Marketing Content Ideas

Content Ideas Generator

Generates 15–25 content ideas mapped to B2B buying stages and content types.

When to Use

• User wants content calendar, ideation, or kanban backlog
• After ICP + personas exist for a service
• User specifies counts per type (e.g. 5 LinkedIn, 3 blog, 2 case study)

Prerequisites

Brand scope. In a multi-brand workspace (2+ brand homes in workspace/firm/brands/), every path this skill reads or writes is brand-scoped: definitional entities (services, ICP, personas, proof, channels) live in firm/brands/{brand}/…, working pipelines carry a {brand}/ segment under their entity root, and the working brand comes from the user's choice or B2BFORCE_BRAND — never guessed. Rules: Brand Scope Gate in AGENTS.md; paths: docs/WORKSPACE.md.

1. workspace/firm/profile.md
2. workspace/firm/services/{service-slug}.md
3. workspace/marketing/icp/{icp-slug}.md + at least one persona
4. Optional APIs: EXA_API_KEY (content landscape), DATAFORSEO_* (keywords), research for service-page opportunities

Hard gate: if the selected service has no linked ICP, or the ICP has no persona, stop and run marketing-icp first. Do not invent ICP/persona context inside the ideas prompt.

Run the central readiness check before generation:

scripts/validate-content-readiness.sh [service-slug] [icp-slug] [persona-slug]

Content types and stage rules

Buying stages: problem → concept → education → decision → vendor

Details: references/buying-stages-and-types.md

Workflow

1. Build context

Assemble from workspace:

• Company: name, tagline, description, industry, brand_voice, specializations
• Service: name, type, challenges, features, outcomes, process, differentiators, fit_criteria
• ICP: target client company segment; pain_points, goals, trigger_events, buying_mode, messaging_angle, job_to_be_done, anti_fit_criteria
• Persona: decision maker or user; job_titles, seniority, pain_points, goals, content_formats, channels, objections
• Persona distribution: if user selects multiple personas, either generate a separate batch per persona or assign explicit counts per persona; every idea must carry one persona slug.
• Exclusion list: firm profile section existing_content, existing idea titles, existing unique_angle, existing buyer_question, and user-provided URLs/titles to avoid
• Content landscape: Exa search — top 10 articles on topic in last 12 months (if API set). If no research API is used, mark ideas as research_mode: dry_run.

2. Service page opportunity research (optional)

When landing_page count > 0 and user requests research:

Run service-page research — 5 queries:

competitors, landing_structure, objections, social_proof, cta_strategies

Optional DataForSEO keyword data.

Save research JSON to include in idea files for landing_page type only.

3. AI generation

System prompt: references/system-prompt.md. A buying-mode-aware appendix is always added on top; it tailors ideas to the ICP's buying_mode — reactive (problem/pain → urgent, problem-solving) vs proactive (challenge/opportunity → aspirational, growth-oriented), with mixed as the default. The ICP's buying_mode, messaging_angle, and market_research are passed in context for the AI to adapt.

User prompt includes:

• Count per type OR total count distributed across stages
• Language from firm context
• Existing ideas from workspace/marketing/content/ideas/ as JSON (dedup)
• Dedupe against titles, buyer questions, unique angles, and problem framing, not only exact title matches
• Big5 topic guidance where relevant (cost, problems, comparisons, alternatives, reviews)
• hook_type per idea: data | question | contrarian | specificity | problem | story
• Quality rubric per idea: buyer question, stage fit, service fit, proof source, non-generic angle, next action
• Persona-aware distribution if more than one persona is selected
• Research mode: market_informed only if Exa/DataForSEO/SERP/market context was used; otherwise dry_run
• SEO discipline: target_keyword only for blog_post and landing_page; set it empty/null for LinkedIn, X, case study, and prospecting
• Proof discipline:
  • do not invent client names, metrics, page counts, revenue numbers, or research claims;
  • use specific numbers only when they exist in firm/service/proof context;
  • if case_study is requested and no real proof exists, create the idea with a [Client] placeholder and proof_source: needs real client proof before draft;
  • never generate a case study draft from a placeholder proof idea.

Output schema: references/output-schema.md

Generate batch ID: batch-{YYYY-MM-DD}-{uuid-short} — same ID in all idea files from this run.

4. Write workspace outputs

One file per idea:

workspace/marketing/content/ideas/{content_type}--{buying_stage}--{slug}.md

Keep all ideas in one folder. Do not create type subfolders. The filename is a human scanning index for large backlogs; frontmatter remains the source of truth.

---
title:
content_type: blog_post
buying_stage: problem
status: new
language: en
service: {slug}
icp: {slug}
persona: {slug}
buyer_question:
big5_topic: problems          # optional: cost|problems|comparisons|alternatives|reviews
target_keyword:                 # optional, from SEO research
generation_batch: batch-2026-06-01-abc123
competitive_research: false     # landing_page only — path to research file if saved
hook_type: data                 # metadata for quality (optional in frontmatter)
unique_angle:
proof_source:
next_action:
recommended_next_skill: marketing-content-blog-post
research_mode: dry_run
---

Description body: 2–3 sentences on content angle.

If landing_page with research, also write: workspace/marketing/content/ideas/{content_type}--{buying_stage}--{slug}-research.json

5. Default distribution (if user doesn't specify)

Channel registry first. If workspace/marketing/channels/ (multi-brand: firm/brands/{brand}/channels/) holds channel files, they decide distribution: enable only the content types some active channel lists in content_types, and size counts to the channels' schedule lines — a channel posting once a week does not need five ideas a week. Read each channel's body ("what to publish here") as context for the angles proposed. Rules: docs/content-generation.md, "Distribution channels".

Fallback (no channel files). Only content types that are enabled are generated. By default just three are enabled: LinkedIn×5, X×5, blog×3 (≈13 ideas). case_study has a default count of 2 and landing_page/prospecting_sequence default to 0, but those are off by default — they are only generated if the user enables them. The legacy "total count" mode (no per-type counts) defaults to 20 ideas spread across all 5 stages.

Quality rules

• Titles must use hook patterns (data, question, contrarian, specificity, problem, story)
• Avoid generic titles ("Complete Guide to X", "10 Best Tips")
• Each idea needs unique_angle vs existing ideas and content landscape
• Match content type to buying stage restrictions
• Early stages → educational; late stages → comparison/proof
• No fabricated proof: specific metrics, client names, research claims, and case study facts must come from workspace context or be marked as needing real proof.
• Every idea must include buyer_question, next_action, recommended_next_skill, and research_mode.
• Run scripts/validate-content-ideas.sh after writing ideas; fix any failures before generating drafts.

Related Skills