Session 11.4: Editing AI vs Human Output

Course → Module 11: Quality Control & The Human Gate

Session 4 of 7

Two Different Editorial Jobs

Editing human writing and editing AI writing look similar on the surface. Both involve reading text, identifying problems, and fixing them. But the problems are fundamentally different, and if you use the same approach for both, you will miss what matters.

When you edit human writing, you are working with someone's voice. The text has a perspective, even if poorly expressed. It has quirks, preferences, and patterns that reflect how a specific person thinks. Your job is to preserve that voice while improving clarity, accuracy, and structure.

When you edit AI writing, there is no voice to preserve. The text is a statistical composite of everything the model has processed. Your job is not preservation. Your job is injection: adding the voice, specificity, and perspective that the model cannot generate on its own.

The Editorial Divide: Editing human text is subtractive (remove errors, tighten prose, clarify muddy thinking). Editing AI text is additive (inject voice, add specificity, insert opinion). Different problems require different tools.

What You Fix in Human Text vs. AI Text

Edit Type	Human Text	AI Text
Grammar and mechanics	Common. Typos, comma splices, subject-verb disagreement.	Rare. AI prose is mechanically clean.
Structure	Occasional. Ideas sometimes presented out of order.	Superficially fine, but often follows a generic template rather than the best structure for the argument.
Voice	Present but sometimes inconsistent. Fix by smoothing, not replacing.	Absent or generic. Fix by adding distinctive patterns, rhythm, and perspective.
Specificity	Variable. Some writers over-generalize, others provide too much detail.	Almost always insufficient. AI defaults to general claims. Every general statement needs a specific example.
Opinion	Usually present, sometimes too strong or poorly supported.	Absent or artificially balanced. AI hedges by default. Opinions must be inserted manually.
Facts	Errors from memory or outdated knowledge. Usually in the right ballpark.	Hallucinations. Can be completely fabricated while sounding authoritative.
Rhythm	Distinctive but may need evening out. Long-winded writers need tightening; terse writers need expansion.	Monotonous. Sentences tend toward uniform length and cadence. Needs deliberate variation.

The Absence Problem

The hardest part of editing AI text is detecting what is missing. Human text has visible problems: a misplaced comma, a run-on sentence, a factual error. These are presences you can point to and fix.

AI text has invisible problems. The absence of a personal anecdote. The absence of a specific number where a vague claim sits. The absence of a strong opinion where balanced nothingness occupies space. These absences are invisible if you read passively. They become obvious only when you read as an editor asking: "What should be here that is not?"

flowchart TD subgraph HumanEdit["Editing Human Text"] H1[Read for errors] --> H2[Fix grammar/mechanics] H2 --> H3[Tighten prose] H3 --> H4[Clarify structure] H4 --> H5[Smooth voice inconsistencies] end subgraph AIEdit["Editing AI Text"] A1[Read for absences] --> A2[Inject voice patterns] A2 --> A3[Add specific examples] A3 --> A4[Insert opinions and positions] A4 --> A5[Verify all factual claims] A5 --> A6[Vary sentence rhythm] end style HumanEdit fill:#222221,stroke:#6b8f71 style AIEdit fill:#222221,stroke:#c8a882 style H1 fill:#6b8f71,color:#111 style H2 fill:#6b8f71,color:#111 style H3 fill:#6b8f71,color:#111 style H4 fill:#6b8f71,color:#111 style H5 fill:#6b8f71,color:#111 style A1 fill:#c8a882,color:#111 style A2 fill:#c8a882,color:#111 style A3 fill:#c8a882,color:#111 style A4 fill:#c8a882,color:#111 style A5 fill:#c8a882,color:#111 style A6 fill:#c8a882,color:#111

The Rhythm Fix

AI prose tends toward a specific sentence length. Not short. Not long. Medium. Every sentence, medium. Read three paragraphs of unedited AI output and you feel the monotony even if you cannot immediately name it.

Human prose has natural rhythm variation. A long explanatory sentence followed by a short declarative punch. A fragment. Then a compound sentence that builds momentum through subordinate clauses before landing on its point.

To fix AI rhythm, count the words in each sentence of a paragraph. If they are all within 5 words of each other, you have a monotony problem. The fix: combine two short sentences into one long one. Break one long sentence into a fragment and a follow-up. Insert a one-word or two-word sentence for emphasis. The content does not change. The reading experience does.

The Specificity Injection

AI writes: "Many companies have found success with this approach." A human editor changes it to: "Stripe reduced their onboarding time from 14 days to 3 using this approach. Shopify reported similar results in their 2024 annual report."

The difference is not style. It is substance. The first version communicates nothing verifiable. The second version communicates two specific claims that a reader can check, believe, or challenge. Specificity is the single most valuable edit you can make to AI output.

Every paragraph of AI output should be interrogated: "Is there a specific name, number, date, or example that could replace a general claim here?" If yes, add it. If no, consider whether the paragraph says anything worth keeping.

Assignment

Edit two 500-word pieces side by side: one you wrote yourself (include some deliberate errors) and one AI-generated on the same topic. Track every edit you make in a table with columns: Edit Location, Type of Edit, Time Spent. After finishing both, compare the edit profiles. What types of edits dominated in the human text? What dominated in the AI text? Write a one-paragraph summary of the difference. This is the editorial skill you are developing.

Editing AI vs Human Output