Reviewed 19 August 2026 by the Cognivy editorial team
Abstract reasoning is the broadest of the non-verbal test families, which is why candidates arrive unsure whether their assessment is abstract, inductive, diagrammatic or simply non-verbal. This guide answers that first, then covers the item types, the provider formats and a scan you can apply to any of them — and six free questions to run the scan on.
- Questions
- 15 to 50, depending on provider
- Time limit
- 12 to 25 minutes, depending on provider, sometimes shared across a whole battery
- Answer format
- Multiple choice: sequences, odd-one-out sets, classifications and analogies
- Delivery
- Standalone or one section of a battery. Pearson's DAT abstract subtest uses computerised adaptive testing
- Scoring
- Percentile against a comparison group. No provider publishes a pass mark
The skill
What is an abstract reasoning test?
An abstract reasoning test presents sets of figures and asks you to discover the relationship between them. It might be a sequence that continues, a pair whose relationship you apply to a new pair, or two groups whose membership rule you have to infer. In every case the shapes carry no meaning of their own.
Pearson describes the abstract subtest of its DAT range as looking at the ability to see sequences in shapes and patterns. Saville sells abstract reasoning as one of the aptitudes inside its Swift batteries. Talogy's Logiks measures abstract ability either on its own or as one section of a general test.
The terminology genuinely is loose. Non-verbal reasoning is the umbrella; abstract, inductive, diagrammatic and matrix reasoning all sit under it, and different publishers use different labels for similar tasks. What matters for preparation is not the label but the shape of the item in front of you.
If the items are grids with a missing cell, matrix reasoning is the closer guide. If they are input-operator-output chains, diagrammatic reasoning is. If your invitation names a specific inductive product such as SHL Verify Interactive Inductive Reasoning or one of Aon's inductive modules, the inductive guide covers that naming directly. All three are linked at the end of this page.
Try it before anything else
Free abstract reasoning practice questions
Six original questions covering the formats above — odd one out, analogy, next in series and classification. Run the attribute scan and commit to an answer before opening the solution; each solution names the rule and why every wrong option is on the list.
Five figures, described left to right. Each figure is a closed outline shape containing a number of small dots.
1: pentagon with 5 dots · 2: triangle with 3 dots · 3: hexagon with 7 dots · 4: octagon with 8 dots · 5: square with 4 dots
Which figure does not follow the rule?
Show the worked solution
Answer: B — figure 3.
- The figures are not in any order, so look for a relationship inside each figure rather than a change between them.
- Compare the two attributes present: number of sides and number of dots.
- Pentagon 5 and 5, triangle 3 and 3, octagon 8 and 8, square 4 and 4 all match.
- The hexagon has six sides and seven dots, so figure 3 is the exception.
Why the other options are there. The dot counts run 5, 3, 7, 8, 4 and tempt a search for a sequence rule. There is none, and there is no rule connecting one figure to the next. When a set is unordered, there is no between-figure sequence to find, so the relationship lives inside each figure.
Two pairs of tiles. Each tile contains some combination of a circle, a cross and a star.
Tile A holds a circle and a cross · tile B holds a cross and a star · combining A and B produces a tile holding a circle and a star. Now tile C holds a circle and a star, and tile D holds a star and a cross.
Which tile results from combining C and D?
Show the worked solution
Answer: A — a circle and a cross.
- Work out the operation from the first pair. A holds circle and cross, B holds cross and star, and the result holds circle and star.
- The cross appears in both inputs and is absent from the result. The circle and the star each appear in one input only, and both survive.
- So the rule keeps elements appearing in exactly one input and cancels elements appearing in both.
- Applying it to C and D: the star is in both, so it cancels; the circle and the cross each appear once, so both survive.
Why the other options are there. Option C keeps everything, which is what a plain overlay rule would produce. Overlay and cancellation look identical until you check an element that appears in both inputs, so always find one before deciding.
Each card below shows the same L-shaped tile: one long arm and one short arm meeting at a right angle. The four cards form a sequence, read left to right, and one rule carries the tile from each card to the next.
Which tile comes next in the sequence?
Show the worked solution
Answer: A — the corner at the bottom left, long arm up, short arm to the right.
- Name the change between the first two cards as a sentence: from Card 1 to Card 2 the long arm goes from pointing up to pointing right, which is a quarter-turn clockwise.
- Test that rule against every remaining step before trusting it. Card 2 to Card 3 turns right to down, and Card 3 to Card 4 turns down to left. The same quarter-turn clockwise holds at every step, and no card changes handedness, so a single rotation rule explains the whole row.
