Chapter 20: Writing and Editing with AI

Writing is the core professional act of knowledge work. You write to persuade, to inform, to document, to build trust, to create a record, and to think. For most professionals, writing takes more time than any other single activity — and it is also the activity where AI assistance, applied thoughtfully, produces the largest productivity gains.

That productivity gain comes with a risk that is easy to underestimate. Writing has a voice. It carries personality, credibility, and relationship. When AI takes over too much of the writing process, the voice flattens. The prose becomes recognizable — not as yours, but as AI's. Clients notice. Readers notice. The very effectiveness that made writing valuable is degraded.

This chapter builds a complete AI-assisted writing workflow that captures the speed advantages of AI without surrendering the voice and judgment that make professional writing worth reading.

1. The Writing Workflow and Where AI Fits

Professional writing is not a single act. It is a sequence of distinct cognitive tasks, each with different demands and different tolerances for AI involvement. Understanding that sequence is the first step toward integrating AI effectively.

The Seven Stages of the Writing Workflow

Stage 1: Ideation Ideation is the generative phase — finding angles, uncovering what is interesting, identifying what you actually want to say. AI excels here because the stakes of a bad idea at this stage are low. You are not committing to anything. You are exploring.

At the ideation stage, use AI to generate lists: ten possible angles on this topic, five counterintuitive framings, three ways a skeptical reader would push back on this premise. The goal is not to use what AI produces but to react to it — to notice what resonates, what you disagree with, what sparks a better idea you would not have generated on your own.

AI fit at ideation: High. Low risk, high generative value.

Stage 2: Outlining Outlining is where ideation becomes structure. You are deciding on the argument arc, the sequence of sections, and the rough proportions of each part. AI is useful here for generating alternative structures and pressure-testing your logic.

Give AI your thesis and ask it to propose three different structural approaches. Ask it to identify what is missing from your current outline. Ask it to play the role of a skeptical editor: "What would a reader find confusing about this structure?"

AI fit at outlining: High. AI is good at structural logic and can see gaps you are too close to notice.

Stage 3: First Draft First drafts are where most writers lose the most time. The blank document is paralyzing. AI can eliminate most of that paralysis — not by writing your first draft for you, but by generating a starting version that you respond to rather than create from nothing.

There is an important psychological insight here: it is far easier to edit than to originate. A mediocre AI-generated first draft that you can improve is more productive than a blank page you approach with creative anxiety. Use AI to generate the first pass, then treat it as raw material.

AI fit at first draft: High for generation, but requires significant human rewriting to retain voice.

Stage 4: Developmental Edit The developmental edit addresses structure and argument. Does the piece make its case? Is the sequence logical? Are transitions effective? Is anything missing or redundant?

At the developmental stage, do not ask AI to rewrite. Ask it to critique. "Read this draft and identify the weakest argument." "Does this piece deliver on the promise of the opening?" "What would a skeptical reader find unconvincing?" Use AI as a critical reader, not as a writer.

AI fit at developmental edit: High for critique, risky for rewriting.

Stage 5: Line Edit The line edit improves prose at the sentence level — rhythm, word choice, clarity, concision. This is where the risk of AI voice bleed is highest. When you ask AI to "improve" or "clean up" your sentences, it will tend to make them sound like AI sentences — smooth, competent, and characterless.

At the line edit stage, use AI surgically. Ask it to work on specific sentences that you know are weak. Ask it to generate three alternative versions of a particular paragraph so you can choose elements from each. Keep the bulk of the line editing as a human task.

AI fit at line edit: Moderate. Useful for stuck sentences, risky as a wholesale process.

Stage 6: Proofreading Proofreading is the mechanical layer — grammar, spelling, punctuation, consistency. This is where AI is unambiguously useful. Tools like Grammarly and the proofreading capabilities in Claude or ChatGPT catch errors reliably without threatening voice.

AI fit at proofreading: Very high. Low risk, high reliability.

