The Five Reading Errors We See Most
Most lost marks in this section come from a handful of repeatable habits rather than from gaps in English. Here are the five, and what each one costs.
Error one: answering a question adjacent to the one asked
This is the largest single source of lost marks in CELPIP Reading, and it feels nothing like an error at the time. You read the stem, you read the options, one of them matches something you clearly remember from the text, and you take it. The option is true. It just answers a different question.
It happens because recognition and correctness produce the same sensation. The stem asked why the council delayed the vote; the option gives a real criticism of the project. Both are in the text. Only one is a reason for the delay.
It is also the error most resistant to rereading the passage, which is what makes it expensive. Candidates who get this wrong and then review by rereading the text find their answer again, still true, and conclude the question was unfair.
The fix. After choosing, read the stem again with your option inserted as one sentence, and ask whether that sentence is a claim the text supports about the thing being asked. Three seconds, and it is the highest-return three seconds available anywhere in this section.
Error two: holding a position without its owner
Specific to Part 4 and to parts of Part 1, and it is why those two parts get named as the hardest. You read the article, you understand every argument in it, and by question 3 you are holding four positions with no names attached.
The distractors are built for exactly this. Moving a real opinion onto a different speaker costs the test writer nothing and produces an option that is true, familiar and wrong. The critic's costing objection under the champion's name. The letter writer's partner's reluctance attributed to the writer.
The reason it happens is that tone attaches to a text as a whole. When you read an argument you build a sense of what is being argued, and the ownership of each piece is the first thing dropped, because in ordinary reading it does not matter.
The fix. Write a ledger while you read: name, position in three words, one detail that is theirs alone. Twenty seconds inside a twelve minute part. In Part 1 the equivalent is asking, for every option about a feeling or an intention, whose feeling and whose intention.
Error three: reading in the wrong order for the part
Every part has an opening move, and using the wrong one costs a third of the part before a single question is answered.
The two expensive versions are opposites. Reading a Part 3 passage closely before looking at the statements means holding four paragraphs of unattached detail against nine questions you have not seen, which nobody can do, so the passage gets read twice. Skimming a Part 1 letter and hunting per question means searching for implications that have no printed form, so the letter gets read five times.
The Part 2 version is the most common of all: reading all twelve feature bullets of the chart closely before knowing the person's constraints. The chart is meaningless until you know what you are comparing for, so the detail does not stick and the chart is read again.
The fix. Decide the four opening moves in advance and rehearse them until they are automatic. Read the letter whole for Part 1. Map the chart's shape for Part 2. Label the four paragraphs for Part 3. Build the ledger for Part 4. Four decisions made once, none of them made under a timer.
Reread the stem with your chosen option inserted as one sentence before you commit. This single three second check defeats the largest error category in the section.
Error four: answering the second text from the second text
Fifteen of the 38 questions are blanks inside a second text: the reply in Part 1, the email in Part 2, the comment in Part 4. All three work the same way, and candidates lose the same marks in all three.
The second text was written by someone who has already read the source, and its load-bearing facts have been cut out and replaced with drop-downs. What remains is the shape of an answer. Fill a blank from the sentence around it and you are choosing whatever is most grammatical and most plausible, which is exactly the property the distractors were given.
The symptom is speed. These blanks feel fast. A candidate who is filling them at fifteen seconds each and finishing the part early is almost certainly answering from the second text, and will be surprised by the score.
The fix. For every blank, point at the line in the source that forces it. If you cannot find that line, you have not answered the question yet. Also read the whole second text through before filling anything, because later sentences routinely settle earlier blanks.
For every blank in a reply, email or comment, point at the line in the source text that forces it. If the blanks are taking fifteen seconds each, you are answering from the wrong document.
Error five: overspending on one question
A question resists. You have understood the passage, so it feels solvable, and each additional twenty seconds feels like it is about to pay off. Ninety seconds later you have an answer you are not confident in, and two questions you will now answer badly.
The arithmetic is unambiguous and almost nobody applies it under pressure. Every question is worth the same. The question in front of you is worth one mark whether it takes ten seconds or three minutes, and the two after it are worth one each. Spending three questions' worth of time on one is a straight loss even when it works.
Part 2 is where this is most costly, because the whole part is eight minutes and one ninety second question is a fifth of it. Part 4 absorbs it better and still not well.
The fix. Set a hard ceiling per question, roughly 90 seconds, and treat it as a rule rather than a guideline. When you hit it, eliminate what you can in five more seconds, choose, and move. There is no penalty for a wrong answer, so an imperfect choice costs nothing you had.
Reviewing for errors rather than for score
These five are recognisable in your own work, but only if you review in a way that can show them, and most review cannot. Marking a test and counting the score tells you where you are and nothing about what to change.
A review that works takes about fifteen minutes per section. For every wrong answer, write one line: which of the five it was, or a sixth category if it genuinely does not fit. Do not reread the passage first, because rereading generates a plausible story about why the question was hard and buries the actual cause.
After three sections you will have twenty or so lines and one category will dominate. That category is your preparation for the next fortnight, and it will be a habit rather than a topic. Habits change quickly, which is why this kind of review produces faster improvement than more practice tests do.
One caution. If the dominant category is unfinished questions rather than wrong ones, none of the five applies yet. That is a fluency and pacing problem, and it is fixed with timed single parts rather than with technique.
Cap any single question at about 90 seconds. Every question is worth the same, so three questions worth of time spent on one is a loss even when you eventually get it right.
Practise it
Five questions on the recurring errors and the specific habit that fixes each one.