SAT inference and command of evidence questions, explained

The three skills in Information and Ideas

Skill names from the Assessment Framework for the Digital SAT Suite, table 10, checked 6 October 2026.

The key idea: prove it from the text

Every passage is short, between 25 and 150 words, and every question has one best answer. For inference and evidence questions, the wrong answers are usually not silly. They are typically too strong ('always', 'proves', 'only'), about a related but different claim, true in the world but not stated, or accurate to a graph but irrelevant to the claim. Your job is to find the answer you could defend by pointing at specific words or numbers.

Read the last sentence of the passage and the question first, because they tell you exactly what is being claimed. Then read the whole text and predict the answer in a few words before you look at the choices. A prediction protects you from the attractive wrong answer that happens to use the passage's vocabulary.

For evidence questions, write the claim down in its shortest form and circle any qualifier, such as 'in dry regions' or 'over time'. A finding that supports the general topic but ignores the qualifier is a classic wrong answer.

Worked example 1: an inference question (original, invented study)

Text: In a study, one group of students reviewed vocabulary flashcards in short daily sessions, while a second group studied the same cards in one long session. On a test the day after studying, the two groups scored about the same. On a test a month later, however, the daily-session group recalled noticeably more words. These results suggest that ______ Which choice most logically completes the text? A) long study sessions do not help students learn vocabulary. B) the advantage of short daily sessions may appear mainly in longer-term recall. C) most students prefer short study sessions to long ones. D) the test given the day after studying was too easy to be useful.

1. Summarise the facts: same score after one day, daily group better after a month.
2. Predict: the benefit of daily study shows up later, not straight away.
3. A is too strong. The long-session group did learn words; they matched the other group the next day.
4. C introduces preferences, which the text never mentions.
5. D blames the test, but nothing in the text says it was too easy.
6. Answer: B. The word 'may' keeps it as cautious as the evidence, which is a good sign in inference answers.

Worked example 2: textual evidence (original, invented researcher)

Text: A biologist hypothesises that a desert beetle collects drinking water from morning fog using the small bumps on its back, where droplets form and then roll towards its mouth. Which finding, if true, would most directly support the hypothesis? A) The beetles are most active in the hours before sunrise. B) Beetles whose bumps were coated with a smooth film collected far less water from fog than uncoated beetles did. C) Fog occurs on many mornings in the beetle's habitat. D) Several other desert insects do not have bumps on their backs.

1. The claim has two parts: water is collected from fog, and the bumps are what do it.
2. A fits the timing of fog but says nothing about the bumps.
3. C shows fog is available, but not that the beetle uses its bumps.
4. D is about other insects and tells us nothing about how this beetle collects water.
5. B changes only the bumps and shows water collection falls. That tests the mechanism directly. Answer: B.

Worked example 3: a quantitative evidence question (invented data)

A student surveyed teenagers and grouped them by average daily screen time. The student claims that, among those surveyed, more screen time was associated with less sleep. Which choice most effectively uses data from the table to support the claim?

Choices: A) Group 2 slept 7.6 hours a night on average. B) Group 3 averaged 6 hours of screen time a day. C) Group 3, with 6 hours of screen time, averaged 7.0 hours of sleep, compared with 8.1 hours for Group 1, with 2 hours. D) All three groups averaged at least 7 hours of sleep. Answer: C. A, B and D are accurate to the table, but only C compares screen time and sleep across groups, which is what an association needs.

Common mistakes

• Choosing an answer that is true in real life but not supported by the passage. Background knowledge is not evidence.
• Picking an answer with strong words (proves, always, never, only) when the text is cautious.
• In data questions, choosing a choice that reads the table correctly but does not address the claim.
• Ignoring a qualifier in the claim, such as a place, time period or group.
• Reading the choices before predicting, then being pulled towards one that repeats words from the passage.
• Confusing support and weaken: read the question stem twice before answering.

Exam technique and how a tutor helps

These questions take longer than grammar or transitions, so bank time on those and allow up to 90 seconds here. If two choices remain after a minute, ask which one could be wrong given the text; the one that could be wrong is the wrong one. There is no penalty for a wrong answer, so never leave a question blank, and use the flag tool to return if time remains in the module.

Inference and evidence skills improve more slowly than grammar because they depend on reading habits. In one-to-one lessons the tutor has the student predict aloud before seeing the choices, then explain in one sentence why each eliminated option fails ('too strong', 'off topic', 'not in the text'). That turns a vague sense of doubt into a reason, and after a few weeks most students eliminate faster. For data questions, the tutor drills reading titles, axes and units first, because many errors come from reading the wrong row or column.

Self-check

• What is wrong with an inference answer containing 'proves that'? (It is usually stronger than a short text can support)
• In a data question, what two tests must the right answer pass? (Accurate to the data and relevant to the claim)
• A claim says a plant grows faster 'in salty soil'. Does a finding about growth in all soils support it? (Not directly; it ignores the qualifier)
• Should you read the choices before predicting? (No, predict first)
• How long should you allow for an evidence question? (Up to about 90 seconds, using time saved elsewhere)