Content MarketingTrendingmedium impact

Thomas Ptacek's Rule: Use LLMs as Copyeditors, Not Writers

By InnovaAI Research2 min readSimonwillison

Developer and writer Thomas Ptacek published a guide in September 2026 arguing that LLMs should function as copyeditors rather than content authors. His core rule forbids accepting any specific word or phrase an LLM suggests, treating the constraint as intellectual self-protection.

Key Facts

01Thomas Ptacek's September 2026 post establishes a single hard rule: no word or phrase suggested by an LLM may appear in the final copy.
02He frames this constraint as 'intellectual personal protective equipment,' not stylistic preference.
03The framework positions LLMs as copyeditors that review human-written drafts rather than co-authors that contribute language.
04Accepting LLM phrasing causes writing to converge on recognizable AI cadences, eroding client voice differentiation.
05Agencies can apply the rule immediately by updating content briefs and internal style guidelines.

Why does this matter for agencies?

▶Content convergence is a measurable business risk: clients and audiences increasingly recognize and discount AI-patterned writing, which undermines the value agencies charge for.
▶The copyeditor model shortens revision cycles without sacrificing voice, offering a productivity gain that does not trade off quality.
▶A clear, auditable policy ('no LLM phrasing in final copy') is easier to enforce across distributed writer teams than vague guidance about 'appropriate AI use.'
▶Search systems and content quality signals are increasingly sensitive to AI-generated patterns, making voice differentiation a ranking and engagement factor, not just a brand preference.

What should agencies do?

Rewrite content production briefs to specify that LLMs may only be used to critique completed human drafts, not to generate or suggest copy, and distribute the updated brief to all writers by end of month.

low effort

Audit a sample of AI-assisted content from the past quarter by reading it aloud across multiple client accounts to identify voice convergence, then flag articles that require rewriting.

medium effort

Configure voice and tone controls in AI writing tools such as Jasper or Anyword to reflect each client's established style guide, so any AI feedback is filtered through client-specific parameters rather than generic defaults.

medium effort