ai-content-workflow
This skill should be used when the user asks to "build an AI content workflow", "set up AI content production", "design an AI writing process", "create an AI content pipeline", "AI-assisted content creation", "how to use AI for content production", "AI content process", "integrate AI into content workflow", or any variation of designing, building, or optimizing end-to-end content production workflows that incorporate AI tools for B2B SaaS.
AI Content Workflow
An AI content workflow is the end-to-end process for producing content where AI handles the heavy lifting (research, drafting, formatting) and humans handle what AI can't (strategy, accuracy, originality, voice). The goal is 3-5x production speed without sacrificing quality.
The trap most teams fall into: they use AI to draft and then publish with minimal editing. This produces high volume of low-quality content that ranks nowhere and builds no authority. The workflow must enforce quality gates at each stage.
The AI Content Production Pipeline
| Stage | AI role | Human role | Time (AI-assisted) | Time (manual) |
|---|---|---|---|---|
| 1. Strategy | Suggest topics from competitive gaps | Final topic selection and prioritization | N/A | N/A |
| 2. Brief | Generate draft brief from template | Refine angle, add proprietary insights, specify POV | 15 min | 30 min |
| 3. Research | Summarize competitor pages, pull data | Verify facts, add original data | 20 min | 2 hours |
| 4. First draft | Generate 70-80% of the draft | Review for direction and completeness | 30 min | 4-6 hours |
| 5. Accuracy edit | Flag claims that need verification | Verify every fact, stat, and product detail | 15 min | 30 min |
| 6. Quality edit | Suggest improvements | Add voice, originality, POV, proprietary data | 30-45 min | 1-2 hours |
| 7. AEO formatting | Apply schema, suggest FAQ questions | Validate structure, test in AI engines | 15 min | 30 min |
| 8. Publish | Format and schedule | Final review and approval | 10 min | 20 min |
Total with AI: 2-3 hours per 1,500-word piece. Manual: 8-12 hours. That's a 3-5x speed improvement.
Stage-by-Stage Workflow
Stage 1-2: Strategy and brief
AI can help identify content gaps and draft briefs, but humans must make strategic decisions.
AI tasks:
- Analyze competitor content for gaps ("What topics do competitors cover that we don't?")
- Generate draft content briefs from a template
- Suggest FAQ questions from "People Also Ask" data
Human tasks:
- Choose which topics to prioritize based on business goals
- Define the angle and POV (AI can't invent your perspective)
- Specify proprietary data or unique insights to include
Stage 3: Research
AI excels at synthesizing information from multiple sources. It's terrible at verifying accuracy.
AI tasks:
- Summarize competitor pages on the target topic
- Pull relevant data points and statistics from provided sources
- Identify common arguments and counterarguments on the topic
- Draft a competitive landscape overview
Human tasks:
- Verify every data point against the original source
- Add proprietary data from your own product/platform
- Identify unique angles competitors haven't covered
Stage 4: First draft
The draft is where AI saves the most time. But the draft is a starting point, not a final product.
Prompt engineering for better drafts:
| Prompt approach | Bad | Good |
|---|---|---|
| Specificity | "Write a blog post about lead scoring" | "Write a 1,500-word comparison of behavior-based vs demographic lead scoring for B2B SaaS. Include a comparison table. Take the position that behavior signals are more predictive. Target audience: VP Sales at $5-50M ARR companies" |
| Voice | No voice instruction | "Write in a direct, peer-to-peer tone. Short sentences. No filler words. Use 'you' not 'one'. Be opinionated, not balanced" |
| Structure | No structure | "Use this H2 structure: [list of H2s]. First 50 words must directly answer the question. Include at least 2 tables. End with a 5-question FAQ section" |
| Constraints | No constraints | "Do not use: 'leveraging', 'streamline', 'in today's fast-paced world', em-dashes, or 'it's important to note'" |
Stage 5-6: Editing (the critical stages)
Editing is where the value is created. AI produces a competent first draft. Editing transforms it into content worth publishing.
