The numbers behind AI writing

Sourced data pages on AI writing: every figure carries its source and its date.

• The measurable markers of AI-generated text
  A referenced catalogue of 16 measurable markers of AI-generated text, from focal words to formatting tics, with the evidence behind each and its limits.
• The words ChatGPT overuses, with the evidence
  The AI words list with receipts: corpus studies, magnitudes, and per-generation dating, because tells decay as models change. Last checked August 2026.
• Are AI detectors accurate? Every published number
  Are AI detectors accurate? On unedited AI text, sometimes; on edited text and on non-native writers, much less. Every published accuracy and false positive figure in one sourced table: the corpus, the number, and who funded each test.
• What stylometry research actually shows
  What peer-reviewed stylometry shows about measuring writing style: which features carry authorship, the accuracy shared tasks actually report, how far it falls when genre changes, and how much text it needs.
• The anatomy of a writing voice: 45 measured attributes across 8 layers
  The complete taxonomy ScriptGrain measures a writing voice with: 45 named attributes across lexical, syntactic, tone, rhetorical, punctuation, function-word, content and cadence layers, each with its unit and how it is extracted.
• How writing voice is measured: the Voice Match method
  The published method behind ScriptGrain's Voice Match score: which features are counted in code, which are judged, the weights and tolerances, the calibration from raw agreement to the number you see, and where the score is unreliable.
• Brand voice, measured: the 45-attribute framework for marketing teams
  How a brand voice is measured rather than described: 45 attributes from the brand's own published writing, a score on every draft, a threshold for the team, and what a written style guide can and cannot enforce.
• The lexical layer: which vocabulary metrics identify a writer
  The six vocabulary attributes ScriptGrain measures (diversity, preferred words, word length, rare words, fillers, contractions), what 299 human pieces look like on each, and how to read your own numbers.
• The syntactic layer: sentence-structure metrics that separate two writers
  The seven sentence-level attributes ScriptGrain measures (length, variance, complexity, clause order, brackets, dashes, semicolons), the reference-corpus distribution for each, and why sentence rhythm is the first thing AI drafts lose.
• Tone and register, measured: the attributes behind 'that sounds like us'
  The six tone attributes ScriptGrain measures (formality, humour register, expressiveness, contraction rate, confidence versus hedging, audience adaptation), how a model judges five of them and code counts the sixth, and why a prompt can describe register but not hold it.
• Punctuation and formatting as a fingerprint: dashes, commas, list habits
  The punctuation and format attributes ScriptGrain measures (commas, exclamation marks, ellipses, questions, capitalisation, lists, plus dashes and semicolons), what 299 human pieces look like on each, and the one mark that gives AI drafts away.
• British versus American English: the stylistic markers you can measure
  The spelling families, vocabulary, punctuation and grammar habits that separate British from American English, which of them a machine can check, and how ScriptGrain sets the variant on a profile rather than asking for it in a prompt.
• AI-isms in marketing copy: 30 phrases and 10 sentence shapes, counted per 1,000 words
  The AI-isms list, counted: 30 phrases in four groups and 10 sentence shapes, the density bands, what 299 human pieces score, and what two AI arms scored in a published study.
• Brand voice guidelines for marketing agencies: a template with measurable criteria
  A tone of voice guidelines template (brand voice guidelines) where every section carries a criterion a machine can check: register, sentences, contractions, punctuation, person, vocabulary, structure, humour, threshold and English variant, filled from your own or a client's writing.
• AI writing voice tools compared: pricing, voice measurement and API access (checked monthly)
  Twenty AI writing and brand voice tools in one table: entry price, free plan, whether a numeric voice score exists, whether the scoring method is published, public API and MCP server access, each with the date it was checked.
• Best brand voice tools for agencies in 2026: what each one measures per client
  Nine brand voice tools compared for agency use: per-client voice profiles, seats, whether drafts are scored against the client, API and MCP access, and what ten client voices cost, with check dates.
• Which AI writing tools have a public API and an MCP server: access by plan
  Fourteen AI writing tools checked for a public REST API and an MCP server, the plan each needs, how they authenticate and what an assistant can actually call, with the date each was read.
• Voice profile JSON schema: 45 attributes across 8 layers, as the API returns them
  The shape of a ScriptGrain voice profile as GET /v1/profiles/{id} returns it: status, confidence, narrative and the 45-attribute object, field by field with JSON types, units and allowed values.
• ScriptGrain MCP tool reference: every tool the server exposes
  All 41 tools on ScriptGrain's MCP server (mcp.scriptgrain.com/mcp): what each does, what it costs, its REST equivalent, the two prompts, how to connect from Claude, ChatGPT, Cursor, Gemini CLI and Muse, and a typical session.
• ScriptGrain API on the free plan: rate limits, quotas and what each tier adds
  What an API key can do on each ScriptGrain plan, from Free upward: the 60-a-minute rate limit, extraction and mimic limits, credits per plan, what costs credits and what is free, and how idempotent retries work.
• Brand voice API: endpoints, fields and the score that comes back
  ScriptGrain's brand voice API end to end: build a profile from samples or a public site, generate in it, score any text with POST /v1/voice-match and read the per-feature deltas, with the request fields, error codes and limits.
• How to evaluate a brand voice AI tool: a twelve-point checklist with test prompts
  Twelve questions to put to any brand voice AI tool before buying, each with a pass condition, plus three test prompts to run in every tool on the shortlist and a way to score the result that does not depend on a demo.
• How do AI detectors work? The four methods, and why each one misfires
  AI detectors work in one of four ways: they measure how predictable the text is to a language model, compare it with a second model, run a classifier trained on human and AI writing, or look for a watermark. How each works, from the papers, and where each fails.
• Best AI humanizer tools in 2026: what each one changes, and what it costs
  Eight AI humanizers compared from their own pricing and product pages, read on 2 October 2026: price, free allowance, detector claims, and whether any of them writes in your voice. Plus what the research says about passing AI detectors.
• Content governance for written content: a checklist for voice, claims and repetition
  Content governance for writing: the checks every piece should pass before it goes out (voice, facts and claims, repetition, AI use, ownership, review dates), who owns each one, and which of them a tool can measure for you.
• A content library for written content: what to keep, and where it should live
  A content library for writing keeps more than files: the pieces, the drafts and their versions, the briefs, and what your writing has already said. What each kind of tool keeps (CMS, docs, content platforms, voice libraries), and how to set one up that your AI can use.