AI Editing and Polishing
AI drafts are raw material. The polishing process — editing for clarity, flow, accuracy, and formatting — is where a mediocre draft becomes publishable content. This lesson covers the multi-pass editing workflow, specific AI-assisted editing prompts for each pass, and the quality gates that every article must clear before publishing.
Part 1 — The 5-Pass Editing Framework
Why Single-Pass Editing Fails
A single editing pass tries to fix everything simultaneously: grammar, flow, accuracy, formatting, tone. The result is that you fix the obvious errors and miss the subtle ones. Each pass should focus on one dimension only.
flowchart TD
A[AI Draft\nRaw Output] --> B[Pass 1: Structural\nH2/H3 flow, section order]
B --> C[Pass 2: Depth\nFiller removal, depth addition]
C --> D[Pass 3: Voice\nHumanization, personality]
D --> E[Pass 4: Accuracy\nFact-check, source verify]
E --> F[Pass 5: Format\nLists, tables, callouts, SEO]
F --> G[Publish-Ready\nArticle]
style A fill:#8B0000,color:#fff
style G fill:#217346,color:#fff
What Each Pass Focuses On
- Pass 1: Structure
- Pass 2: Depth
- Pass 3: Voice
- Pass 4: Accuracy
- Pass 5: Format
Question: Does the article flow logically from H1 to the final section?
- Does the H2 order make sense? (Would a reader be confused by the sequence?)
- Does each H2 fulfill its promise? (No sections that drift off-topic)
- Are there redundant sections that say the same thing differently?
- Is the article too long? Could any section be cut without losing value?
AI assist: "Review this outline and flag any H2 sections that are redundant, out of order, or don't serve the main topic."
Question: Does every paragraph add new information?
- Identify "filler paragraphs" — text that exists to increase word count
- Identify "surface sections" — topics mentioned but not explored
- Every claim must be supported by evidence, example, or data
- Delete any sentence where the reader thinks "so what?"
AI assist: "Review each paragraph. If a paragraph can be deleted without the reader missing any information, flag it as filler."
Question: Does this sound like a human expert wrote it?
- Apply the 8 humanization transforms from the previous lesson
- Check for hedging language ("could potentially," "might help")
- Verify at least 1 opinion/position is stated per section
- Ensure no "In conclusion" or "It is important to note" phrases remain
AI assist: "Flag every sentence that sounds AI-generated. For each, explain the specific pattern and suggest a human alternative."
Question: Is every fact verifiable?
- Apply the 3-pass verification process from the previous lesson
- Every stat has a named source and year
- Every product feature is verified against current documentation
- No "studies show" or "experts agree" without named sources
AI assist: Note — AI cannot reliably fact-check itself. This pass requires human verification against primary sources.
Question: Is the article optimized for both skimmers and readers?
- 300-word rule: No 300+ consecutive words without a non-paragraph element
- Bold text marks 3–5 key insights per section
- Tables are used for comparisons, lists for sequences/options
- Callouts are used sparingly (max 1/500 words)
- Primary keyword appears in H1, first 100 words, and meta description
- Featured snippet opportunities are structured correctly
AI assist: "Audit this draft for formatting. Flag any section over 300 words without a list, table, or callout. Identify paragraph content that should be a table or list."
Part 2 — Editing Prompts for Each Pass
Ready-to-Use Prompts
Structural Edit Prompt
Role: Senior content editor Task: Review this article's structure:
- Can any H2 sections be merged without losing value?
- Is the H2 order logical — does each section build on the previous?
- Are there redundant sections that repeat the same idea?
- Flag any section where the content doesn't match the H2 heading's promise
- Is the article the right length, or does it have sections that should be cut? Input: [Paste Draft]
Depth Edit Prompt
Role: Content quality auditor Task: Identify every weak section in this article:
- Flag any paragraph that could be deleted without losing information (filler)
- Flag any claim without evidence, example, or data (unsupported)
- Flag any section that mentions a topic but doesn't explore it (surface-level)
- For each flagged item, suggest: Delete, Expand, or Add Evidence Input: [Paste Draft]
Voice Edit Prompt
Role: Brand voice editor Task: Make this draft sound like a human expert, not an AI:
- Replace all hedging language with direct statements
- Delete all instances of "It is important to note" and "Furthermore"
- Add conversational transitions where the tone feels flat
- Where possible, add a specific opinion or recommendation
- Flag any paragraph that feels emotionally neutral — suggest where humor, surprise, or frustration would fit Input: [Paste Draft]
Format Edit Prompt
Role: Web content formatter Task: Optimize this article for web readability:
- Identify any 300+ word blocks without visual elements — suggest where to add tables, lists, or callouts
- Convert paragraph-form comparisons to tables
- Convert paragraph-form sequences to numbered lists
- Verify bold text marks key insights (not everything, not nothing)
- Check the primary keyword appears in: H1, first 100 words, one H2, and natural body text (once per 300–500 words) Input: [Paste Draft]
Part 3 — The Pre-Publish Quality Gate
The Final Checklist Before Publishing
Every article must clear these gates. If any gate fails, the article goes back for revision.
| Gate | Criteria | Pass / Fail |
|---|---|---|
| Structure | H2s pass the skim test. No redundant sections. Logical flow | ☐ |
| Depth | Zero filler paragraphs. Every claim has evidence | ☐ |
| Voice | No AI patterns detected. Opinion/position stated. Personality present | ☐ |
| Accuracy | All stats sourced. All product features verified. No hallucinations | ☐ |
| Format | 300-word rule met. Tables, lists, callouts used correctly. SEO elements in place | ☐ |
| Intent alignment | Article matches search intent (verified by SERP check) | ☐ |
| Differentiation | At least one clear reason this article is better than the current #1 result | ☐ |
Part 4 — Bad vs. Good Examples
- ❌ Single-Pass Edit
- ✅ 5-Pass Edit
Writer's process: "I ran the AI draft through Grammarly, fixed the typos and grammar errors, and published."
(Why it fails: Grammarly catches mechanics — not depth, not voice, not accuracy, not structure. The article is grammatically perfect and substantively empty.)
Writer's process:
- Structure: Moved the "Common Mistakes" section before "Best Practices" — readers need to know what NOT to do before learning what TO do. Cut the redundant "Why This Matters" section (already covered in intro)
- Depth: Added a case study to section 3. Replaced "email marketing is effective" with specific ROI data. Expanded the thin "Implementation" section from 50 to 200 words
- Voice: Deleted the AI opener. Added personal experience in section 2. Changed "It is important to consider" to "You need to decide"
- Accuracy: Verified the HubSpot stat — it was wrong ($36/ROI, not $42). Removed the fake McKinsey citation
- Format: Added a comparison table to the tools section. Bolded 4 key insights. Added a :::tip callout for the most actionable advice
(Why it wins: Each pass catches different issues. The final article is structurally sound, genuinely deep, human-sounding, factually accurate, and visually scannable.)
Part 5 — Output Checklist
- 5-pass framework: You apply structure, depth, voice, accuracy, and format passes separately.
- Pass-specific prompts: You have saved, reusable prompts for each editing pass.
- Pre-publish gate: Every article clears the 7-gate checklist before publishing.
- Filler detection: You can identify and remove content that exists only for word count.
- Quality standard: Your final articles require 30–60 minutes of editing, not 3+ hours.
- AI limitations acknowledged: You know that AI cannotFact-check itself — accuracy requires human verification.
Internal use only. Do not distribute externally. For questions or suggested updates, raise with the content lead.