# Boldface in running prose Part of [The AI Tells Index](https://feedsquad.com/ai-tells). A tell signals low effort. It does not identify an author. Skilled writers produce every shape listed here, some of them daily, and automated detectors misread those writers for it at rates measured above 60 percent on non-native English prose. Nothing in this index proves that a machine wrote anything. Read an entry as one piece of evidence to weigh against the false-positive notes printed beside it. ## Facts - Id: boldface-in-running-prose - Category: Formatting (formatting) - Subcategory: emphasis - Also known as: WP:AIBOLD, key-term bolding, emphasis carpet-bombing - Status: Active. Currently signals low-effort writing. - Severity: medium - Evidence grade: primary-doc (platform policy, vendor documentation, or a model card) - Languages: en - Added: 2026-08-15 - Updated: 2026-08-17 - Page: https://feedsquad.com/ai-tells/boldface-in-running-prose ## Description Chosen terms bolded through ordinary paragraphs with no list anywhere near them. Every appearance of a phrase gets the same treatment, so the emphasis stops marking anything. Bolded openers inside list items belong to a separate entry and the specimens are kept apart. ## Why it reads as machine-written Wikipedia files the habit under WP:AIBOLD and traces it to readmes, fan wikis, how-tos, sales pages, slide decks and listicles in the training material, all of which bold heavily. Anthropic's dated system prompts instruct Claude away from over-formatting, which is a vendor writing down the pull it wants suppressed. Count rather than judge. One bolded term in a long piece is editing. ## Detection Type: statistical (a measured metric against a threshold with a stated basis) Metric: bold-spans-per-1000-words-outside-lists Threshold: 6 (fires when above) Threshold basis: > No published human baseline for bold density in prose exists in anything this project read. The one measurement located is a suppression experiment on models rather than a human comparison: Freeburg reports markdown features dropping to roughly zero for eleven of twelve models when they are instructed to write prose, while em dashes persist. Six bold spans per 1,000 words outside lists is therefore a FeedSquad review trigger and not a published cutoff. A reading above it means read the paragraph. ## Examples Constructed specimens. Written for this index. Never quoted from anyone's posts. ### Specimen 1 Before, exhibiting the tell: > The **billing migration** was the quarter that taught us the most. **Billing** touches everything we do, so the **migration** was never going to be simple, and the **team** knew that going in. What we came away with is that preparation matters more than pace, which is the sort of lesson most teams end up learning twice. After, repaired: > The billing migration was the quarter's largest project. It touches every subscription we hold, so it had to run without downtime. We rehearsed it twice against a copy of production and hit the same ordering bug both times, which is why the live run took 11 minutes rather than the 40 we had planned for. Note: Figures in this repair are invented for the specimen. The tell is the repeated emphasis, not the emptiness underneath it. The specimen is written in asterisks because a JSON field cannot carry rendered bold, which is why the raw-markdown-leak rule also matches it. ## False positives Who legitimately writes this way. Textbooks, revision guides and exam material bold keywords on purpose, and so do plenty of developer documentation house styles. Product marketers bold the one term a skimmer has to catch. Newsletter and content management templates apply bold to a first clause automatically, so the writer never chose it. A reviewer separates the two by asking what removing every bold span would cost the reader: deliberate key-term bolding marks terms the piece goes on to define, while the habit marks the same phrase every time it appears and defines nothing. ## Model attribution Anthropic documents an instruction against over-formatting in dated system prompts, which is first-party evidence of the pull rather than a measurement of what ships. OpenAI documents that GPT-5 in the API emits no Markdown by default, so one set of weights bolds through one door and not another. Attribute the habit to a surface and a date, never to a family on its own. ## Platform notes - linkedin: The composer has no bold at all, so asterisks publish as asterisks. That artifact is the raw markdown entry rather than this one. - wikipedia: Shortcut WP:AIBOLD. The manual of style already restricts boldface in article text, so the habit reads as a style violation there before it reads as anything else. ## Sources 1. Wikipedia: Signs of AI writing https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing (tier: community; accessed 2026-08-14) 2. Anthropic dated system prompts, release notes https://platform.claude.com/docs/en/release-notes/system-prompts (tier: vendor; accessed 2026-08-14) 3. anti-slop-writing system prompt and pattern list https://github.com/adenaufal/anti-slop-writing (tier: community; accessed 2026-08-14; inverted evidence, cited for what it catalogues rather than what it advocates) 4. vale-ai-tells, 111 machine-checkable rules https://github.com/tbhb/vale-ai-tells (tier: community; accessed 2026-08-14) ## Status history Ids are permanent. A retired tell keeps its id and its page. - 2026-08-15, Active: Active. The vendor evidence is first-party and dated: Anthropic publishes system prompts that instruct against over-formatting, which documents the tendency without measuring output. Wikipedia records the same habit in article space and names the training genres it comes from. The threshold is ours and says so, because no human baseline for bold density has been published. ## License CC BY 4.0. https://creativecommons.org/licenses/by/4.0/ Attribution: The AI Tells Index, feedsquad.com/ai-tells Reuse the data, including commercially. Keep the attribution line. --- Dataset version 1.0.0. Schema version 1. Part of [FeedSquad](https://feedsquad.com). Built by [Herman Foundry](https://hermanfoundry.com) from Levi, Finnish Lapland.