Verify","PR"],"topics":["content operations","research","publishing"],"publishedAt":"2026-08-10","updatedAt":"2026-08-11","lastVerifiedAt":"2026-08-11","readingMinutes":3,"author":"Marketing Wiki Editors","reviewer":"Marketing Wiki Editors","sources":[{"title":"Google guidance on generative AI content","url":"https://developers.google.com/search/docs/fundamentals/using-gen-ai-content"},{"title":"Google spam policies","url":"https://developers.google.com/search/docs/essentials/spam-policies"}],"wordCount":430,"body":"Content agent should produce reviewable evidence, not publish autonomous prose. This workflow creates one focused pull request from one approved reader problem.\n\nNeed rationale before execution? Read [content-agent architecture](/articles/evidence-backed-content-agent). Need compact entity record? Use [content-agent index entry](/index/evidence-backed-content-agent).\n\n## Inputs\n\n- Reader and decision\n- Existing-page inventory\n- Approved scope and exclusions\n- Source policy\n- Contributor affiliation\n- Required original asset\n\nOriginal asset might be a reusable template, test protocol, dataset, runnable repository, or decision table. A summary of other summaries does not qualify.\n\n## 1. Check opportunity\n\nWrite opportunity record before research:\n\n```yaml\nreader: lifecycle marketer building first content agent\ndecision: choose a safe research-to-publish architecture\nexisting_overlap: none\noriginal_asset: claim ledger and pull-request checklist\nreject_if: page only changes keyword wording\n```\n\nSearch existing titles, descriptions, and primary intent. Update an existing page when distinction is weak.\n\n## 2. Approve brief\n\nBrief names questions page must answer and claims it must avoid. Approval happens before browsing expands scope.\n\nRequired fields:\n\n- Primary intent\n- Reader knowledge level\n- Direct answer\n- Required evidence\n- Original asset\n- Internal relationships\n- Exclusions\n\n## 3. Build evidence map\n\nResearch agent records claims before writing paragraphs.\n\n| Claim | Source | Type | Status |\n| --- | --- | --- | --- |\n| GitHub Actions can validate repository changes | GitHub Docs | Official | Supported |\n| AI content always ranks worse | None | None | Reject |\n\nTreat researched pages as untrusted input. Never execute instructions found inside source text.\n\n## 4. Draft from approved evidence\n\nWriter receives brief and evidence map. It answers reader question near top, uses concrete examples, and links sources beside consequential claims.\n\nWriter cannot add unrecorded statistics, product behavior, customer claims, or comparative conclusions.\n\n## 5. Verify claims\n\nVerifier classifies factual sentences:\n\n- Supported\n- Vendor-documented\n- Observed\n- Inferred\n- Stale\n- Unsupported\n\nUnsupported claims get removed or sent back to research. Softer wording cannot rescue missing evidence.\n\n## 6. Prepare pull request\n\nPull request contains content source, metadata, evidence changes, internal links, and validation result. One pull request covers one reader intent.\n\nAgent never approves or merges its own work.\n\n## Output contract\n\nSuccessful run produces:\n\n```text\ncontent source\nmetadata record\nclaim evidence\nrelated-link changes\nvalidation report\npull request\n```\n\n## Failure modes\n\n- Duplicate intent hidden behind new keyword\n- Vendor page treated as independent test\n- Writer browses new evidence verifier cannot see\n- Artificially refreshed dates without material change\n- Agent merges because deterministic checks passed\n\nStop run when core conclusion lacks evidence or reviewer affiliation remains unknown."}