AI Often Makes Writing Worse (Even if It Makes It Better)

Britain analysis-side companion to the cover Leader Dont Let AI Kill the Author (Economist) and to the reporting-side AI-Written Speeches Are Taking Over Politics (Economist). This piece is the prose-quality leg of the three-part arc: the mechanism by which AI-drafted text corrodes the standard-setting layer even as it raises the median.

The load-bearing datapoints

"Writing using artificial intelligence is everywhere. It has crept into elite newspapers (the Wall Street Journal ran an op-ed drafted by it) and elite publishing houses (**Grasset in France* finds itself defending a star author accused — falsely, he claims — of using it in a prize-winning novel). It is in the speeches of British politicians, in the homework of British students and abundantly there in LinkedIn posts (41% of longer ones, according to a study by Pangram, a detector, use it). Ask Gmail to draft a love letter to your partner and it will instantly offer up 'My Dearest [Husband's Name]…'. It is quick. It is not quite Keats."*

Four specific first-in-vault datapoints:

1. 41% of longer LinkedIn posts detected as AI-drafted (per Pangram study). Vault's first specific chamber-share number outside of Parliament — the LinkedIn number is the professional-content-canonical anchor.
2. Wall Street Journal ran an op-ed drafted by AI — first-in-vault named elite-newspaper AI-drafted op-ed instance.
3. Grasset (French elite publisher) defending its star author against AI-authorship charge in "a prize-winning novel". First-in-vault named French elite-publishing AI-authorship dispute.
4. Gmail default-behaviour verbatim: "My Dearest [Husband's Name]…" — the tell of an LLM that hasn't been given the specific target's name and defaults to a template variable it should have interpolated. The vault's first specific LLM-tell-from-a-major-consumer-product anecdote.

The sub-head: "Why is complicated. But it's not about the damn em-dash." — the piece pushes back against the popular em-dash-as-AI-tell meme; the real signal is deeper.

The editorial move

The title captures the load-bearing tension: AI often makes writing worse (even if it makes it better). Median writer → LLM-augmented prose is clearer, better-structured, more polished. Elite writer (op-ed / novelist / professional persuader) → LLM-augmented prose is worse because it homogenises voice, defaults to generic frames, and defeats the "only you could have written this" signal that anchors persuasion.

This is the same shape the cover Leader identifies: the median rises, the right tail collapses, and the right tail is where the standard is set.

Connects to AI Writing Detection

Adds the specific-share anchors:

• 41% of longer LinkedIn posts (Pangram)
• Elite-publication instances (WSJ op-ed, Grasset novel dispute)
• Consumer-default-behaviour tell (Gmail "My Dearest [Husband's Name]…")

Cross-refs

• Dont Let AI Kill the Author (Economist) (cover Leader)
• AI-Written Speeches Are Taking Over Politics (Economist) (UK Commons reporting)
• AI Writing Detection (concept updated)
• Pangram (first-in-vault detection tool)
• Grasset (first-in-vault French publisher — held for second-touch)
• Wall Street Journal (first-in-vault named AI-drafted op-ed instance — held for second-touch)