Stage 7: Audience Adaptation Audience adaptation transforms a piece written for one context into a version suited to a different audience. Taking a technical report and creating an executive summary. Taking a long blog post and creating a LinkedIn version. Adapting a US-market piece for a UK audience.

AI is excellent at audience adaptation because the structural rewriting involved is systematic. You are not asking AI to create — you are asking it to transform according to clear parameters.

AI fit at audience adaptation: High. Define the target audience explicitly and clearly.

2. Maintaining Your Voice While Using AI

The most common complaint about AI-assisted writing is that the output sounds like AI. This is not inevitable. It is the result of asking AI to do too much of the writing, with too little guidance about the voice you want.

Why AI Default Style Is Recognizable

AI language models are trained on enormous corpora of human text — but the distribution of that text is skewed toward professionally written, published, and formally edited content. The result is prose with distinctive characteristics: a preference for active voice; smooth transitions; a tendency toward hedging ("it is important to note that"); a particular rhythm of sentence variety that sounds calibrated rather than natural; and a fondness for three-part lists.

When you read prose that was clearly AI-generated, you are reading the statistical average of its training distribution. That average is competent but characterless. It sounds like a corporate website or a LinkedIn thought leadership post — smooth, readable, and empty of personality.

The "AI as Ghostwriter" vs. "AI as Editor" Distinction

There is a fundamental difference between using AI as a ghostwriter — giving it a brief and asking for prose — and using AI as an editor — giving it your prose and asking for feedback, improvement suggestions, or alternative phrasings.

When AI is the ghostwriter, it brings its own voice because you have not provided one. When AI is the editor, it works within the constraints of your existing voice. The second mode is far less likely to produce AI voice bleed.

As a general principle: the more of your own prose you put into the context window, the more AI will write prose that sounds like you. The less of your own prose you provide, the more AI will write prose that sounds like AI.

Style Extraction Technique

Style extraction is a technique for capturing your voice in a reusable prompt. The process:

1. Collect three to five examples of your best writing — pieces that you feel represent your voice accurately.
2. Submit them to an AI model with this prompt: "Analyze the writing style of these samples. Identify the distinctive characteristics: sentence length variation, vocabulary level, use of transitions, tone, use of examples, paragraph structure, and any other notable features."
3. Ask the AI to produce a style guide from its analysis: "Write a brief style guide based on these samples that another writer could use to approximate this voice."
4. Save that style guide. It is now a reusable prompt component.

When you ask AI to help with any piece of writing, include the style guide: "Write in the style described in the following style guide: [insert]."

This does not produce perfect voice replication, but it dramatically narrows the gap between AI default style and your own.

Voice Injection: Writing Samples in Context

Beyond the style extraction technique, the most reliable voice preservation method is volume. When you submit a piece for AI editing or expansion, include multiple paragraphs of your own writing in the prompt before the section you want help with. More of your prose in the context window pulls AI output toward your voice rather than away from it.

A practical prompt structure for voice injection:

Here are several paragraphs I have written. Please study my voice carefully before completing the task below.

[SAMPLE 1: your paragraphs]
[SAMPLE 2: your paragraphs]
[SAMPLE 3: your paragraphs]

TASK: [Edit/expand/rewrite this section in my voice:]
[the section you want help with]

3. AI for Different Document Types

Different document types have different writing conventions, different audiences, and different tolerances for AI involvement. Here is a practical guide.

Blog Posts

Blog posts are the document type where AI assistance is most seamlessly productive. The genre tolerates and rewards iteration. First drafts are expected to be rough. The relationship between writer and reader is relatively informal.

Recommended workflow: Use AI to generate three possible angles for the post topic. Choose one. Use AI to build an outline. Write the introduction yourself. Ask AI to generate draft versions of the middle sections. Rewrite those sections in your voice, keeping the strong moves and discarding the weak ones. Write your own conclusion. Use AI for proofreading.

Reports

Reports — whether internal, consulting, or analytical — require more structured AI involvement and more careful voice and accuracy management.

