You read the question, glance down, and two of the four choices instantly sound right. A minute later you've picked the smoother one and couldn't say why beyond "it felt right." That feeling isn't comprehension — it's the answer choices doing your thinking for you, and one habit shuts it down.
Why the answer choices are not information
On every Reading & Writing question, one choice is the answer and three are engineered to look like it — built from your own reflexes: the most familiar dictionary meaning, the most recognizable transition word, the passage fact that doesn't answer the question. Read them before you've committed to anything and you don't evaluate them — you audition them, and the most fluent one wins. That reflex is anchoring, the single behavioral root of misses across four question types. This page is the fix; the full anatomy of the leak lives in the answer-anchoring guide.
The real science: the generation effect
Predict-Then-Peek is not a folk trick. It leans on a documented finding from cognitive psychology called the generation effect: information you produce yourself is processed more deeply than information you merely recognize. Recognition is cheap — it feels like knowledge without being knowledge. Generation is effortful, and the effort is the point. A generated answer gives you a standard to judge four choices against; without one, you judge the choices against each other — the exact contest they were engineered to win.
The protocol: cover, commit, peek, match
Four steps, in order, every time.
1. Cover. Before your eyes reach the choices, stop. On paper, put a hand over them. On the digital test, it's eye discipline: read the passage and the question stem only.
2. Commit in plain words. Produce an actual answer and write it down: a word for the blank, a direction, a line in the passage. Two rules. Plain beats fancy — "confusing" is a perfectly good prediction; you don't need "labyrinthine." And the commitment must be concrete enough to be wrong: "something negative" is a vibe, not a prediction.
3. Peek. Now read the choices.
4. Match. You're looking for the choice that means what you wrote — not the one that sounds best. Anything that fails the prediction dies, however fluent it sounds.
The whole loop on an example:
Maria's directions were so ___ that even lifelong residents of the neighborhood got lost following them.
Cover the choices. What has to fill the blank? Something like confusing. Peek: (A) elaborate (B) convoluted (C) thorough (D) memorable. Match: convoluted. Notice elaborate — a plausible-sounding word for describing directions that would have auditioned well if you'd read the choices first. Against a committed prediction, it never gets a hearing.
How the prediction changes by question family
The protocol is constant; what you commit to changes shape.
- Words-in-context: predict one plain word for the blank. The classic trap offers a real dictionary meaning of the tested word that's wrong in this sentence — it only works on the uncommitted. The full method is in the words-in-context guide.
- Command of evidence: don't predict a word — point at a line. Before the choices, put your finger on the sentence in the passage that would prove the claim. The right choice must contain that line's fact, not a fact you happen to remember.
- Transitions: predict the direction first — does the second idea continue the first or push against it? That's The Same-or-Opposite Test, and it kills half the choices on contact.
- Inference: state, in your own words, the one thing that must be true given the passage. That protects you from the true-but-irrelevant choice: a statement that's accurate and still not an answer to the question.
The plug-back check: finish the rep
Matching isn't the last step. Plug your winning choice back into the sentence — or re-read the two sentences around the blank with it in place — and read the whole thing fresh. If it only works for half the logic, it doesn't work. Ten seconds, and it catches the near-miss matches that survive step four.
The self-test: ten questions, five covered
Run this today; it takes one practice set.
- Pick ten questions from the predict-first families (words-in-context, transitions, inference, evidence).
- On the first five, run the full protocol: cover, commit in writing, peek, match, plug back.
- On the last five, work the way you normally do.
- Compare accuracy.
Many students find the covered half better — which should be impossible with less information on screen, and isn't, because the choices were never information. If both halves are equally rough, be honest: that's not an anchoring problem, it's a content gap in that question type, and the cure is learning, not habit-catching. The behavior-versus-knowledge sorting logic is in the behavioral leaks guide.
Building it into daily practice
A protocol you run five times is a demo. A protocol you run daily for two weeks is a habit.
- Start untimed. The first sessions will feel slow. That's the generation effect being effortful — let it be. Speed returns as the prediction compresses from a written word to a beat of thought.
- Write the prediction down. For the first week, every prediction goes on scratch paper before you peek. A written word can't be quietly revised after the fact.
- Small daily sets beat a weekend binge. Ten predict-first questions a day. Spaced short sessions are where habits consolidate; a four-hour Sunday is where they don't.
- Track your slips. When your eyes jump early — they will — mark the question, finish normally, and afterward compare accuracy on predicted versus slipped questions.
- On test day, the cover is eye discipline plus scratch paper: read the stem, jot the prediction, then scroll to the choices.