Detect Anthropic Claude content across Sonnet, Opus, and Haiku in a single scan. Claude has a recognisable register, structured bullet headers, careful hedging openings, and a heavier em-dash cadence than other large language models. TextSight reads the prose, flags Claude-shaped sentences with colour-coded highlights, and runs the same scan against ChatGPT and Gemini at no extra step. Free to try. No card.
Anthropic Claude is the second-largest large language model in production use after ChatGPT. Output is shaped enough to be recognisable, but most detectors trained primarily on OpenAI samples underrate Claude content. TextSight is trained on multi-model data and weights Claude-specific patterns alongside ChatGPT and Gemini signals.
TextSight detects all three Claude tiers currently in production. Claude Sonnet is Anthropic's balanced workhorse and the most common source of Claude content in marketing and code documentation. Claude Opus is the heavy-reasoning tier that shows up in academic essays, technical analysis, and long-form research notes. Claude Haiku is the fast tier used for quick rewrites and short-form content. All three share the same stylistic spine.
You do not need to tell TextSight which model produced the text. The classifier reads the prose and flags Claude-shaped sentences, ChatGPT-shaped sentences, and Gemini-shaped sentences in the same pass. Mixed-source documents (one paragraph drafted in Claude, another reworded in ChatGPT) score correctly because each sentence is scored on its own pattern.
Colour-coded sentence highlights point to specific lines that carry Claude markers: structured bullet headers, hedging openings, parenthetical asides, and dense em-dash usage. Reviewers see exactly which sentences drove the score rather than guessing from a single percentage.
Output coming through the Anthropic API at console.anthropic.com, the Claude.ai web interface, or any downstream tool wrapped around the API all carry the same fingerprints. The classifier treats Claude as a model, not as a product surface, so detection works regardless of where the user pasted from.
Claude has its own register. It tends toward measured, thoughtful, and structured prose with longer sentence variance than ChatGPT, but the patterns are consistent enough that a classifier trained on Claude samples picks them up reliably. The most useful tells fall into four families.
Claude reaches for explicit structure quickly. Three to five clearly headed bullet points, numbered sections, or step-by-step framing show up even in prose contexts where a human writer would write flowing paragraphs. The structure is helpful but recognisably uniform across responses, and the headers themselves carry a similar phrasing rhythm from one scan to the next.
Claude opens with phrases like I would like to suggest, I would be happy to help, or Let me think through this with you. Hedging language runs softer and more personal than ChatGPT's institutional It is important to note. The result is prose that reads earnest and thoughtful but lands in a narrow stylistic band that the classifier learns quickly.
Claude uses em-dashes considerably more often than other large language models. A paragraph with three or four em-dashes inside a few hundred words, especially used to insert parenthetical asides mid-clause, is a strong calibration signal. Em-dash frequency on its own is not a verdict, but it sits high in the classifier's feature ranking because it survives light human editing.
Claude's burstiness profile is higher than ChatGPT's, which makes the cadence feel more human at first read. The vocabulary range, however, sits in a narrower band: certain phrases recur (worth noting, it is fair to say, that said) and the classifier reads the lexical fingerprint underneath the sentence-length variance.
Claude often narrates its own thought process: stepping back, I should note, let me reconsider that. The meta-reasoning is helpful in chat but tends to survive into pasted prose unless the user edits aggressively. When it does survive, sentence highlights pick it out immediately.
Pro at $19.99 a month standard, $14.99 a month on yearly, is the right fit for solo editors, instructors, and reviewers running steady individual scans. Business at $39.99 a month standard, $29.99 a month on yearly, fits teams scanning fifty or more pieces a month with shared history and REST API access. Full details on the pricing page.
Billed $89.88/year — Save $30
Billed $179.88/year — Save $60
Billed $359.88/year — Save $120
Yearly billing saves 25%. View full pricing →
Detector disagreement on Claude is common. The first generation of AI detectors trained primarily on OpenAI ChatGPT output because that was the dominant model in 2023. Claude samples were under-represented in those training sets, and the classifiers learned ChatGPT patterns deeply and Claude patterns shallowly.