- Apply the rule once more to Card 4: a quarter-turn clockwise carries the long arm from left back to up and the short arm from up back to right, returning the tile to Card 1's orientation. Four quarter-turns make a full turn, so the sequence restarting is expected rather than suspicious.
- Run the handedness check on the answer: standing at the corner, the swing from long arm to short arm is clockwise on every card in the sequence, and option A keeps it clockwise.
Why the other options are there. Option B is the mirror image of the correct tile: corner and arms all in believable places, but the swing from long arm to short arm runs anticlockwise, and no rotation produces a mirror image. Option C turns the right amount in the wrong direction, a quarter-turn anticlockwise that walks the sequence back to Card 3's orientation. Option D repeats Card 4 unchanged, which is what remains when you notice the cycle but forget to apply the final turn.
Five cards, each showing an L-shaped tile: one long arm and one short arm meeting at a right angle.
Which card does not belong with the other four?
Show the worked solution
Answer: D — Card 4.
- The five cards are a set, not a sequence, so there is no between-card rule to chase. Look for a property four cards share and one lacks.
- Scan the obvious attributes and count the splits. Long arm direction: up appears twice, down, right and left once each, so no four-against-one there. Corner position: Cards 4 and 5 share the bottom right and the rest differ, so no four-against-one there either. Neither attribute can be the group's rule.
- Run the handedness check: stand at the corner of each tile and swing from the long arm to the short arm. Cards 1, 2, 3 and 5 all swing clockwise, however differently they sit.
- Card 4 swings anticlockwise. It is the mirror image of the tile the other four share, and no rotation can produce a mirror image, so it is the card that does not belong.
Why the other options are there. Every wrong card passes the swing test. Cards 1, 3 and 5 sit at less familiar angles than Card 2's upright pose, and under time pressure unfamiliar gets read as wrong, but each is the same tile merely turned. Card 5 pulls hardest because it shares its bottom-right corner with Card 4, and a glance at corners cannot separate the pair — only following the arms does. Card 2, the most natural pose of all, is there so that tidiness cannot be mistaken for the rule either.
Two groups of figures, each defined by an unstated rule. Every figure is one outline shape drawn inside another. A new figure has to be assigned to the right group.
Group 1: a triangle inside a square · a triangle inside a hexagon · a square inside a pentagon. Group 2: a hexagon inside a square · a pentagon inside a triangle · a hexagon inside a pentagon. New figure: a square inside a hexagon.
Which group does the new figure belong to?
Show the worked solution
Answer: A — Group 1.
- Work out Group 1's shared property first. Count sides: a triangle in a square is 3 inside 4, a triangle in a hexagon is 3 inside 6, and a square in a pentagon is 4 inside 5. In every figure the inner shape has fewer sides than the outer.
- Check that Group 2 holds the opposing rule rather than assuming it: 6 inside 4, 5 inside 3 and 6 inside 5 all put more sides inside than out.
- Refine the rule against the evidence. The triangle inside the hexagon differs by three sides and still sits in Group 1, so the rule is fewer, not exactly one fewer. The size of the gap carries nothing.
- Assign the new figure: a square has 4 sides and a hexagon has 6, and 4 inside 6 is fewer inside more, which satisfies Group 1's rule and violates Group 2's.
Why the other options are there. Group 2 is what you choose if you compare in the wrong direction — the hexagon does have more sides than the square, but the rule runs from the inner shape outwards. Neither group tempts anyone who read Group 1's rule as exactly one side fewer from its first and third figures; the triangle inside the hexagon is in the group precisely to rule that reading out. Either group fails because the two rules are opposites, so a figure that satisfies one can never satisfy the other.
Four boxes form a sequence, read left to right. Each box holds only circles and triangles.
Box 1: one circle and four triangles · Box 2: two circles and three triangles · Box 3: three circles and two triangles · Box 4: four circles and one triangle
What does the fifth box hold?
Show the worked solution
Answer: A — five circles and no triangles.
- Track each shape separately rather than reading the boxes whole. Circles run 1, 2, 3, 4 — rising by one each box.
- Triangles run 4, 3, 2, 1 — falling by one each box. Two independent rules, each verified across all three steps.
- Apply both rules once more: five circles and no triangles.
- A constant runs underneath as a cross-check: every box holds five shapes in total. That alone eliminates the options holding six or four shapes, and the remaining wrong option fails both counts.
