Data Insights is the section of the GMAT Focus where a single prompt can hand you a bar chart on the left, a six-row table on the right, and two short paragraphs of commentary underneath, then ask a question that none of the three sources can answer on its own. Candidates who treat those prompts the same way they treat a single-source graph or a single-source table quietly bleed two to three minutes per question, and that time tax is exactly what separates an 84 from a 78. The skill that closes the gap is not faster arithmetic, faster reading, or faster chart reading in isolation; it is learning to read three sources as a single argument.
This article lays out the read order, the annotation habits, and the question-decoding moves that let you handle mixed-format prompts inside the 2 minutes 30 seconds the section allows. Everything below is written for the current GMAT Focus edition, where Data Insights is a 45-minute, 20-question section scored on a 60-90 scale, and where graph-plus-table-plus-text prompts are routinely the hardest 20% of the section. If you can master the protocol below, the single-source prompts start to feel like a controlled warm-up.
What a mixed-source Data Insights prompt actually looks like
Most candidates who say they struggle with mixed-format prompts have not stopped to define what a mixed-format prompt actually is. On the GMAT Focus, the Data Insights section draws from five item families: Data Sufficiency, Multi-Source Reasoning, Table Analysis, Graphics Interpretation, and Two-Part Analysis. Mixed-source prompts are not a sixth family; they are the prompts inside Graphics Interpretation, Table Analysis, and Multi-Source Reasoning where the test deliberately gives you more than one visual or textual source, and asks a question that requires you to combine them.
The defining feature of a mixed-source prompt is that no single source, on its own, contains the answer. A graph might show you revenue by region, a table might show you unit volume by product line, and a short text paragraph might explain a one-off cost that hit in the third quarter. A question that asks 'Which region's revenue growth was most understated by the change?' forces you to read the graph for the top-line number, the table for the underlying volume, and the text for the one-off adjustment. Read only two of the three and the correct answer is unreachable.
This is also the prompt type where the test is most willing to make sources disagree. A graph might report a figure rounded to the nearest million, a table might show the unrounded version, and the text might mention a restatement. The candidate's job is to notice the disagreement, decide which source is authoritative for the question being asked, and answer accordingly. Most candidates never learn to look for the disagreement, because their read order is graph-first, question-second, table-third, text-when-stuck. That order is exactly backwards on a mixed-source prompt.
Item families most often producing mixed-source prompts are Multi-Source Reasoning, where three discrete tabs are the entire point of the question type, and the harder end of Graphics Interpretation, where a chart and a table sit on the same screen. Table Analysis prompts are technically single-source, but the very best Table Analysis questions hide a small graph inside the on-screen calculator view, which counts as a second source for read-order purposes. The protocol below works for all three.
The five-pass read order that protects your timing
On a mixed-source Data Insights prompt, the default mistake is to read top-to-bottom-left-to-right the way a web page is read. That habit costs you a re-read, and a re-read on a 2 minutes 30 seconds budget is the most expensive thing you can do. The protocol I teach in our diagnostic sessions is a five-pass read order that takes roughly 40-50 seconds of the question and that turns the rest of the time into answer evaluation rather than source re-reading.
Pass 1, the question stem (5-7 seconds). Read only the actual question, not the introductory context. The question stem almost always ends in a question mark and contains the verb that defines the task: 'Which value is greatest?', 'Which option is supported?', 'What is the minimum percentage change?', 'In the table below, which row...'. The verb tells you what kind of answer you are looking for, which tells you which source is the answer source and which sources are just supporting context. Candidates who skip this pass end up reading the entire prompt, then re-reading the question, then re-reading the sources in a different order. That is three reads before the first candidate evaluation.
Pass 2, the text paragraph (10-15 seconds). Text paragraphs on mixed-source prompts almost always contain the definitions, the units, the restatements, the assumptions, or the one-off events that the chart and table cannot show. Read the text paragraph second, even if it sits below the chart on screen, because the text is where the test tells you what the numbers mean. Underline, mentally or on the scratch pad, the units, the time period, and any sentence that contains a hedge word like 'approximately', 'subject to', 'excluding', or 'adjusted for'.
Pass 3, the chart (15-20 seconds). Only now do you look at the chart. The chart's job is usually to give you a shape: which region is biggest, which quarter was the anomaly, which line crossed which line. Do not read every data label. Identify the axis, the legend, and the one or two visual features that the question is asking you to compare. If the question is about which segment grew fastest, your chart work is to identify the two segments and the two endpoints, not to read every bar height.
