The Load-Bearing Tic, Now With a Growth Chart

For a year, load-bearing has been a punchline about Claude's writing, but this week someone put a number on it. A developer named Louis Abraham sorted 461,000 GitHub pull request descriptions by vocabulary alone and found one writing style, the one built around calling everything load-bearing, climbed from under one percent of the corpus in early 2025 to thirty-nine percent by mid-2026. See the chart on our trend board.

None of this means Claude invented the phrase; humans use load-bearing too, just far less often and less predictably. What is new is that developers have started fighting back with code. Guides now circulate CLAUDE.md bans and prompt hooks built specifically to suppress the word, alongside older workarounds for Claude's habit of calling every function the heart of something (AIToolCrunch, Jul 2026). Naming the tic in a system prompt sometimes backfires, since it can make the model reach for a synonym instead of writing plainly.

Elsewhere the models kept moving. Wikipedia's own guide to AI writing signs now credits OpenAI's GPT-5.1 as the chatbot most notably tuned to cut back on em dashes, while separate OpenAI release notes describe reworking GPT-5.5 Instant specifically to produce fewer long, bullet-heavy answers and adjusting GPT-5.3 Instant to drop teaser lines like you'll never believe. Each fix quietly retires one tell and, often, opens space for the next.

None of this changes the basic advice for a working writer. A single load-bearing or bulleted list proves nothing on its own, but density does. If your editor of choice is Claude, expect the tic more than with other assistants; if it is ChatGPT, expect shorter, less bulleted answers than a few months ago. Run a suspicious paragraph through our draft checker or the fix-your-ai guide before you publish.