Why the other options are there. Option B keeps the circles rising but leaves one triangle behind — the answer for anyone reluctant to let a count reach zero — and its six shapes break the steady total of five. Option C gets the triangles right but stalls the circles on four, applying one rule and forgetting the other. Option D restarts the row from Box 1, reading it as a cycle, and nothing in the four boxes repeats, so there is no evidence of any cycle to restart.
A short abstract reasoning set at the real per-question pace, scored server-side like the full product, with a worked solution after every answer.
Start the timed free setDraws on your 20 free practice questions. No account needed.
What employers are looking at
What do abstract reasoning tests measure?
Employers use abstract reasoning to estimate how someone handles a problem with no template: how quickly a rule is formed, tested, and abandoned when it fails. It is deliberately independent of vocabulary, arithmetic and job knowledge, which is why the same test travels across languages and educational backgrounds.
Three habits separate strong from weak performance. Scanning systematically rather than looking. Testing a hypothesis against every figure rather than the first one. Preferring the simplest rule that explains the whole set.
Reading your result
How abstract reasoning tests are scored
Abstract results are norm-referenced: what reaches the employer is a comparison against other people, not a count of correct answers. On adaptive delivery even the questions you met were partly chosen by your own answers, which changes what a good result means and what is worth optimising for.
- Scores are norm-referenced, not marked out of ten. Abstract scores are reported against a comparison group, and what an employer does with the resulting percentile depends on the role and on the rest of the process. The raw number of correct answers is not what travels.
- No provider publishes a pass mark. None of the providers reviewed on this page publish a universal passing score. Any threshold belongs to the employer rather than to the test, so if a recruiter has disclosed a target, that is the only target worth working to.
- On an adaptive test, difficulty is not feedback. Pearson's DAT Next Generation uses computerised adaptive testing: harder questions follow correct answers and easier ones follow incorrect answers. Rising difficulty is what doing well produces, so treat it as normal behaviour rather than as a verdict on your performance.
- The number of questions completed is not the score. On adaptive delivery, correct and incorrect responses change the difficulty of what follows, so the number of questions you get through does not explain the result on its own. Two candidates can answer the same number and score very differently.
- Question banks make memorised answers worthless. Talogy draws Logiks questions from a bank so that no two assessments match, and memorising questions is worthless against that design. The preparation that transfers is the attribute scan and the rule families, because they apply to figures you have never seen.
Question formats
What do the questions look like?
Six item shapes cover almost everything you will meet. Practise them separately before mixing them.
- Next in series. A row of figures changing by a consistent rule, where you choose the next one. SHL describes its inductive task as a logical sequence of five boxes where you decide which box completes it.
- Odd one out. A group where all but one obey a shared rule. Aon publishes a module that asks you to find the item in a series of nine that does not fit.
- Classification into groups. Two sets defined by an unstated rule, and a new figure to assign. Aon describes its scales cls module as understanding the common rules in the patterns given, then assigning new images to the relevant patterns.
- Visual analogy. Figure A is to figure B as figure C is to what. You name the transformation and apply it to a new starting figure.
- Combination and overlay rules. Two figures merge, with shared elements either kept or cancelled. Frequently stacked on top of a rotation or a count rule.
- Matrix completion. A grid with one missing cell. Covered in depth on the matrix reasoning guide, because its providers, pacing and scoring differ enough to need their own page.
Five boxes in a row. Across them, a shape rotates by a fixed angle at each step, a small dot moves one corner anticlockwise, and the fill alternates between solid and empty. Three independent rules running at once, and the answer options each satisfy two of them. Getting it right means finding all three before choosing, not after.
Launch coverage
Which providers use abstract reasoning tests?
Abstract material appears across four of Cognivy's launch families, under four different product names.
- DAT Abstract Reasoning. Pearson describes the subtest as looking at the ability to see sequences in shapes and patterns, delivered with computerised adaptive testing at 15 items in approximately 12 minutes.
- Abstract Reasoning. Usually delivered inside a Swift battery. Saville lists Swift Executive Aptitude at 18 minutes across numerical, verbal and abstract reasoning, and Swift Global Aptitude at 12 minutes across number reasoning, abstract reasoning and memory recall.
- Logiks Abstract (Advanced). Available as a standalone ability or as one section of Logiks General. Talogy publishes the range at 12 to 25 minutes and 20 to 50 items depending on version, drawn from an item bank so that no two assessments match.
- scales cls. Aon files scales cls under inductive-logical thinking and describes it as understanding the common rules in the patterns given, then assigning new images to the relevant patterns. It is a classification task rather than a sequence.
