Grammarly

Grammarly's free web checker got every correction we could read right and never flagged clean text, and it also let the word recieve sail through without so much as an underline. Our limited 3/5 covers five automated, anonymous fixtures on the Free Grammar Checker demo page on September 15, 2026, not Grammarly as a whole, not the signed-in product, not the browser extension and not Pro.

What we tested. T1 planted six errors in two sentences: a subject-verb error (Each of the engineers were), the misspelling recieve, a doubled the the, Its for It's, there for their, and everyone agree. T2 planted six punctuation and capitalization errors: Wheres, a question ending in a period, lowercase monday, a comma splice around however, dont, and a missing final period. T3 was a clean three-sentence control paragraph. T4 planted four context homophones: effect for affect, then for than, loose for lose, and complement for compliment. T5 accepted one suggestion to confirm the documented hover-and-click flow. Expected corrections were written down before anything was typed, and each fixture was typed once into a fresh editor.

The grammar fixture. The Suggestions counter showed 5. Underlines sat on were, Its, there and agree, plus a trailing double period that our own typing automation introduced and the fixture did not contain; that artifact stands in the evidence. The cards we could open said Correct subject-verb agreement: was; Use the right word: It's; Use the right word: their; and Punctuation problem: a single period. All four match the intended fix. The card for agree would not open for our automation, so that replacement is unverified. Nothing at all was underlined on recieve or on the the, and the count stayed at 5 for the roughly five minutes the text sat in the editor. Four of six planted errors were flagged.

The punctuation fixture. The counter again showed 5. Underlines sat on Wheres, report., monday, however and dont. The missing final period drew no flag. None of the cards would open for our automation on this fixture, so we can say which words were flagged but not what the product proposed for them. Five of six planted errors were flagged.

The clean control. The library paragraph produced no underlines and no counter at all. Zero false positives is the result a proofreader must have, and this one delivered it.

The homophones. The counter showed 3. Underlines sat on effect, then and to (the phrase to loose). complement, used where compliment was meant, drew no flag. As with T2 the cards would not open, so the replacements are unverified from our side. Three of four planted errors were flagged, which is a fair result for the page's own claim that it catches correctly spelled words used in the wrong context.

Accepting a suggestion. Our fixture named recieve as the card to accept; since recieve was never flagged, that could not run as written and is recorded as unavailable rather than replaced. As a separate attempt we accepted the was card on were. The editor text changed exactly there, every other character stayed identical, and the counter dropped from 5 to 4. Every card also carried a Sign up for Grammarly link, and the panel holds a hidden Sign up to keep going prompt promising tone and clarity insights after account creation; we did not create an account, so we cannot say what that unlocks.

Where this leaves the reader. Used as a second pair of eyes on a short piece of text, the free checker's suggestions were right every time we could read them and it did not invent problems in clean prose. Used as the only pass, it is not enough: in our tests a common misspelling and a doubled word produced no signal, and neither did a missing full stop or a wrong homophone. Read no underline as not flagged, not as correct. That is why the limited rating is 3/5 rather than 4/5 or 2/5. It is an editorial judgment, not a pass count, an external review average or a Fit Score calculation.

What this review cannot establish. Testing was automated, anonymous and confined to five short fixtures on one day, in an automated Chromium browser at a 1280 x 1280 viewport. The product disclosed an embedded client build (Inkwell 2.12.5-alpha) but no model or plan version, so whether the misses belong to the anonymous demo, the Free suggestion set or the model is unknown. Our typing introduced one artifact in T1, and our automation could not open the suggestion cards on T2 and T4 or on agree in T1. We did not test the signed-in Free account, the browser extension, the desktop or mobile apps, Word or Google Docs integration, Pro features, generative prompts, tone detection, other languages, long documents or repeated runs. These gaps bound the review rather than lower its score. No product-wide rating follows from these fixtures.