Deductive Reasoning Test 2026: How to Pass Must Be True

Must be true, and nothing weaker

Read the question wording first, because it sets the bar. Three wordings are common, and they want different things.

1. Which must be true? The conclusion holds in every situation the facts permit. One counterexample kills it.
2. Which cannot be true? The conclusion fails in every situation the facts permit. One example that works kills it.
3. Which could be true? The conclusion holds in at least one situation the facts permit. This is the easiest bar, and candidates trained on the first two often over-reject here.

The test for "must be true" is mechanical. Try to invent one world where all the facts hold and the conclusion fails. If you can, the answer is wrong, however plausible it looks. If you genuinely cannot, it is forced.

That single habit, actively looking for the counterexample rather than checking whether the conclusion feels supported, is what separates candidates on this format. Plausibility is not a criterion here, and the wrong options are written to be plausible.

The same line separates deductive from inductive reasoning, and the difference between deductive and inductive reasoning is worth naming here because the two tests sit side by side in one product family. Deductive reasoning examples run from stated facts to a conclusion the facts force. Every analyst on the Lisbon team files a weekly timesheet, Marta is on the Lisbon team, so Marta files a weekly timesheet. The conclusion adds nothing the premises did not already contain, which is exactly why it cannot fail while they hold.

Inductive reasoning examples run the other way, from observations to a general claim that reaches past them. The last four Lisbon timesheets arrived on a Friday, so the fifth will arrive on a Friday. That conclusion is useful, it is how most real analysis works, and a single late Friday refutes it. A deductive item marks you wrong for taking that second kind of step, however sensible it is, and SHL sells a separate inductive reasoning test for the pattern-spotting this one refuses to credit. Held against each other, inductive vs deductive reasoning comes down to the direction of travel. Deduction runs down from a rule to a case the rule already covers, so its conclusion cannot fail while the premises hold. Induction runs up from cases to a rule that reaches past them, so its conclusion can be useful and still be wrong. That is the whole of it, and it is the only piece of theory this format pays for. Carry one sentence into the sitting: the question asks what the facts force, and never what they suggest.

SHL describes what its deductive product measures as "the ability to draw logical conclusions based on information provided, identify strengths and weaknesses of arguments, and complete scenarios using incomplete information." The last four words matter. The facts are deliberately incomplete, and the gaps are where the wrong answers live.

What the publisher prints about the deductive reasoning test

SHL's fact sheet for Verify deductive reasoning gives the shape of the sitting in plain numbers: "Allowed Time 20 minutes," "Maximum Number of Questions 18 questions," "Question Format Multiple Choice, Adaptive," and "Designed for Unproctored Environment Yes." It is published at shl.com and carries a 2022 date in its own footer, checked 29 September 2026.

Twenty minutes for up to eighteen items is roughly a minute each, and the adaptive design means the difficulty tracks your answers. Items getting harder as you go is the sign of a good run rather than a bad one.

The same document says the test "utilizes computer adaptive technology" for three reasons it lists: "Shorter testing administration time," "Ability to offer the test in an unsupervised setting," and "A more precise test score." Note the middle one. The whole design assumes you are sitting alone at home, which is what makes the next section matter.

One caution on naming. SHL sells more than one deductive product, and the interactive versions differ from the multiple choice ones in length and response style. Any page that gives you a single duration for "the SHL deductive test" is describing one product and calling it the family.

The verification retest, which nobody mentions

The fact sheet lists the sittings as "One plus Verification Test." Across the whole first page of results for this query, it goes unmentioned.

What it means in practice: the unsupervised test you take at home can be followed by a supervised version, sat under observation, to check that the first result was yours. A large gap between the two is the signal the mechanism exists to catch.

That turns the usual advice about getting help on an online test into a straightforward risk. Help inflates the unsupervised score, the verification test is where the inflation shows, and the candidate who scored honestly has nothing to reconcile. It also reframes practice. You are preparing to reproduce a score twice over, rather than to produce one on a single afternoon.