- scales clx. Aon's inductive logical reasoning module presents two three-by-three example grids that follow a rule and asks which two of four further grids follow the same rule. Aon lists it at 6 minutes and states there is no set number of questions, because the system creates a new one each time you finish. It is taught here rather than on the inductive reasoning guide because the task is rule classification, exactly as its sibling scales cls is, rather than continuing a sequence. Cognivy's practice serves that reasoning move as a single-select figure classification: nobody's bank does select-two-of-four, and saying so is better than implying the format is reproduced.
Provider names identify the assessment format. Cognivy is independent and is not affiliated with or endorsed by these providers.
What varies, and what does not
Timing, delivery and scoring
Abstract assessments are typically short and often adaptive. What varies most is whether the abstract section stands alone or is one part of a combined battery.
- Standalone or combined changes everything. Pearson's DAT abstract subtest is roughly 12 minutes on its own. Saville's Swift Executive Aptitude spends 18 minutes across three aptitudes, so the abstract share of that is much smaller.
- Adaptive delivery is common. Pearson's DAT Next Generation uses computerised adaptive testing: harder items follow correct answers and easier ones follow incorrect answers. An adaptive test may follow a correct answer with a harder item, but you cannot read your score from how hard the questions feel, so treat rising difficulty as normal behaviour rather than as feedback.
- Item banks mean no two tests match. Talogy states that the Logiks item bank makes every assessment different from the last. Memorising items is worthless against that design.
- No provider publishes a pass mark. Results are reported against a comparison group, and any threshold belongs to the employer rather than to the test.
A repeatable approach
A scan that works on any item
One scan, applied in the same order to every item, does most of the work.
- 1
Run the attribute scan. Count, shape, size, position, rotation, shading, line style, symmetry. Always in the same order, so nothing gets skipped when you are rushed.
- 2
Describe the change in words. Say what happens between adjacent figures as a sentence. A rule you cannot state is a rule you cannot check.
- 3
Test the simplest rule against every figure. One-step explanations first. If a simple rule holds for all but one figure, the exception is usually your misreading rather than evidence of a more complex rule.
- 4
Expect more than one rule. Independent attributes often change at the same time. Count how many options your first rule eliminates; if several survive, another rule is running.
- 5
Use the options to falsify. Take your candidate rule to the options and try to kill it. The rule that survives every option is the rule.
What slows progress
Common mistakes, and the fix for each
- Assuming every visible feature matters. Fix: identify which attributes are constant across all figures. Constants are decoration. Only what changes can carry the rule.
- Testing a hypothesis on one example. Fix: check the rule against every figure in the set before looking at the options. A rule that works twice and fails once is not the rule.
- Reaching for a complex rule too early. Fix: exhaust one-step explanations first. A complicated hypothesis is usually a symptom of a misread attribute.
- Confusing rotation with reflection. Fix: track a single asymmetric feature through the change. Rotation carries it around the figure; reflection swaps its side.
- Practising only one item shape. Fix: sequences, odd-one-out, classification and analogy each reward a different first move. Practise at least two so an unfamiliar layout does not cost you the opening minute.
You have seen the method. Employers set the full battery.Every provider, every track, a worked solution on every question.
Get full accessIn practice
How do you find the rule when every figure looks different?
Run the attribute scan in a fixed order — count, shape, size, position, rotation, shading, line style, symmetry — and in the same order every time. Whatever stays constant across the figures is decoration, and only an attribute that changes can carry the rule, so the scan's real product is a shortlist of two or three attributes worth testing.
Then describe the change as a sentence, because a rule you cannot state is a rule you cannot check. In sample question 6, saying that circles rise by one while triangles fall by one settles the answer immediately and exposes every option that breaks a count. Vague recognition rejects nothing.
Prefer the simplest rule that survives every figure, and test it against all of them rather than the first two. If a one-step rule holds everywhere except a single figure, the likeliest explanation is a misread attribute rather than a deeper rule, so recheck before you rebuild.
In practice
How do you tell rotation from reflection?
Track one asymmetric feature through the change. Rotation carries a feature around the figure and keeps the direction everything winds; reflection swaps its side and reverses that winding. No rotation within the plane can turn a figure into its mirror image, which is exactly why mirror images are the standard wrong option wherever rotation is the rule.
On the L-shaped tiles in sample questions 3 and 4 the check takes a second: stand at the corner, swing from the long arm to the short arm, and note whether that swing runs clockwise or anticlockwise. Every rotation of the tile keeps the same swing, and a mirrored tile always reverses it. The mirrored option in sample question 3 puts the corner and both arms in believable places and still fails this one test.