Pass 4, the table (15-20 seconds). The table is where the precise numbers live. By the time you reach the table, you already know which row, which column, or which cell the question is pointing at. Locate that cell, read the value, and check that the units match the question's units. This is also the pass where you catch the rounding mismatch: a chart label of 1.2 and a table value of 1.18 are the same number at the test's level of precision, and the prompt is allowed to round either way.
Pass 5, the answer choices (60-90 seconds). Only after all four sources are anchored do you evaluate the answer choices. This is the pass that actually takes the bulk of the time, and it is the pass that the four preceding passes are designed to protect. A candidate who reads sources and choices in parallel is constantly re-reading. A candidate who reads sources first, in the order above, evaluates choices against a fully loaded short-term memory.
Why this order works
Reading the question first sounds obvious, but most candidates read the question last, because the question is physically at the bottom of the prompt. The test designers know this, and they place question stems at the bottom precisely so that candidates will read the sources, form a hypothesis, and then read a question that has been carefully written to be consistent with a plausible-sounding but wrong reading of the sources. Putting the question first inoculates you against that design choice.
Reading the text second works because text is the slowest source to parse but the smallest in information volume. Reading it second, with the question already in working memory, lets you read it once instead of twice. Reading the chart third puts the fastest-to-skim source in the slot right before the table, which forces you to be selective about which chart features matter, since the table will give you the precise numbers anyway.
Annotation: what to write on the scratch pad and what to leave in your head
Candidates who try to write down everything from a mixed-source prompt run out of scratch pad and out of time. Candidates who write nothing spend half of the question's time re-reading. The right move is to annotate only the items that will answer a follow-up question, not the items that answer the current question. Mixed-source prompts are frequently clustered in the section, and the chart or table from prompt 7 will reappear as a reference in prompt 8. Your annotations need to survive that reappearance.
Three annotations cover roughly 80% of mixed-source prompts. First, write the unit and the time period at the top of your scratch pad as soon as you see them. 'Revenue, USD millions, Q1-Q4 2019' is a five-second write that prevents the most common mixed-source error, which is using a 2018 figure when the question asked about 2019. Second, draw a tiny arrow on the chart or sketch the relevant cell of the table, and label it with the question's verb. 'Grew fastest' next to two cells is a five-second write that turns the answer evaluation into a glance rather than a re-derivation. Third, if the text paragraph contains a one-off event, write the event in three words. 'Tax dispute settled' or 'FX headwind' or 'Divestiture of X' is enough; you do not need the full sentence.
What you should not annotate is the value of every cell, the height of every bar, or the wording of every text sentence. Candidates who annotate at that level of detail spend more time writing than the 2 minutes 30 seconds budget allows, and the annotations are useless by the next prompt. The point of annotation on Data Insights is to convert the prompt into a question you can answer in one look, not to produce a faithful copy of the prompt on paper.
Resolving the disagreement between sources
The single most diagnostic move in mixed-source prompts is learning to spot when two sources disagree, and to know which one to trust. The GMAT Focus builds this feature into roughly a third of mixed-source prompts, and the candidates who never learn to look for it will answer those prompts correctly only by accident.
The three disagreement patterns are rounding, period, and definition. Rounding is the simplest: the chart shows 4.2, the table shows 4.18, and the question asks for the percentage change. Both sources are correct; you just need to use the more precise one for the arithmetic. Period disagreement is more subtle: the chart covers fiscal year 2019, the table covers calendar year 2019, and the question asks about 2019 without specifying which. Definition disagreement is the hardest: the chart defines 'active customer' as anyone who logged in during the quarter, the table defines it as anyone who made a purchase, and the question asks about active customers in Q3. The right move is to find the definition that the text paragraph uses, because the text is the test's tie-breaker on definition.
A useful rule of thumb is the following: if two sources disagree on a number, the more precise source is correct. If two sources disagree on a period, the text paragraph is correct. If two sources disagree on a definition, the text paragraph is correct. If no text paragraph exists, then the chart and the table are both candidates and the question stem will tell you which to use. In my experience tutoring candidates stuck in the 76-80 band, this rule of thumb alone accounts for one to two correct answers per section.