SHL positions the cognitive range at shl.com/products/assessments as "automated, remotely proctored, and ask fewer questions thanks to our unique partial-scoring methodology," and names deductive reasoning among the products in it. Fewer items with partial scoring is why eighteen questions can separate a field, and why one careless answer costs more here than on a long paper.

Worked case one: the conclusion that only looks forced

Every analyst in the Lisbon office speaks Portuguese. Marta is an analyst in the Lisbon office. Paulo speaks Portuguese.

"Marta speaks Portuguese." Must be true. She is in the group the first fact covers, so the conclusion is forced with no room left.

"Paulo is an analyst in the Lisbon office." Could be true, and it is not forced. The facts say every Lisbon analyst speaks Portuguese. They say nothing about whether anyone else does, and Portuguese speakers plainly exist outside that office. Invent a world where Paulo works in Madrid and speaks Portuguese: every fact still holds, the conclusion fails, so it is not a must.

"Some analysts in the Lisbon office speak only Portuguese." Could be true. The facts establish that Lisbon analysts speak Portuguese, and nothing about what else they speak. The word "only" adds a restriction the facts never gave.

This item is the single most common error shape on the format. A rule of the form "all A are B" does not license "this B is an A." Read the direction of every rule before you use it, and keep the two halves in the order the facts put them.

Worked case two: when the answer is cannot be true

No consultant on the Bilbao project has worked in retail. Alia has worked in retail. The Bilbao team has six members.

"Alia is on the Bilbao project." Cannot be true. The first fact excludes everyone with retail experience, the second puts Alia in that group, so the two cannot hold together with the conclusion.

"Every consultant on the Bilbao project has worked in energy." Could be true. Nothing in the facts mentions energy at all, so the statement is neither forced nor excluded. Candidates reject this one because it feels unsupported, and unsupported is the correct state for a "could be true" answer.

"The Bilbao team has fewer than six members with retail experience." Must be true. The team has six members and none has retail experience, so the count of members with retail experience is zero, and zero is fewer than six.

That third statement is worth sitting with. It is technically weaker than what the facts establish, and a weaker claim that the facts guarantee is still a must. Some candidates reject an answer for understating the case, which costs a mark on a correct option.

Worked case three: an arrangement, solved by elimination

Five interviews are scheduled in slots one to five, one candidate each: A, B, C, D and E. A is in slot one. B is immediately before D. C is after D.

Work the constraints in order of how much they forbid. A is fixed in slot one, so B and D occupy an adjacent pair somewhere in slots two to five, with B first.

• B in 2 and D in 3. C can then be 4 or 5, and E takes the other.
• B in 3 and D in 4. C must be 5, so E is 2.
• B in 4 and D in 5. C would need slot 6, which does not exist, so this pair is dead.

Three arrangements survive:

Now test candidate conclusions against all three. "D is in slot three" fails on option three. "C is in slot five" fails on option one. "E is in slot two" fails on options one and two. "B is not in slot five" holds in all three, and the dead pair above shows why it is forced rather than lucky.

The method generalizes. Enumerate the arrangements that survive the most restrictive constraint, then read the answer options against your list. Trying to reason about the options directly, without the list, is what turns a one minute item into a four minute one.

The publisher's own example, and what it teaches

SHL prints an example on the same fact sheet. It gives four short facts about two people and their cars, then five options, and asks which one MUST be true. The published answer walks the reader through a single chain: Susan drives a blue car, blue cars get a stated mileage, "the fact that Susan drives a blue car means that her car gets 33 miles per gallon of gas."

Two things are worth taking from it. The correct option uses exactly two of the four facts, and the discarded options each require one extra assumption the facts never gave, such as where someone lives. That ratio is typical. A must-be-true answer is usually a short chain through two or three facts, and the distractors are longer chains that need a bridge you were never handed.

