Spot the tells of machine-written prose
AI-generated text has a recognizable signature: too much passive voice, hedges stacked on hedges, filler transitions, and sentences that all open the same way. This checker scans your text for those exact patterns plus standard issues like comma splices and run-ons, so you can revise output into something that reads like a human wrote it.
How it works
The tool runs a set of rule-based heuristics entirely in your browser. A regular expression looks for be-verb plus past-participle pairs to flag passive voice; a dictionary of hedge words and known AI filler phrases highlights weak or machine-sounding language; a clause-joining pattern catches likely comma splices; and a frequency pass on sentence openers flags monotone structure. It also notes very long sentences and double spaces. Each finding shows the issue type, a short explanation, and an excerpt so you can locate it fast.
What each check targets — and why it matters
Passive voice overuse. A sentence like “The report was reviewed by the team” hides the actor and adds words. Models default to passive when uncertain, and heavy use creates a detached, bureaucratic tone. The checker flags be-verb + past-participle pairs so you can convert them selectively — some passives are correct and intentional; many are just habit.
Hedging chains. Phrases like “It is worth noting that,” “It is important to consider,” and “One might suggest that…” compound across a model response until every claim is qualified into meaninglessness. The checker flags known hedge words and stacked phrases so you can convert them to direct statements.
AI filler phrases. “Delve into,” “certainly,” “of course,” “as an AI language model,” “it’s crucial to” and similar expressions appear at far higher rates in LLM output than in human writing. Each one is a minor signal; a paragraph with five is clearly machine-written.
Repetitive sentence openers. Three consecutive sentences starting with “The” or “This” creates a mechanical rhythm. The checker counts opener frequency and flags starters that appear three or more times in succession.
Comma splices and run-ons. Models occasionally fuse independent clauses with a comma where a period, semicolon, or conjunction belongs. The heuristic looks for patterns like ”…, it is…” and ”…, they are…” to catch the most common cases.
Common revision moves
| Issue | Bad (model output) | Better |
|---|---|---|
| Passive | ”Mistakes were made" | "We made mistakes” |
| Hedge chain | ”It is worth noting that this may" | "This tends to” |
| AI filler | ”Let’s delve into the details" | "Here are the details” |
| Repeated opener | ”The system. The process. The output.” | Vary: “Once processed… Output from…” |
Tips and notes
Treat the output as a revision checklist, not a verdict — the passive-voice and comma-splice checks are heuristic and will occasionally misfire on legitimate sentences. The highest-value fixes are usually the warnings: too many passives, repeated openers, and stacked hedges. Because everything runs locally, you can paste confidential drafts without them leaving your browser. Pair this with a human read-through; the checker surfaces patterns quickly, but only you can judge whether a flagged sentence actually needs changing.