The mistakes that actually cost CELPIP points
Our score data page shows which criterion comes out lowest. This page answers the next question: what, exactly, was the marker objecting to? We grouped every recurring problem it wrote down and counted them.
Last updated (first published 22 September 2026) · a fixed analysis of the window below, refreshed by hand · How to cite
Key findings
- Four mechanical categories — verb forms (15.0%), articles and prepositions (14.1%), spelling (13.0%) and word choice (10.9%) — are 53% of 1,085 marker notes on CELPIP Writing answers.
- Readability is the lowest of the four Writing criteria in 45–60% of answers at every level from CLB 5 to CLB 9.
- Fluency problems (fillers, false starts, pauses) are 27.1% of 2,633 Speaking notes, and vocabulary problems 19.4%.
- 14.3% of Speaking notes — about one in seven — flag an answer that simply stopped: mid-sentence or with no closing line.
- Median speaking pace rises from 111 words a minute at CLB 5 to 147 at CLB 8.
336 Writing answers from 120 learners (26 July – 28 September 2026); 750 Speaking answers from 116 learners (18 August – 28 September 2026). Practice answers marked by AI, not official CELPIP results.
Method: how this was counted
Our marker writes a short free-text note for each recurring problem it finds in an answer. We took every note it wrote on genuine attempts — at least 60 words and a score of CLB 5 or above, staff and test accounts excluded — assigned each note to one category from a single fixed set, and counted each category. That is 1085 notes on 336 Writing answers from 120 learners (26 July – 28 September 2026) and 2,633 notes on 750 Speaking answers from 116 learners (18 August – 28 September 2026).
Each share below is a share of notes, not of answers: one answer usually contributes several. 19 writing notes and one speaking note fitted no category and are left out of the cards (they stay in the total). Bands are the overall CLB rounded down. These are practice answers marked by our AI on the four criteria CELPIP publishes — not official results, and nothing to do with Paragon Testing Enterprises. And the people here chose to practise online, so they are not a random sample of test takers.
No learner's words appear on this page. Every example phrase in the descriptions is written by us.
Writing: what the marker flagged
Verb form and agreement
15.0%The wrong tense, a missing auxiliary, or a subject and verb that do not match — “every staffs are”, “would be appreciate”.
Articles and prepositions
14.1%A missing or wrong a / an / the, or the wrong preposition inside a fixed phrase — “concerned for the delay”.
Spelling
13.0%Ordinary misspellings, most often in exactly the longer words the writer reached for on purpose.
Word choice and idiom
10.9%A real word used imprecisely, or a pairing no one says. The meaning survives; the score does not.
Unspecific support
8.9%The reason is stated but never made concrete — no date, number, name or example behind it.
Capitalisation
7.1%A lower-case “i”, an uncapitalised proper noun, or a sentence that starts small.
Run-on sentences
6.7%Two complete sentences joined by a comma, or a paragraph that never stops for a full stop.
Transitions and connectors
5.0%No signposting at all, or the same three connectors doing all of the work.
Repetition instead of development
4.2%The same idea returns in new clothes instead of a new idea arriving.
Missing task content
4.0%One required content point left out — often the option you were meant to turn down.
Other punctuation
3.4%Missing apostrophes, stray or missing commas that do not run sentences together, a question without its question mark.
Sentence fragments
2.3%A clause standing on its own without a subject or a main verb.
The four largest categories — verb forms, articles and prepositions, spelling, word choice — come to 53% of every note. Not one of them is about what the writer had to say. The ideas arrive; the sentences carrying them are what gets marked down.
Counted a second way, answer by answer, the same thing shows up: readability was the lowest of the four criteria (a tie counted as a share) in 51% at CLB 5, 51% at CLB 6, 60% at CLB 7, 46% at CLB 8 and 45% at CLB 9 — a majority at the level most Express Entry candidates are aiming past.
Speaking: what the marker flagged
Grammar under time pressure
16.2%The same articles, prepositions, tenses and agreement as in Writing — with no chance to go back and fix them.
Repeated words and sentence frames
15.6%One handful of words and one sentence shape carrying the entire answer.
The answer never lands
14.3%The recording stops mid-sentence, or simply stops, with no closing line at all.
Filler words
11.2%“Uh”, “um”, “like”, “you know” — often enough that the marker counts them against fluency.
False starts and restarts
10.7%A sentence begun, abandoned and begun again, usually while hunting for one word.
Weak organisation
7.3%No sequence a listener can follow, or one connector doing all the signposting.
Idea repeated, not developed
6.3%The same point made twice in different words instead of a second point arriving.
Underdeveloped content
6.2%Reasons listed but never explained, with no example behind any of them.