For reports, AI is most valuable in two places: structural scaffolding (generating a comprehensive outline that ensures nothing is missed) and the executive summary (taking a long, dense document and distilling it into a tight, readable summary). The body of a report should be written largely by hand, with AI available for stuck passages and for ensuring internal consistency.

Emails

Emails are the underappreciated high-volume writing task. Most professionals write dozens of emails per day; AI can accelerate email writing significantly.

Use AI for: drafting responses to complex requests where you are uncertain how to structure your answer; toning difficult emails (asking AI to make a message more direct, warmer, or more diplomatic); generating a first draft when you are procrastinating on a message you do not want to write.

Do not use AI for: brief, casual correspondence where AI overhead exceeds writing time; emails that require deep knowledge of a personal relationship that AI cannot access.

Marketing Copy

Marketing copy requires precise emotional calibration. AI is useful for generating variations — three different versions of a headline, five different CTA phrasings — and for first drafts that a human marketer then sharpens. AI is not useful as the final voice for marketing copy. The flattening effect is most damaging in writing where voice is literally the product.

Social Media

Social media content is well-suited to AI assistance because volume is high, the posts are short, and iteration is fast. Use AI to generate a batch of post drafts from a single source piece ("give me ten LinkedIn posts based on this blog post, each with a different angle"). Then select and edit the strongest two or three.

Technical Documentation

Technical documentation is one of the best AI use cases in writing. The genre prioritizes clarity and completeness over voice. AI is excellent at generating comprehensive docstrings, README sections, step-by-step instructions, and API documentation when given accurate technical inputs.

The caveat: technical documentation must be accurate, and AI has no way to verify the technical details you give it. If you feed AI incorrect specifications, it will produce clear, readable, incorrect documentation. Human technical review is non-negotiable.

Proposals

Proposals are high-stakes persuasive documents. They typically require a combination of technical specification, narrative case-making, and careful understanding of the evaluating audience. AI can help with structure, with generating boilerplate sections (team bios, methodologies, company background), and with editing for concision and clarity. The core argument — why you, why now, why this — should be developed by the human who understands the relationship and the context.

4. The "Expand-Then-Contract" Technique

One of the most reliably productive AI writing techniques is expand-then-contract. The principle is simple: generate more than you need, then select and edit down to the final version.

Most writers experience creative constraint — the sense that the words they are producing are the best available and that any deviation represents a degradation. This constraint is largely an illusion. There are usually many good ways to express an idea, and the first one that comes to mind is not necessarily the best.

AI eliminates the creative constraint problem because it generates alternatives without psychological friction. You can ask AI for five versions of a paragraph, select the structural moves you like from two or three of them, and synthesize a version that is better than any of the original options.

How to use expand-then-contract:

1. Write your core idea in rough notes or bullet points.
2. Ask AI to write a full section based on those notes — typically 20-50% longer than your target length.
3. Read the AI output and mark the passages, phrases, and moves that work.
4. Write your own version, using the marked elements as raw material.
5. Contract to target length by cutting what is weakest.

The key insight is that you are not adopting AI's prose — you are using it as a quarry for ideas, structures, and phrasings that you then recut in your own voice.

5. AI Editing Modes

When you ask AI to "edit" a piece, the word is ambiguous. There are at least four distinct editing modes, and being explicit about which one you want produces dramatically better results.

Substantive Editing

Substantive editing addresses argument, structure, and completeness. Ask for it explicitly:

"Act as a substantive editor. Do not improve the prose. Instead, identify: (1) sections where the argument is unclear or unconvincing, (2) places where the structure would benefit from reorganization, (3) anything important that is missing, and (4) anything that should be cut."

Line Editing

Line editing addresses prose quality at the sentence level. When you ask for it, constrain the scope:

"Line edit the following paragraph for clarity and concision. Preserve my voice — do not rewrite sentences that are working, only fix ones that are unclear, wordy, or awkward."