A worked example of resolving disagreement
Imagine a prompt where a bar chart shows Q3 revenue of 142, a table shows Q3 revenue of 142.4, and the text mentions that Q3 includes a one-off gain of 2.1. A candidate who only reads the chart and the table will compute a percentage change off 142 or 142.4, both of which include the one-off. A candidate who reads the text second will see the one-off, subtract 2.1 mentally, and compute the percentage change off roughly 140.3 versus 140.4 from the prior quarter. That is a small adjustment in the arithmetic but a large adjustment in which answer choice survives. The text is the source that makes the difference.
Question-type decoding: matching the verb to the source
Mixed-source prompts hide their difficulty in the question stem, not in the sources. The same three sources can support a 'which is greatest' question, a 'which is supported' question, and a 'what is the minimum' question, and each of those three questions will send you to a different cell of the table, a different bar of the chart, and a different line of the text. Learning to decode the verb is what turns a 90-second source-re-read into a 30-second answer confirmation.
The verb families that appear most often on mixed-source Data Insights prompts are comparison, supported, inferable, minimum or maximum, and percentage change. Comparison verbs ('which is greatest', 'which is smallest', 'which exceeds') point you to the chart for shape and the table for precision; you do not usually need the text unless the comparison is over a time period that the text redefines. Supported verbs ('which option is supported', 'which statement is true') point you to all three sources, because support is the one task that the test most often makes a single source unable to provide. Inferable verbs ('which can be inferred', 'which must be true') are similar to supported but with a tighter logical standard: the conclusion has to follow from the sources, not just be consistent with them.
Minimum and maximum verbs ('what is the minimum percentage', 'the greatest possible value') send you to the cell or bar that defines the boundary, then to the text to check for a constraint, then to the table for the precise number. Percentage change verbs ('by what percent did X increase', 'what is the year-over-year change') send you to the table first, because percentage change needs precise numbers, and then to the text, because the change might be partially driven by a one-off. Candidates who skip the verb decoding step spend 30 seconds on the wrong source for the wrong reason.
Timing budget for a mixed-source prompt
The 2 minutes 30 seconds the GMAT Focus allows for each Data Insights question is the same budget for a single-source prompt and a mixed-source prompt, which is exactly why mixed-source prompts feel rushed. The timing budget below is what I walk candidates through in our diagnostic work; it is not a hard rule, but it is a useful default to drill before you personalise it.
Pass 1, the question stem, takes 5-7 seconds. Pass 2, the text paragraph, takes 10-15 seconds. Pass 3, the chart, takes 15-20 seconds. Pass 4, the table, takes 15-20 seconds. That leaves 60-90 seconds for answer evaluation, of which roughly 30 seconds should be spent on the first candidate answer and roughly 30-60 seconds on the second pass if the first candidate answer does not survive. The total is in the 105-152 second range, which fits inside 150 seconds with a small buffer. Candidates who exceed 90 seconds on the first two passes together have lost the prompt and should pick their best guess rather than continue, because one lost prompt is recoverable inside a 20-question section but two lost prompts usually is not.
The buffer exists because the second prompt in a Multi-Source Reasoning pair is often easier than the first, and a candidate who finishes the first prompt in 110 seconds has 40 seconds of buffer to spend on the second. Candidates who do not internalise this timing budget end up in the situation where the first prompt takes 2 minutes 10 seconds, the second prompt takes 1 minute 50 seconds, and the third prompt never gets read. The protocol's value is not the protocol itself but the predictable timing it produces.
Common pitfalls and how to avoid them
Five pitfalls account for the majority of mixed-source Data Insights errors in our diagnostic work. None of them is about the underlying math; all of them are about how the sources are read and in what order. Recognising them is half the fix; the other half is the read-order protocol above.
Pitfall 1, reading the chart before the question. A candidate who reads the chart first forms a hypothesis about what the question is asking, and then reads the question to confirm. The test exploits this by writing questions that look like chart questions but are actually text questions. The fix is to make pass 1 of the protocol the question stem, no exceptions, even on prompts where the question looks obvious from the chart.
Pitfall 2, treating the text as a caption. The text paragraph on a mixed-source prompt is not a caption explaining the chart; it is a source in its own right. Candidates who skim the text as if it were a caption miss the one-off events, the redefinitions, and the assumptions. The fix is to read the text as slowly as the question, not as fast as the chart.
Pitfall 3, picking the first plausible answer. Mixed-source prompts are designed so that a chart-only reading, or a table-only reading, produces a plausible but wrong answer. A candidate who evaluates one answer against one source and selects it has fallen into the trap. The fix is to require at least two sources to support any answer choice before you select it.