Pauses and pacing
5.2%Long silences, or a rate so slow the answer never fills the time it was given.
Part of the task missed
3.2%One thing the prompt asked for never arrives — how you felt, what happens next, which option you chose.
Grouped, the picture is stark. Fluency problems — fillers, false starts, pauses — are 27.1% of all notes. Vocabulary problems, counting repeated words, imprecise choices and a narrow range together, are 19.4%. Between them, nearly half of everything the marker noticed, and neither has anything to do with whether the speaker had something to say.
The one worth reading twice is third on the list. In 14.3% of cases the problem is simply that the answer stopped — mid-sentence, or with no closing line. About one problem in seven is running out of time, which is a habit rather than a level of English, and the cheapest thing on this page to fix.
Vocabulary also tightens as the score rises: it was the single lowest criterion in 38% at CLB 6, 39% at CLB 7, 54% at CLB 8 and 58% at CLB 9 — though CLB 9 is only 13 answers, so read that last one as a direction, not a number.
Speaking pace rises with the score
Writing length famously does not separate the bands — our score data page shows longer emails do not score higher. Speaking behaves the opposite way. Because the clock is part of the task, hesitation gets measured whether or not anyone intends it to be.
| Band | Answers | Median words | Median pace |
|---|---|---|---|
| CLB 5 | 113 | 120 | 111 wpm |
| CLB 6 | 272 | 134 | 127 wpm |
| CLB 7 | 256 | 150 | 137 wpm |
| CLB 8 | 95 | 163 | 147 wpm |
| CLB 9 | 13 | 184 | 148 wpm |
Median pace rises by about a third between CLB 5 and CLB 8, from 111 to 147 words a minute, and the amount said in the same 60 or 90 seconds rises from 120 to 184 words by CLB 9. Practising until you stop searching mid-sentence is worth more than practising new content — and it is the same fix as the unfinished answers above.
Pace is the transcript's word count divided by the length of the recording. The first version of this table (22 September) showed 57–104 words a minute; we could not reproduce the time base it used, so it now uses the recording itself, which gives higher figures. The pattern — pace rises with the band — is the same.
Both start here: timing and pacing drills and buying time gracefully.
Frequently asked questions
What is the most common mistake in CELPIP Writing?
Verb form and agreement — the wrong tense, a missing auxiliary, or a subject and verb that do not match. It accounts for 15.0% of every recurring problem our marker recorded across 336 marked Writing answers, ahead of articles and prepositions (14.1%) and spelling (13.0%).
What do CELPIP Speaking answers lose marks for?
Across 750 marked Speaking answers, fluency problems — filler words, false starts and long pauses — are 27.1% of everything flagged, and vocabulary problems 19.4%. A further 14.3% are answers that simply stopped: they ended mid-sentence or had no closing line, which is a timing habit rather than an English problem.
Are CELPIP mistakes mostly about grammar or about ideas?
Grammar and word choice, by a distance. In Writing, the four largest categories — verb forms, articles and prepositions, spelling, and word choice — are 53% of everything the marker flagged, and none of them is about what the writer had to say. Content problems such as unspecific support and missing task points together account for far less.
Which CELPIP criterion holds people back most?
In Writing it is readability: it was the lowest of the four official criteria in 45–60% of marked answers at every level from CLB 5 to CLB 9. In Speaking it is vocabulary, and its share grows as the score rises — 38% at CLB 6, 54% at CLB 8.
Is this data from the official CELPIP test?
No. These are practice answers submitted on Celpip.me and marked by our AI on the four criteria CELPIP publishes. They show where practice answers lose points; they are not official results, and Paragon Testing Enterprises has no connection to them.
Related: writing & speaking score data · reading & listening data · CELPIP Writing · CELPIP Speaking.
How to cite this data
Please do. The figures are published under CC BY 4.0, which asks only for attribution and a link. A line you can paste:
Celpip.me (2026). The mistakes that actually cost CELPIP points [Data set]. 1,085 marker notes on 336 Writing and 2,633 on 750 Speaking answers, updated 28 September 2026. https://celpip.me/celpip-common-mistakes
- URL
- https://celpip.me/celpip-common-mistakes
- Last updated
- 28 September 2026 (window: 26 July – 28 September 2026 for Writing, 18 August – 28 September 2026 for Speaking)
- Licence
- CC BY 4.0 — credit “Celpip.me”
Writing something and need a figure cut a different way? Ask at support@celpip.me — we would rather you had the real number than an estimate.
Find out which of these is yours
Write or speak one answer and get it marked on content, vocabulary, readability and task fulfilment — with the exact sentences that held each one down.
Start practising free →