Do not lean on position instead. A rotation and a reflection can put a corner in the same place — Cards 4 and 5 in sample question 4 share theirs — so where a feature sits cannot separate the two. Which way the figure winds always can.
In practice
How do you spot the odd one out when everything looks similar?
Decide first whether the set is ordered. Odd-one-out groups usually are not, and an unordered set has no between-figure sequence to find, so the rule lives inside each figure or across the whole group. Sample question 1's dot counts run 5, 3, 7, 8, 4 precisely to bait a sequence hunt with nothing at the end of it.
Then look for the split where all but one figure agree, rather than for the figure that feels strange. Count how many figures share each attribute value: a three-against-two split, like the vertical and horizontal long arms in sample question 4, cannot be the group's rule, because the rule has to hold for every figure except exactly one. The attribute that splits four against one is the answer.
Reject on a named violation, never on unfamiliarity. Question writers draw rule-obeying figures at awkward angles because looking odd is the proxy candidates fall back on when rushed. If you cannot say which shared property a card breaks, you have not found the odd one out yet.
A clear route
How to prepare in the days you have
What an abstract item asks for is mostly procedure rather than concept, so a short run-up is best spent in this order.
- 1
First, build the attribute scan. Write the checklist down and run it on every item until you no longer need it. This is the highest-return hour of abstract preparation.
- 2
Next, one rule family at a time. Rotation, count, position, shading, overlay. Untimed, until you can name each one in a couple of seconds.
- 3
Then stack two rules. Real items rarely run only one. Practise stating both before choosing an option, and treat a one-rule answer as an incomplete answer.
- 4
Match the item shape to your invitation. A classification test and a next-in-series test feel different under time pressure. Practise the one you were actually sent.
- 5
Add the clock last. Short timed sets, then one full-length rehearsal. If your assessment is adaptive, resist rushing the early items.
Cognivy uses your assessment date to choose the route rather than asking you to predict a study schedule. When you sit down to practise, you choose the session length that fits that day.
Pace
Getting faster without losing accuracy
Speed follows from the scan being automatic. Until it is, going faster only means missing more.
- Fix the scan order. A memorised order is faster than looking, because you stop spending attention on deciding where to look next.
- Describe, then choose. Naming the rule before reading the options is faster overall, and it makes you immune to an attractive distractor.
- Kill options, do not rank them. Reject on a named violation. Comparing options against each other is slower and less reliable than testing each one against the rule.
- Move on from an unresolved item. On a short adaptive test, one stubborn item is worth less than the several you could answer in the same time.
Direct answers
Abstract reasoning test FAQs
A non-verbal test that presents unfamiliar shapes and asks you to discover the rule governing them, then apply it. Items usually appear as sequences, odd-one-out sets, classifications, analogies or grids. Pearson describes the abstract subtest of its DAT range as looking at the ability to see sequences in shapes and patterns.
Non-verbal reasoning is the umbrella term for tests built from shapes rather than words or numbers. Abstract reasoning is the main family within it. Inductive reasoning describes the same underlying skill and is also a specific product name, because SHL and Aon both publish tests called inductive reasoning. Match your practice to the item shape rather than to the label.
Yes. A matrix item is an abstract reasoning item in grid form, with one cell missing rather than a sequence to continue. It is common enough, and its providers distinctive enough, that Cognivy keeps a separate matrix reasoning guide for it.
It depends on whether it stands alone. Pearson lists its DAT abstract subtest at 15 items in approximately 12 minutes. Talogy publishes the Logiks range at 12 to 25 minutes and 20 to 50 items depending on version. Saville usually delivers abstract inside a Swift battery of roughly 12 to 24 minutes covering several aptitudes.
Yes, for the process. You can practise recognising rule families, running a consistent attribute scan and testing each hypothesis against every figure. Cognivy does not claim that practice raises an underlying ability score.
Some are. Pearson's DAT Next Generation uses computerised adaptive testing, and Talogy draws Logiks items from a bank so that no two assessments are identical. An adaptive test may follow a correct answer with a harder item, but you cannot read your score from how hard the questions feel. Follow the instructions attached to your own assessment.
There is no universal pass mark. Scores are reported against a comparison group, and the benchmark is set by the employer for the role rather than by the test publisher.
No. Cognivy is independent and is not affiliated with or endorsed by any assessment provider. All practice items are original and written to teach the rule families and the checking method.