Published 1 May 2026 · Updated 3 September 2026 · 15 min read · Cross-provider verbal reasoning hub
| What it measures | Ability to evaluate written information, identify what's stated vs implied vs unsupported |
|---|---|
| Common formats | True / False / Cannot Say · synonyms / antonyms · sentence completion · critical reasoning |
| Used by | Every major aptitude provider — and the standard for legal, policy, and analyst roles |
| Most-failed question type | "Cannot Say" vs "False" — the distinction that costs most candidates marks |
Verbal reasoning tests measure how precisely you read text and how rigorously you evaluate what's stated, implied, and unsupported. The most common format — True / False / Cannot Say — looks deceptively simple but has the highest pass-rate gap of any cognitive test format: the gap between trained and untrained candidates is enormous because the test rewards a specific reading discipline that most people don't naturally have.
Almost every mark lost on a true/false/cannot say item goes on one boundary, and it is not the one candidates revise. It is the line between False and Cannot Say. Most people collapse the two into a single bucket meaning "not true", and the answer key never does. False means the passage says otherwise; Cannot Say means the passage does not say. That distinction is the scoring mechanism, and it is why fluent readers come out average.
Underneath it is one rule: the passage is the entire universe of admissible facts. What you know about employment law, shipping or immunology counts for nothing unless the words on the page put it there. A statement you are certain is true in the world, but which the passage never touches, is Cannot Say. Marking it True because it is obviously correct is the commonest error in the format.
Practising for a real assessment? The Verbal Reasoning Practice Pack runs these question types at full length — 5 timed tests, 130 questions, a worked explanation on every one.
One label covers at least six mechanically different item types, differing in how many responses you pick from, whether outside knowledge is admissible, and whether an item carries one judgement or two.
The response set is fixed at three and identical for every item, so nothing is authored as a distractor; the difficulty sits entirely in where the statement falls relative to the text. True requires the passage to entail the statement, False requires it to entail the negation, and Cannot Say is everything else. It is correct far more often than candidates expect, because most statements about a short passage are simply unaddressed by it. The three responses are not a confidence scale: Cannot Say is a claim about the passage, not a weaker False.
Four or five authored options, one keyed. The standard shifts from entailment to best supported, and that is the mechanical difference: several options may be consistent with the passage while only one answers the question asked, so the work is comparative rather than absolute. There is usually no cannot-say response, so a candidate trained only on three-way items hunts for a certainty the format does not require.
The stimulus is an argument, not a description, and you are asked what it assumes, what would weaken it, or which conclusion it supports. Mechanically you work on structure, premise and unstated assumption and conclusion, not on matching facts to text. That inverts the three-way discipline: an assumption is by definition absent from the passage, so absence signals the answer rather than ruling it out. Negation is the tool; if denying a candidate assumption breaks the argument, it was assumed.
A word is picked out of the passage and you choose its nearest meaning as used there. This is the one format where outside knowledge is not merely admissible but required: you must already hold the word's senses, and the passage only narrows among them. Distractors are typically its other genuine dictionary senses, so answering from memory of the word rather than from the sentence reliably picks one.
Gap-fill items turn on connective logic, whether the missing word signals contrast, consequence or concession. Text ordering hands you scrambled sentences to rebuild into a coherent paragraph. Both score cohesion rather than inference, so the cues are pronouns, definite articles and discourse markers, not facts. Ordering also differs in scoring: some publishers mark the whole sequence as a single item, so one misplaced sentence loses everything, while others credit each correctly adjacent pair.
A passage is followed by two numbered conclusions and a five-way response set: only I follows, only II follows, both follow, neither follows, or either I or II follows. You make two independent judgements and map the pair onto one response, so one misread conclusion sinks the item and no partial credit exists. The "either or" option covers complementary pairs where the passage guarantees exactly one holds without fixing which.
Two scoring models are in use. Under raw scoring you gain a mark for each correct answer and lose nothing for a wrong one, so leaving anything blank is strictly worse than answering. Under formula scoring, often called negative marking, a fraction of a mark is deducted for each wrong answer while unanswered items cost nothing. Where a wrong answer is penalised and a blank is not, a guess made without reading the passage is a cost you are choosing to take on.
