readability

Cut passive voice

MetricSpot estimates the share of sentences in passive voice. Active voice is shorter, clearer, and converts better. Keep passive under 10%.

What this check does

Counts the share of sentences written in passive voice — a form of “to be” plus a past participle, where the agent (the person doing the thing) is either after “by” or missing entirely. The check fails above 10% of all sentences.

Why it matters

Passive voice is grammatically fine. The problem is that it makes prose harder to read in three measurable ways.

  • Longer sentences for the same idea. “The migration was completed by the platform team” is 28 characters and three concepts longer than “The platform team finished the migration.” Multiply by every paragraph.
  • Hidden subjects. “Mistakes were made.” Who made them? Passive voice drops the agent, and readers either guess wrong or stop trusting the writer.
  • Slower comprehension. Eye-tracking studies (Nielsen Norman, plainlanguage.gov citations) consistently show that active-voice prose is parsed faster and recalled more accurately.

The 10% threshold is a heuristic — Hemingway uses 10%, Yoast uses 10%, the US plain-language guidelines push for under 5%. Anything above 20% reads as bureaucratic or evasive.

How to fix it

Find the verb, find the doer, put the doer first.

Pattern: “X was done by Y” → “Y did X”

Before: The dashboard was redesigned by the product team in Q3.
After:  The product team redesigned the dashboard in Q3.

Pattern: “It is believed that…” → just say it

Before: It is believed that response time has improved.
After:  We measured a 40% drop in response time. (or: Response time dropped 40%.)

Pattern: “X is considered…” → “Most X consider…” (or just remove the hedge)

Before: A four-hour SLA is considered industry standard.
After:  Most providers offer a four-hour SLA.

Pattern: When the agent is unknown or unimportant — keep it passive

The server was rebooted at 3 a.m.

If you don’t know who rebooted it and it doesn’t matter for the reader, passive is correct. The 10% threshold gives you room for these legitimate cases.

The “by zombies” test. A linguistics tip from Rebecca Johnson: if you can add “by zombies” after the verb and the sentence still works, it’s passive.

  • “The audit was completed [by zombies].” ✓ Passive.
  • “Zombies completed the audit.” ✗ Active.

Tools.

  • Hemingway Editor (web, free) — highlights passive sentences in green. Aim for the count widget to read “X uses of passive voice” with X ≤ 10% of total sentences.
  • Grammarly / ProWritingAid — same detection plus inline rewrite suggestions.
  • writegood / retext-passive — Node CLI tools. Add to your build / CI to fail Markdown PRs that exceed a threshold.

The exceptions. Scientific writing, lab reports, legal disclaimers, and incident postmortems often use passive intentionally to focus on the thing not the actor (“The sample was heated to 60 °C”, “The outage was contained at 14:32”). MetricSpot’s 10% threshold accommodates this. If your site is mostly disclaimers, the threshold may not be the right tool — but for marketing pages, blog posts, product copy, support docs, it’s the right floor.

Frequently asked questions

Does this rule apply to all languages?

The detector understands English passive constructions. For Spanish (fue + participio), French (a été + participe), German (wurde + Partizip II), Italian (è stato + participio), and Portuguese (foi + particípio), MetricSpot uses language-specific patterns from each locale’s NLP rules. The 10% threshold is the same.

I write technical docs and need passive sometimes.

Use it when the thing matters more than the actor — incident postmortems (“The pod was OOM-killed”), security writeups (“The token was leaked via the URL”), or scientific procedure. The 10% threshold gives you headroom; the rule fails above it, not at it.

What’s the difference between this and “Long sentences”?

Long sentences flags individual sentences over 25 words. Passive voice is independent of length — a short passive (“Cookies are set.”) still counts. Both can fail on the same page for different reasons. See also Flesch–Kincaid grade level for the overall readability score.

Sources

Last updated 2026-05-11