Reading a paper without reading all of it

Nobody reads papers front to back

Researchers do not start at the first word and finish at the last. They triage. For an undergraduate reading twelve papers for an essay, the difference between triaging and not is the difference between finishing and not.

The order that works:

1. Title and abstract. What did they do, what did they find. Decide whether to continue. Most papers stop here, and that is a successful outcome, not a failure.
2. Figures and tables, with their captions. This is the surprising one. The figures usually contain the actual result, and captions in a good paper are written to be readable alone. You can often learn what a paper found in ninety seconds this way.
3. The last paragraph of the introduction. Where the specific aim and hypothesis live.
4. The first and last paragraphs of the discussion. What they think it means, and what they concede.
5. The limitations section. Read this properly. It is where the honest authors tell you what the paper cannot support, and it is exactly what you need in order to cite it accurately.
6. Methods, only if the result matters to you. Then read it slowly, because this is where a study is strong or weak.
7. Results text, to check the numbers match what the abstract claimed. They sometimes do not.

Where the model helps and where it must not

The safe pattern: read the paper yourself in that order, and use the model on the parts you cannot follow, grounded on the pasted text.

Here is the methods section. Explain what a mixed-effects model is doing here and why they would have chosen it over a repeated-measures ANOVA. Use only what is in the text; say what is missing rather than filling it in.

That is a good use. It is a comprehension aid on material in front of you.

The unsafe pattern is asking for a summary and citing the summary. Two things go wrong. Summaries drop conditions and hedges (the in older adults only, the when baseline was controlled) and those qualifications are frequently the entire finding. And a summary you did not check gives you no page number and no sentence, so you cannot cite honestly.

The questions to ask of any study

Whatever the field:

• How many participants or observations, and how were they selected? A study of 24 undergraduates supports different claims from one of 24,000 adults.
• Compared with what? No control group means no causal claim. A comparison with nothing is not a comparison.
• Was anything randomised? This is the difference between associated with and caused.
• What exactly was measured? Wellbeing measured how? Self-report on one item is not the same as a validated scale.
• How many analyses were run? Twenty tests produce a significant result by chance.
• Who funded it, and who benefits from the finding? Not disqualifying; relevant.
• Is this the first report or a replication? First reports are much more often wrong than the confidence of their abstracts suggests.

A model can be asked to apply this list to a paper you have pasted, and that is a good use: it is applying a checklist to supplied text rather than recalling anything. Verify the answers against the paper.

Abstract inflation

Worth knowing as a specific hazard: abstracts systematically overstate. They are written to be read and cited, they omit the qualifications, and there is a documented tendency for abstract conclusions to be stronger than the results section supports. In several fields this has been measured, and the mismatch is not rare.

So if a claim matters, check it against the results, not the abstract. This is the single highest-yield habit in reading research, and it takes one minute.

Note-taking that survives

For each paper, four lines in your own words:

Claim. What they say they showed.

Evidence. Design, n, key number, with the page or figure.

Limitation. What it cannot support.

Use. Which part of my argument this is for.

Written in your own words, because a copied summary is a copying-into-the-essay accident waiting to happen and because paraphrasing is where you find out whether you understood. Zotero holds these notes attached to the reference.

Thirty papers read this way is a working knowledge of a literature. Thirty AI summaries is a folder.