The three-way response set complicates that more than it does elsewhere. A blind pick lands one time in three, better odds than a five-option comprehension item, and candidates assume this makes guessing safe. Formula scoring is normally calibrated to the number of options, so the deduction on a three-way item is larger and the advantage disappears. What survives is elimination: rule out True on solid textual grounds and a choice between False and Cannot Say is worth submitting under either model.
Timing models change what a passage is worth. A fixed item count against one clock lets you bank time on short passages. An indefinite bank, where nobody reaches the end, punishes rereading; the item you polish costs you two you never saw. Per-section timers stop you moving time between formats. Per-question timers remove skipping, which bites here because items cluster on shared passages, so skipping means abandoning reading you have already paid for. Adaptive versions weight early items heavily and rarely allow review.
Sitting it soon? TestSolve reads each question off your screen and shows the worked answer, over 90% correct on our own practice tests.
Passage: "The Brantwood depot began collecting food waste separately in March. Tonnage sent to landfill from the depot has fallen since. Residents answering the council survey rarely reported difficulty with the new caddies. A second round of caddy distribution has been costed but has not yet gone to committee."
Judge each statement.
1. Landfill tonnage from the depot was higher before March than after it. True. The passage states tonnage has fallen since March, which entails the comparison. 2. No resident reported difficulty with the new caddies. False. "Rarely reported" entails that some did. The passage contradicts the statement, so this is not Cannot Say. 3. Separate food waste collection caused the fall in landfill tonnage. Cannot Say. The passage reports two things happening in sequence and says nothing about one producing the other. 4. The council will run a second round of caddy distribution. Cannot Say. Costing establishes a present state; it fixes nothing about a future decision or an intention. 5. Food waste sent to landfill produces methane. Cannot Say. True in the world and entirely absent from the passage, which makes it inadmissible.
Each wrong answer comes from one identifiable misapplied rule. Marking 2 as Cannot Say treats "rarely" as vague rather than as a quantifier with entailments. Marking 3 as True reads sequence as causation. Marking 4 as True reads a costed option as a decision. Marking 5 as True imports world knowledge. Marking 1 as Cannot Say is the opposite failure, refusing an inference the words force.
Drill the False versus Cannot Say boundary on its own rather than working whole papers. For every statement you mark Cannot Say, write the sentence the passage would have needed for it to be False. If the passage does not contain that sentence, you were right; if you find yourself quoting text that already contradicts the statement, the answer was False.
Mark the quantifiers on first read: all, none, some, most, only, never, rarely, always, and learn which way each entails. "Rarely" entails "sometimes", so "never" is contradicted and the answer is False. "Some" does not entail "not all", so a passage saying some managers agreed leaves "all managers agreed" at Cannot Say unless a counter-instance appears. Getting those two the right way round removes most errors of this kind.
Build a reflex for two verb classes. Causal verbs, caused, led to, resulted in, and intent verbs, plans, intends, will, aims, are Cannot Say candidates unless the passage uses a verb of the same class. Passages describing states and sequences almost never license them.
For the two-statement variant, decide each conclusion separately and write it down before reading the five combined responses, which are designed to pull a settled judgement out of place.
It means the passage neither supports the statement nor contradicts it. It is not an admission of uncertainty, nor a middle setting between True and False. Every response is a claim about the text, so Cannot Say asserts something definite: the words in front of you settle nothing either way.
Cannot Say. The passage is the only admissible evidence, so accuracy in the world is irrelevant if the text does not supply it. This is the trap the format exists to set, and it catches specialists hardest: someone who knows the subject reads their own background into the passage without noticing.
When the passage entails the opposite of the statement, usually through a quantifier or a comparison. If the passage says a process rarely fails, "it never fails" is False, because rarely entails some failures. If the passage gives an instance of something, a statement denying any such instance is False. Absence of mention is Cannot Say; contradiction is False.
That depends on the scoring model, which the instructions normally state. Under raw scoring, answer everything: a blank and a wrong answer are worth the same. Under formula scoring the deduction is scaled to the three-way response set, so a blind pick has negative expected value. Eliminating one response first makes an answer worth submitting either way.
Because comfortable reading works against you. Skilled readers build a model of what a passage is about and answer from that model, which smooths over exactly the gaps the items target. The format rewards a pedantic return to the specific words behind each statement, close to the opposite of ordinary reading.
Budget by passage, not by item, because the reading is a fixed cost shared across every statement attached to it. One read for structure, then a targeted return for each statement, beats repeated full rereads. Where the clock allows movement, take the passages carrying the most statements first.