Proofreading

Proofreading is mechanical error correction. Distinguish it from editing clearly:

"Proofread the following text. Correct grammar, spelling, punctuation, and capitalization errors only. Do not change word choice or restructure any sentences."

Tone Check

Tone check is a specialized editing mode where you ask AI to evaluate whether the piece's tone is appropriate for its context and audience:

"Read this email and assess its tone. I am writing to a client who is frustrated with a service failure. Is the tone appropriately apologetic and professional? Does anything sound defensive? Does anything sound overly formal given our relationship?"

6. The "Multiple Drafts" Technique

One of the most underused AI writing techniques is the multiple drafts approach. Instead of asking AI for one draft and then editing it, you ask for three significantly different versions and then synthesize.

The process:

1. Define your brief clearly: topic, audience, tone, key points to include, approximate length.
2. Ask AI for three versions: "Write three different versions of this section. Make each version genuinely different in its approach — not just paraphrased, but structurally and rhetorically different. Label them Version A, B, and C."
3. Read all three. Note what works in each: the opening of A, the middle argument structure of B, the conclusion of C.
4. Write a synthesis: "Now write a fourth version that incorporates these specific elements: [list what you liked]. Keep my voice in mind."
5. Edit the synthesis in your voice.

This technique produces better outcomes than single-draft AI generation for two reasons. First, it forces AI to explore the solution space more fully. Second, it gives you genuine choice — you are selecting among alternatives rather than accepting or rejecting a single option.

7. Collaborating on Long-Form Content

Long-form content — a 5,000-word report, a white paper, a book chapter — requires workflow adaptations that go beyond what works for short-form pieces.

The core challenge with long-form AI collaboration is coherence. When you generate sections individually, the resulting piece often lacks consistent voice, has internal contradictions, and loses the thread of an argument. Managing this requires deliberate structure.

Section-by-Section Workflow for Long Form

1. Build a detailed outline before generating any prose. The outline should specify the argument or content of each section, not just its topic.
2. Write the introduction yourself. This establishes the voice and the argumentative frame that everything else must align with.
3. Generate sections one at a time. For each section, include in the prompt: (a) the full outline, (b) your introduction, and (c) any previously written sections. This gives AI the context to maintain consistency.
4. After generating each section, read it against what came before. Catch inconsistencies and voice drift immediately — they are much harder to fix after the whole piece is assembled.
5. Conduct a final coherence pass. After assembling all sections, read the whole piece and look for: repeated phrases, contradictory claims, abrupt tonal shifts, and missing transitions.

The Context Window Limitation

AI models have context window limits that affect long-form collaboration. For pieces longer than approximately 4,000-5,000 words, you may not be able to include the entire document in a single prompt. In this case, work in the largest chunks your model supports, and be deliberate about what context you include with each generation request.

8. Fact-Checking in Writing Workflows

AI writes confidently about things it does not know. In writing workflows, this manifests as invented statistics, misattributed quotes, incorrect dates, and fabricated examples. These errors are not obvious — they look exactly like correct facts.

The rule is absolute: every factual claim in AI-assisted writing must be verified against a primary source before publication.

This is not a conditional rule. It does not apply only to technical writing or high-stakes documents. It applies to every factual claim in every AI-assisted piece. The embarrassment risk of publishing invented facts is too high, and the verification time is low compared to the writing time AI saves.

Practical workflow integration: during your editing pass, highlight every factual claim. Then, in a separate verification pass, check each claim against a source. Do not combine these tasks — the verification pass is easily skipped if it is entangled with editing.

Fact-checking in writing workflows is treated more fully in Chapter 30, which covers verification across all professional contexts.

9. Attribution and Disclosure

As AI writing assistance becomes prevalent, professional and ethical norms around attribution and disclosure are evolving. Different contexts have different expectations.

In academic and journalistic contexts, disclosure expectations are stringent and sometimes formally required. In professional and commercial writing contexts, norms are less codified but are developing rapidly.