So when an option requires you to explain why it follows, that explanation is usually the assumption that disqualifies it. The correct answer tends to need no explaining beyond pointing at the facts in order.

Five ways the deductive reasoning test is failed

Four failure modes, in rough order of cost.

• Reversing a rule. Using "all A are B" to conclude something about a B. Case one above.
• Importing the ordinary world. Filling a gap with what is usually true. The facts are incomplete on purpose.
• Rejecting a weaker claim. A conclusion that understates what the facts guarantee is still guaranteed.
• Reasoning without enumerating. On arrangement items, refusing to write the surviving options and trying to hold them in your head.

There is a fifth that only shows under the clock: spending three minutes on one arrangement item and losing two easy ones at the end. On a twenty minute test with up to eighteen items, an item that passes ninety seconds should get a committed guess.

One platform detail decides whether guessing is free. TestGorilla tells candidates that "Once skipped, you cannot go back and answer a question. We do not recommend skipping questions. Skipped questions count as wrong answers. If you don't know the answer, guess." Its guide is at candidates.testgorilla.com, checked 29 September 2026. Read your own instruction screen, and let it override any general rule.

How it is scored, and what stays private

SHL's candidate report prints a percentile against a comparison group that is named on the report itself, plus a letter grade on a five point scale shown as E, D, C, B, A. On the sample report SHL publishes, results are compared with a dated general population group, so the benchmark is a specific population rather than an absolute standard.

The report also states that "test results remain valid for about 12 to 18 months," which makes a second attempt inside the same recruiting cycle unlikely to be offered.

The cutoff is not published by anyone. SHL reports the percentile and the band, the employer decides which band continues, and the same test serves different roles at different bars. The score bands quoted around this subject, the ones that tell you seven out of ten is a pass, are illustrations written by the sites quoting them.

Two further things sit outside anyone's reach. Which publisher your test comes from varies by firm, office and cycle. So does the composition of your sitting, because publishers assemble an assessment from modules the employer selects. Your invitation email names the platform, and it is the only current source on that.

That is also why a search by firm lands somewhere else. An EY assessment test, a McKinsey assessment test, a BCG assessment test, a Deloitte assessment test, a KPMG verbal reasoning test, an Oliver Wyman numerical reasoning test, a PwC logical reasoning test and an L.E.K. assessment test are eight separate questions, and the answer to each one sits with that firm rather than on a publisher's fact sheet. What each firm publishes about its own screening step is on its guide: EY, McKinsey, BCG, Deloitte, KPMG, Oliver Wyman, PwC and L.E.K.. One firm in that set names its two publishers outright, and a Bain assessment test is either Sova or Test Gorilla depending on the email you get, which the Bain case interview guide takes apart rule by rule.

Practicing for the deductive reasoning test against a clock

The three deductive reasoning questions above are printed with their answers and with the step that forces each one, which is the part a question bank usually leaves out. Three shapes cover almost every item: a conclusion drawn from stated premises, an arrangement under constraints, and the assumption a claim needs in order to hold. Practice them separately, because they fail differently and the corrections do not transfer.

Work timed from the first session, and after each set write down which of the five failure modes caused each miss. One will dominate, and that one is your preparation. Untimed practice trains the logic, which most candidates already have, and leaves the pacing untouched.

SHL publishes example questions and a full length practice test on its own candidate site, covering numerical, verbal, inductive, calculation and checking formats, and its deductive fact sheet carries the worked example described above. It is the publisher's own material and it is free, so start there before paying for a question bank.

We do not have a deductive drill, and claiming one would be easy and false. The free drills train the case version of the same discipline, where a recommendation has to follow from the numbers on the page rather than from what sounds right, and how an answer is assessed sets out the weights they score against, read from the code that does the scoring. Our assessment simulators cover firm built formats rather than publisher tests, and none of the six is live yet, which that page states itself.

For the adjacent formats, verbal reasoning covers the cannot say rule and numerical reasoning covers where the clock goes on exhibit questions.