Claim
AI cannot credibly ghostwrite because all users who draw on the same model converge toward the same voice. Embedding idiosyncratic personal markers before using AI as a refiner preserves authorial distinctiveness in a way that fully AI-generated prose cannot.
Mechanism
Ghostwriters and editors serve opposite functions. An editor suppresses their voice to amplify the author's. A ghostwriter blends their voice with the author's. Because AI systems are trained on the same corpus and optimized toward the same outputs, all writers using the same model converge toward the same prose patterns. Readers calibrated on large volumes of AI-assisted writing detect this convergence even without tools. Tunguz names this the age of slop: output that is competent but distinguishable precisely because it lacks the irregularities that mark individual authorship.
The fix is not avoiding AI. It is sequencing the work. The author writes first, embedding their idiosyncratic markers. AI then refines without replacing. Markers that make writing detectably yours are the same markers that make it immune to AI-slop detection.
Conditions
Holds when: the author needs their individual voice to build audience trust, maintain platform differentiation, or avoid detection by AI-trained readers. The effect is strongest in high-volume content environments where AI-assisted writing has already standardized the baseline.
Fails when: the content type does not require a distinctive individual voice (reports, documentation, structured output), or the audience has no pattern recognition for AI-generated prose. Also fails when platform labeling requirements supersede voice authenticity as the primary compliance goal.
Evidence
"The problem with AI as a ghost writer: everyone uses the same ghost writer."
"In the age of slop, readers test authenticity."
"AI is a poor ghost writer, but a great editor."
Tunguz describes his own practice: compose first, then use AI for editorial refinement. The practical fix includes unconventional punctuation (ampersands), neologisms, grammatical fragments, and deliberate stylistic mimicry of specific authors. These choices embed variance that survives AI polish and signals authentic human authorship.
Signals
- Your AI-assisted posts and fully AI-drafted posts produce indistinguishable engagement rates, indicating readers cannot separate your voice from the model's.
- You can predict the first sentence of your AI-generated drafts before reading them.
- Competitors' content on the same topic is indistinguishable from yours even topic-by-topic.
Counter-evidence
The distinction between ghostwriting and editing is not universally clean. Executive ghostwriting and brand voice work require convergence toward a defined style, not individual distinctiveness. In those cases AI's convergent output is the feature, not the limitation. The claim also assumes readers can detect AI convergence, which varies sharply by audience and platform. Platform-level AI labeling requirements would change the calculus: if all AI-assisted content must be disclosed, voice fingerprinting becomes less critical than compliance.
Cross-references
- Claude Sonnet 4.6 makes LinkedIn ghostwriters obsolete by replacing prompts with agents.: the cost-and-replacement angle on AI and ghostwriting; Tunguz's card is the complementary voice-and-authenticity angle.
- Five AI agents replace LinkedIn ghostwriters for $1500/month.: five AI agents replacing LinkedIn ghostwriters on volume; the limitation Tunguz identifies is why volume-produced AI content converges and loses competitive distinctiveness.