Detectors trained mostly on ChatGPT output learn the institutional hedging, uniform sentence cadence, and stock transitional phrasing of GPT prose. A Claude paragraph with structured bullets, parenthetical asides, and softer hedging does not light up the same features. The detector reads it as low confidence and returns a human-ish score even when the prose is straightforwardly Claude.
TextSight was trained on samples from Anthropic Claude, OpenAI ChatGPT, Google Gemini, and other large language models. Claude-specific markers, including the em-dash density and meta-reasoning openings, activate the right signals. Cross-model scoring stays calibrated rather than collapsing to whichever model the training set leaned on.
When TextSight reports 85 percent AI on a paragraph and a GPT-tuned detector reports 20 percent, the disagreement is usually a calibration gap, not a contradiction. The two detectors are reading different distributions. Sentence-level highlights make this concrete: a reviewer can point to the specific lines carrying Claude markers and decide whether to act on the signal.
Anthropic retrains Claude regularly and the stylistic distribution drifts. TextSight refits the Claude classifier against fresh samples on a rolling cadence. The page you are reading reflects the current distribution; detection accuracy on new Claude versions stays within the published band.
Claude output appears in three high-volume contexts: academic essays where the structured reasoning and hedging fit a research register, marketing copy where the polished tone reads professional, and code documentation where the step-by-step framing maps cleanly onto technical explanation. Each context calls for a slightly different read of the scan.
Students reach for Claude on essays where the hedged, measured register feels appropriate to academic prose. Instructors reviewing submissions see structured bullet headers in unusual places, meta-reasoning that survived light editing, and a tendency toward perfectly paragraphed five-section essays. Sentence highlights make the pattern explicit, which is more useful in an integrity conversation than a single percentage.
Content teams use Claude for blog drafts, landing page copy, and email sequences because the prose reads polished out of the gate. The same polish is the tell. Em-dash density spikes, parenthetical asides cluster, and hedging openings on subheads recur. Reviewers running a pre-delivery scan catch these before the client does.
Engineering teams use Claude to draft README files, API references, and inline documentation. The structured framing fits, but the prose around the code reads identifiably Claude. Detection here is less about catching academic misconduct and more about flagging documentation that has not been read by a human before publication, which is a separate quality concern.
Claude often handles longer technical Slack messages and internal memos. The hedging and structured framing carry over, which is fine internally but creates problems when those notes get pasted into public-facing pages without editing. A quick scan catches the lift-and-paste case.
A single percentage is not a fix path or an evidence trail. The TextSight result panel surfaces which sentences carried Claude markers and why, with paragraph-level rollups for longer pieces, so reviewers can point to specific lines rather than negotiating headline numbers.
Every sentence is colour-coded by its own AI-likeness score. Red sentences clustered around structured bullet headers and hedging openings are a stronger signal than scattered yellows. The visual makes the pattern legible without forcing a reviewer to study the percentage.
Longer pieces get paragraph-level rollups so reviewers can see which paragraph is dragging the headline score. On Claude content this usually points at intros (which carry hedging) or step-framing sections (which carry the structure). Targeting the lowest paragraph first is the fastest way to confirm the read.
Perplexity measures how predictable word choices are to a language model. Claude prose runs slightly higher perplexity than ChatGPT prose because the vocabulary range is wider, but specific recurring phrases drop the per-sentence number sharply. The diagnostic context helps decide whether a flag is real Claude residue or a well-rehearsed product description.
Burstiness measures sentence-length variance. Claude has higher burstiness than ChatGPT, which is why the cadence feels human at first read. Low burstiness across a passage where the bullet structure and hedging fingerprints fire is a particularly strong Claude signal: the variance dropped because the model was operating in a templated reply mode.
More LLM-specific detection guides.
OpenAI ChatGPT detection with the same multi-model classifier and sentence highlights.
For ChatGPT →The main detector page covering accuracy, methodology, and the multi-model classifier.
Main detector →Light, Balanced, and Maximum modes for editing Claude-shaped passages without losing voice.
Read the guide →Free, Starter, Pro, Business. Yearly billing saves 25%. Solo to team tiers.
See pricing →Free to try. No card. Pro at $14.99 a month on yearly for solo reviewers; Business at $29.99 a month on yearly for detection teams.