The 5-pass editing framework:
- Accuracy — verify every fact
- Specificity — replace vague with specific
- Voice — eliminate AI markers, add personality
- Structure — AEO formatting
- Originality — add proprietary data and unique POV
See the ai-content-editing skill for detailed editing rules.
Quality Gates
Every piece must pass these gates before publishing. No exceptions.
| Gate | Check | Pass criteria |
|---|---|---|
| Accuracy gate | Every fact, stat, and product detail verified | Zero unverified claims |
| Specificity gate | No vague claims ("many companies", "significant improvement") | Every claim has a number, name, or detail |
| Voice gate | No AI markers (hedging, filler intros, false balance) | Reads like a human expert wrote it |
| Originality gate | Contains something AI couldn't have written | At least 1 proprietary data point, unique POV, or expert insight |
| AEO gate | Structured for AI search extraction | Answer-first, tables, question H2s, schema |
| Quality score | Scored on the AI content quality scorecard | Score 8+ out of 10 |
If a piece fails any gate, it goes back for editing. Never skip gates to meet a deadline. Low-quality content damages your brand more than no content at all.
AI Tool Selection
| Task | Recommended tools | Notes |
|---|---|---|
| First draft generation | Claude, ChatGPT, Gemini | Claude produces the most structured, rule-following output |
| Research and summarization | Claude, Perplexity | Perplexity for real-time research; Claude for synthesis |
| SEO optimization | Clearscope, SurferSEO, Frase | Layer on after draft, not before |
| Schema generation | ChatGPT, Claude | Generate JSON-LD from page content |
| Image generation | Midjourney, DALL-E | For original graphics and visualizations |
| Grammar and style | Grammarly, Hemingway | Final pass for clarity and readability |
Team Roles in an AI Content Workflow
| Role | Responsibility | AI proficiency needed |
|---|---|---|
| Content strategist | Topic selection, brief writing, quality oversight | Medium — needs to write good prompts |
| AI content producer | Draft generation, prompt engineering, first-pass editing | High — expert prompt engineer |
| SME reviewer | Accuracy verification, expert insight | Low — reviews content, doesn't use AI |
| Editor | Voice, quality, AEO compliance | Medium — uses AI for suggestions, makes final decisions |
| Publisher | Formatting, schema, CMS upload | Low-medium — uses AI for schema generation |
Pre-Workflow Setup Checklist
- [ ] AI tools selected and accounts provisioned
- [ ] Content brief template created with AI-specific fields (prompt instructions, constraints)
- [ ] Quality gates defined with pass criteria
- [ ] AI content quality scorecard adopted
- [ ] Editing workflow defined (5 passes with specific owners)
- [ ] AI voice guidelines documented (banned phrases, tone targets)
- [ ] Prompt library built for common content types (comparison, how-to, definition)
- [ ] SME reviewer identified for each content area
- [ ] Publishing checklist includes AEO requirements
- [ ] Team trained on the workflow (especially the editing stages)
Anti-Pattern Check
- Using AI to draft and publishing with minimal editing → The draft is 40% of the work. Editing is 60%. A pipeline that skips editing produces high volume of low-quality content that ranks nowhere
- Over-relying on AI for accuracy → AI hallucinates. Every fact, statistic, product detail, and URL must be human-verified. The accuracy gate is non-negotiable
- No prompt engineering discipline → "Write a blog post about X" produces generic content. Detailed prompts with structure, voice, constraints, and audience produce dramatically better drafts. Invest in prompt quality
- Same prompt for every content type → Comparison pages, how-tos, and definitions require different prompts with different structures. Build a prompt library per content type
- No originality gate → If every piece could have been written by anyone's AI, you have no content moat. Enforce the originality gate: every piece must contain something AI couldn't have written
- Speed over quality → The point of AI is to maintain quality at higher speed, not to sacrifice quality for speed. If your AI workflow produces more content but lower quality, the workflow is broken
Want agents that use skill files like this?
We customize skill files for your brand voice and methodology, then run content agents against them.
Book a call