The current practical guidance: - When AI drafts significant portions of work, the professional who publishes that work is responsible for its accuracy, appropriateness, and voice — regardless of disclosure. - Disclosure is appropriate when: the audience has a reasonable expectation to know, when the credibility of the work is tied to human authorship, or when organizational policy requires it. - Never falsely claim human authorship in contexts where AI assistance matters — academic submissions, legal documents, contexts where the audience is explicitly relying on the author's personal judgment and expertise.

Chapter 33 covers attribution, disclosure, and the evolving ethics of AI-assisted work in full.

10. Alex Scenario: Full Blog Post Workflow from Brief to Published Draft

🎭 Scenario Walkthrough: Alex's Blog Post

Alex manages content marketing for a mid-sized B2B software company. She has been asked to write a blog post: "Why Remote Teams Underinvest in Async Communication Tools." Her audience is mid-level managers and team leads at companies with distributed teams. Target length: 1,200 words. Tone: authoritative but conversational. The post should lead to a CTA for the company's async communication tool.

Step 1: Ideation (AI-assisted)

Alex submits the brief to Claude and asks for ten different angles on the topic. Among the ten, two catch her attention: "The meeting paradox: teams that schedule more meetings communicate less," and "The cost-of-context-switching case for async." She decides to combine both angles.

Step 2: Outline (AI-assisted)

She submits: "Help me outline a 1,200-word blog post for this audience [describes audience]. The post combines these two angles [states them]. Key points to make: [lists five core arguments from her knowledge]. Structure the outline with estimated word counts per section."

Claude returns an outline. Alex reviews it, moves one section, and removes a proposed conclusion that feels too sales-heavy.

Step 3: First Draft (AI-generated, human-revised)

Alex submits the revised outline and asks Claude for a first draft. She specifies: "Match this voice [includes two paragraphs from a previous high-performing blog post]. Write for the audience I described. Do not make the CTA aggressive — it should feel like a natural next step."

The draft comes back at 1,340 words. Alex reads it immediately with a critical eye. She marks three passages that are strong and use them nearly verbatim. She rewrites the introduction from scratch — the AI version is competent but lacks the specific opening hook she has in mind. She rewrites two middle paragraphs where the AI made generalizations she knows are not supported. She adjusts the tone throughout to match her company's brand voice.

Step 4: Developmental Check (AI-assisted)

Alex submits her revised draft with the prompt: "You are a developmental editor reviewing this piece for a B2B marketing blog. Does this piece make its case effectively? Is anything missing or redundant? Does the CTA feel earned or forced?" Claude identifies one argument that lacks support and suggests she add a specific example. She agrees and adds one from a client case study she knows.

Step 5: Line Edit (human-led)

Alex does her own line edit, using AI only for two sentences she finds clunky. She submits those sentences: "Give me three alternative versions of this sentence that are more direct: [sentence]."

Step 6: Proofreading (AI-assisted)

She runs the piece through Grammarly, then submits the full text to Claude: "Proofread for grammar, spelling, and punctuation only. Do not change any wording."

Step 7: Final Review (human)

Alex reads the final piece aloud. This is her non-negotiable final step — reading aloud catches rhythm problems and voice anomalies that silent reading misses. She makes three small adjustments.

Total time: ninety minutes, for a post she estimates would have taken three and a half hours without AI assistance. More importantly: the post sounds like her, not like AI.

11. Elena Scenario: Consulting Report Workflow — Structure, Draft, Edit, Client Adaptation

🎭 Scenario Walkthrough: Elena's Consulting Report

Elena is a strategy consultant delivering a market assessment to a financial services client. The deliverable is a 25-page report covering competitive landscape, regulatory environment, and strategic recommendations. She has four days and is working on two other engagements simultaneously.

Structure Phase (Day 1)

Elena has her primary research: interviews, market data, regulatory filings, and analyst reports. She submits a summary of her research findings to Claude: "Based on these research findings [summarizes key themes], help me build a comprehensive outline for a 25-page consulting report to a financial services client. The report should follow McKinsey-style structure: executive summary, current state assessment, market opportunity analysis, competitive positioning, strategic options, recommendation, and implementation roadmap."

