Session 11.3: The AI Signature Problem

Course → Module 11: Quality Control & The Human Gate

Session 3 of 7

From Detection to Correction

In Module 1, you learned to identify the 15 forensic markers of AI-generated content. That was diagnosis. This session is surgery. Each marker gets a specific correction method, not a vague "make it sound more human" instruction, but a concrete editorial action you can apply consistently.

The goal is not to hide the fact that AI was involved. The goal is to produce content that meets your quality standard. AI markers are quality problems, not disclosure problems. A hedge phrase is weak writing regardless of whether a human or a model produced it.

The AI Signature: A collection of stylistic, structural, and rhetorical patterns that signal machine-generated text. These markers reduce readability, erode trust, and flatten voice. Eliminating them is an editorial skill, not a deception technique.

The 15 Markers and Their Fixes

#	Marker	Example	Correction
1	The "comprehensive guide" opening	"In this comprehensive guide, we'll explore..."	Start with a specific claim, question, or scenario. No meta-commentary about the article itself.
2	Tricolon abuse	"efficient, effective, and engaging"	Pick the one word that matters. Delete the other two. If all three matter, give each its own sentence.
3	The false bridge	"But here's the thing..."	State the contrasting point directly. The bridge adds nothing.
4	Premature summarization	"In summary" at paragraph 3 of 10	Remove. Summaries belong at the end, if at all. Let the argument build.
5	The hollow metaphor	"Think of it as a Swiss Army knife for your content"	Replace with a specific comparison grounded in the actual subject. If no good metaphor exists, use none.
6	Over-attribution	"According to experts..." (no specific expert named)	Either name the expert and cite the source, or remove the attribution entirely and state the claim directly.
7	The enthusiasm spike	"This is truly game-changing!"	Delete the exclamation mark. Replace the superlative with a specific, measurable outcome.
8	Synonym cycling	"utilize," "leverage," "employ," "harness" in one paragraph	Pick one word and repeat it. Repetition is clearer than variety for the sake of variety.
9	The safety disclaimer	"It's worth consulting a professional..."	Remove unless legally necessary. If legally necessary, move to a footnote or end note.
10	Parallel structure overuse	Every paragraph starts with "When you..." or "By doing..."	Vary sentence openings. Start with a noun. Start with a verb. Start with a dependent clause. Mix.
11	The non-answer answer	500 words that avoid committing to a position	Force a position. "The answer is X, because Y." If no clear answer exists, say so in one sentence.
12	Context-free confidence	"Studies show..." with no citation	Add the specific citation, or remove the claim. Vague authority references are worse than no reference.
13	The empathy prefix	"I understand this can be frustrating..."	Delete. Move directly to the useful information. Performative empathy is transparent.
14	Temporal vagueness	"In recent years..."	Specify the year. "Between 2023 and 2025" is information. "In recent years" is filler.
15	The closing platitude	"The future is bright for those who..."	End with a specific action, a concrete recommendation, or nothing. A strong article does not need a motivational sendoff.

The Correction Workflow

Applying these fixes is not a single pass. It is a layered process, each pass targeting a specific category of markers.

flowchart TD A[Raw AI Output] --> B["Pass 1: Structural Markers
(#4 Premature summary, #10 Parallel structure, #11 Non-answer)"] B --> C["Pass 2: Voice Markers
(#1 Generic opening, #3 False bridge, #7 Enthusiasm, #13 Empathy prefix, #15 Platitude)"] C --> D["Pass 3: Precision Markers
(#2 Tricolon, #5 Hollow metaphor, #6 Over-attribution, #8 Synonyms, #12 Context-free, #14 Temporal vague)"] D --> E["Pass 4: Cleanup
(#9 Safety disclaimers)"] E --> F[De-Slopped Output] style A fill:#c47a5a,color:#111 style B fill:#c8a882,color:#111 style C fill:#c8a882,color:#111 style D fill:#c8a882,color:#111 style E fill:#8a8478,color:#ede9e3 style F fill:#6b8f71,color:#111

Why This Order Matters

Structural fixes come first because they change the shape of the text. Fixing a non-answer might mean rewriting an entire section. If you fix voice markers first and then restructure, you lose your voice edits.

Voice markers come second because they affect tone across the whole piece. Once the structure is solid, you can hear the voice problems more clearly.

Precision markers come third because they are surgical. Replacing a vague attribution with a specific one does not affect structure or voice. It is a scalpel operation.

Safety disclaimers come last because they are the simplest: keep or cut.

The Economics of Correction vs. Regeneration

If you find more than 8 of these 15 markers in a single piece, regenerate. The cost of fixing 8+ markers exceeds the cost of a new generation with a better prompt. If you find 3-7, correct. If you find fewer than 3, your prompt engineering is working and you only need a light editorial pass.

Track your marker counts over time. A decreasing count means your prompts are improving. A stable count means your prompts have plateaued and need revision.

Assignment

Take the AI Detection Checklist you created in Session 1.6. Add a "Correction Method" column for each marker based on this session's guidance. Then take a real AI-generated article and apply the four-pass correction workflow. Count how many markers you found in the original. Count how many survive after correction. If more than 8 survived your prompt, revise the prompt and regenerate instead.

The AI Signature Problem