GPT-6 Astra Brings Bullets Back, Even as ChatGPT Cuts Them

OpenAI's own developer documentation for GPT-6 Astra says the model tends to use lists, tables, and Markdown to make responses scannable by default. That habit is exactly what a sibling model, GPT-5.5 Instant, had been tuned away from earlier this summer. Prompting guides published in the last two weeks now recommend banning stock phrases and capping bulleted lists at twice per document just to get plain paragraphs out of Astra, a sign that writers, not the lab, are doing the tuning this time. See how the two models compare on our trend board.

Wikipedia's guide to spotting AI writing is quietly reconsidering one of its oldest signs. A tracking effort following the page's September revisions found a new banner proposing to move em dashes out of the current tells and into a section for historical indicators, since a July study found only Claude still overuses the punctuation while ChatGPT has fallen below the human rate. The debate has not settled yet, but it confirms what our own draft checker has shown for months: the dash is now a single-vendor habit, not an industry one.

None of this means every bulleted answer or dash is machine-made; plenty of human writers reach for both. What is useful is noticing where the defaults sit before you edit. If you are drafting with Astra, expect more lists and bold headers than you asked for; if you are drafting with Claude, expect the occasional stray dash and a metaphor stretched past its usefulness. Run a suspicious paragraph through our draft checker or the fix-your-ai guide before you publish.

Elsewhere the numbers held steady. GPTZero's AI Vocabulary page and EQ-Bench's Slop Score page are both unchanged from last week's data pull, still built on the same human baseline and multiplier figures cited all month. A quiet week for one tracker is not a quiet week everywhere else, which is the whole reason this board checks in weekly rather than trusting a single snapshot.