Claude returns an outline. Elena cross-checks it against the research gaps she knows exist and adds a section on regulatory risk that Claude did not include.

Drafting Phase (Days 1-2)

Elena drafts the executive summary herself — she has been burned before by AI executive summaries that sounded generic and lost the sharp analytical point of the underlying research. For the body sections, she uses AI to generate first drafts of the more factual and descriptive sections (competitive landscape, regulatory overview), while writing the analytical sections herself.

For each AI-generated section, she feeds in her research notes: "Draft the competitive landscape section [3-4 pages] based on these findings: [notes]. Use this outline structure: [structure]. Write in a professional, analytical consulting tone — avoid hedging language and make claims directly."

Editing Phase (Day 3)

Elena uses AI for substantive editing: "Act as a critical editorial partner for a strategy consulting report. Identify: sections where the logic is unclear, any recommendations that are not fully supported by the analysis, and any section that a CFO-level reader would find vague or unconvincing."

The feedback identifies two places where her strategic logic has a gap. She addresses both.

Client Adaptation (Day 3-4)

The client has asked for a 2-page executive brief in addition to the full report, formatted for their board meeting. Elena submits the full report to Claude: "You are adapting this consulting report for a board-level audience. Create a 2-page executive brief that: captures the three most critical findings, states the recommendation clearly in the first paragraph, and uses the simplest possible language — assume the board has not read the underlying report."

Elena reviews the brief carefully, adjusts the recommendation language to reflect her judgment about the political dynamics of the client's board, and adds a specific risk caveat that she knows is important given recent regulatory changes.

12. Common Writing AI Failure Modes

Understanding the ways AI writing assistance fails makes you a better user of it.

AI Padding

AI has a tendency to fill space. When you ask for a certain length, AI will often produce that length through redundancy — repeating points in slightly different language, adding qualifier clauses that add no information, inserting transitional summaries that restate what was just said.

The fix: edit for concision aggressively. When you are editing AI-generated text, cut first and add second. Every sentence should earn its place.

AI "Voice" Bleeding In

Even when you provide writing samples and style instructions, AI voice will bleed into output if you are not vigilant. The signs: sentences that are unusually smooth and balanced; an absence of the imperfections and idiosyncrasies of natural prose; a tendency to state things in three parallel clauses; an avoidance of colloquialisms or humor that are characteristic of your real voice.

The fix: read your AI-assisted pieces aloud. Voice anomalies are immediately obvious when spoken that are invisible when read silently.

Overconfident Claims

AI makes factual claims with the same confident tone whether they are well-established facts or plausible-sounding inventions. This is particularly dangerous in research-backed writing where specific claims carry evidentiary weight.

The fix: treat all AI-generated factual claims as unverified until you have checked them.

Lost Nuance in AI Editing

When you ask AI to "tighten" or "clean up" writing, it frequently removes nuance — qualifications that were deliberate, hedges that were appropriate, complexity that was necessary for accuracy. The smooth, clear version AI produces is sometimes wrong.

The fix: when requesting line editing, be explicit: "Do not remove qualifications or hedges that seem deliberate. Only fix clear grammatical errors and awkward constructions."

13. Research Breakdown: AI Writing Assistance and Quality Studies

📊 Research Breakdown

The empirical evidence on AI writing assistance is growing but contested. Here is what the current research suggests:

Speed gains are consistent and substantial. Multiple studies of knowledge workers using AI writing assistance find productivity gains in the range of 30-50% for first draft generation tasks. A 2023 MIT study found that workers using AI assistance for writing tasks completed them 37% faster on average, with lower-skilled writers seeing the largest gains.

Quality effects are mixed and task-dependent. For pure writing quality as judged by human raters, AI assistance tends to improve the quality of writing from lower-skilled writers more significantly than from higher-skilled writers. For expert writers, AI assistance can actually introduce quality regressions if the workflow is not carefully managed.