Pitfall 4, re-reading on the second prompt of a pair. Multi-Source Reasoning presents two or three prompts against the same three tabs. A candidate who finishes the first prompt and then re-reads the tabs for the second prompt loses the time benefit of the shared source. The fix is to leave yourself a one-line annotation on the tabs that summarises the key value, so the second prompt can be answered without a re-read.
Pitfall 5, ignoring the on-screen calculator. Table Analysis prompts include a small calculator that the test designers use to embed a tiny graph or a tiny table inside the on-screen view. A candidate who treats the on-screen calculator as a calculation tool rather than a second source will miss the embedded visual. The fix is to expand the calculator view on every Table Analysis prompt and to read it as a second source, not as a tool.
Drilling the protocol: a 10-prompt practice loop
Reading about the protocol is not enough; mixed-source prompts are a skill that has to be drilled, and the right drill is short and frequent. The 10-prompt practice loop below is what I assign to candidates in the 76-82 score band who are trying to break into the 84-88 range. It is designed to take roughly 35-40 minutes, which fits inside a single focused study block.
Start with 10 mixed-source Data Insights prompts, all of them Multi-Source Reasoning or hard Graphics Interpretation. Use the official GMAT Focus practice materials, because third-party materials often mis-label question families. Set a timer for 2 minutes 30 seconds per prompt and run the five-pass protocol exactly. After each prompt, write one sentence on the scratch pad: which source held the answer, and where the protocol broke down if it did. After all 10 prompts, count how many of the 10 you finished inside 2 minutes 30 seconds, and how many of the 10 you got right. The target is eight of 10 finished on time and seven of 10 correct. If you are not at seven of 10, the issue is source-mapping, not arithmetic, and you should rerun the loop focusing on pass 1 and pass 2 only.
The second iteration of the loop should be the same 10 prompts in a different order, this time annotating only the three items the protocol calls out. The third iteration should be a fresh set of 10 mixed-source prompts, this time forcing yourself to skip the chart on pass 3 if the table answers the question, and skip the table on pass 4 if the chart answers it. By the third iteration, the protocol starts to feel automatic, and that automaticity is what protects your timing on test day.
| Pass | Time budget | Source | Output to scratch pad |
|---|---|---|---|
| 1 | 5-7 seconds | Question stem | Verb, units, time period |
| 2 | 10-15 seconds | Text paragraph | One-off events, definitions, hedges |
| 3 | 15-20 seconds | Chart | Shape, two visual features |
| 4 | 15-20 seconds | Table | Precise cell, rounding check |
| 5 | 60-90 seconds | Answer choices | First candidate, second pass if needed |
How the protocol shows up on test day
Test day is not the day to learn the protocol, and it is also not the day to override the protocol because a prompt looks easy. Mixed-source prompts are designed to look easy on the surface, which is the test's main way of separating the candidates who read the protocol from the candidates who read the chart. A candidate who runs the five-pass protocol on every mixed-source prompt will finish the section with a small time surplus, and that surplus is what gets spent on the one or two prompts in the section that genuinely require the full 2 minutes 30 seconds.
For most candidates, the score movement on Data Insights comes from the 78-84 band, and the bulk of that movement comes from converting the prompts that are currently being misread into prompts that are being read correctly. The protocol above is the read-side half of that conversion; the source-resolution rules in the disagreement section are the answer-side half. Together, they account for most of the score movement you can realistically expect from preparation.
As a final tactical note, when you sit down for the official GMAT Focus, do the first 3-4 prompts of the Data Insights section using the full five-pass protocol, even if they look like single-source prompts. That is your calibration block, and it sets the timing pattern for the rest of the section. Once the pattern is set, you can compress the protocol on the easy prompts and expand it on the hard ones, but the pattern itself should not change.
Conclusion and next steps
Mixed-source Data Insights prompts reward a specific read order: question first, text second, chart third, table fourth, answers fifth. The protocol protects your 2 minutes 30 seconds budget, prevents source re-reads, and surfaces the disagreements between sources that the test uses to separate the upper band from the mid band. Drilling the protocol on 10 mixed-source prompts at a time, with the source-resolution rules in mind, is the shortest path to a higher Data Insights score. TestPrep İstanbul's mixed-format drill set is the natural next step for candidates ready to put the protocol into timed practice on Multi-Source Reasoning and hard Graphics Interpretation prompts.