Voice authenticity is measurable and matters. Research on reader perception of AI-generated content finds that readers can detect AI voice at rates significantly above chance, and that detected AI voice negatively affects credibility perceptions — particularly in professional and expert contexts. A 2024 study found that readers rated identical content as less credible when they believed it was AI-generated.

The workflow design effect. The most important variable in AI writing quality is not the model used but the workflow design — how AI is integrated, at which stages, and with what human review. Well-designed workflows consistently outperform both no-AI and poorly-designed AI workflows.

The practical implication is clear: AI writing assistance works best for knowledge workers who treat it as a workflow integration challenge rather than a capability deployment challenge. The tool is widely available; the skill is in using it within a workflow that preserves what matters.

14. Workflow Prompts for Each Stage

🗣️ Script/Template: Writing Workflow Prompts

Ideation:

I am planning to write a [document type] on [topic] for [audience].
Generate 10 different angles I could take on this topic.
For each angle, suggest: the core argument, the most compelling opening hook,
and the likely objection a skeptical reader would raise.

Outlining:

Help me build a detailed outline for a [length] [document type] on [topic] for [audience].
My core argument is: [argument]
Key points I must include: [list]
The piece should end with: [CTA/conclusion type]
Include estimated word counts for each section.

First Draft:

Write a first draft of [section] based on this outline: [outline]
Voice guidance: [paste style guide or writing samples]
Audience: [describe]
Key constraints: [any specifics about tone, technical level, etc.]
Write at approximately [X] words.

Developmental Edit:

Act as a developmental editor reviewing this [document type].
Evaluate:
1. Does the argument hold together?
2. Is the structure logical?
3. What is missing or should be cut?
4. What would this audience find confusing or unconvincing?
Do not rewrite anything — only provide editorial feedback.

[PASTE DOCUMENT]

Line Edit:

Line edit the following [section] for clarity and concision.
Preserve my voice — only fix sentences that are unclear, wordy, or awkward.
Do not add new information or remove deliberate qualifications.

[PASTE SECTION]

Proofreading:

Proofread the following text. Correct grammar, spelling, punctuation,
and capitalization errors only. Do not change any wording or sentence structure.

[PASTE TEXT]

Audience Adaptation:

Adapt the following [document type] for [new audience].
Original audience: [describe]
New audience: [describe]
Key changes needed: [formal → informal, technical → plain language, etc.]
Target length: [length]

[PASTE ORIGINAL]

Summary

Writing is the highest-leverage AI use case for most knowledge workers, and it is also the use case where failure modes are most consequential. The framework in this chapter — AI at every stage, with human judgment governing voice, accuracy, and final quality — allows you to capture AI's speed advantages without surrendering what makes your writing worth reading.

The core discipline is knowing which stage of the workflow you are in and matching your AI use to that stage. AI for generation. AI for critique. Human for voice. Human for verification. AI for proofreading. Human for the final read.

Voice is not a vague aspiration. It is a technical challenge with specific solutions: style extraction, voice injection, reading aloud, the multiple drafts technique. These are not sophisticated techniques — they are straightforward workflow steps that most AI writing users skip because they have not been told about them.

Do not skip them.

✅ Best Practice Before using AI to draft any piece of professional writing, write your own introduction — even a rough one. It establishes the voice and the frame that AI generation must work within. AI completions are strongly influenced by what precedes them; your prose is the best possible steering input.

⚠️ Common Pitfall Asking AI to "edit" without specifying which type of editing you want. Developmental, line, and proofreading edits require completely different instructions. Undifferentiated "edit this" prompts produce outputs that change things you wanted to keep and miss things you wanted fixed.

💡 Intuition Think of AI in writing as a very skilled collaborator who has read everything and has no taste. It knows the conventions of every genre, can generate competent prose on any topic, and has no personal stake in whether the result is actually good. Your job is to provide the taste, the voice, and the judgment. AI provides the speed and the raw material.