How do I remember what was in papers I read months ago?

Forgetting is not a failure of attention. Reading a paper builds an understanding that lives in working memory and dissolves within weeks unless something durable was produced. The durable thing has to be written, and it has to be written by you, because the act of putting the finding in your own words is what makes it retrievable later.

This is why summarizing the abstract does not work. Copying the authors' phrasing produces text you can read but did not think, and reading it back six months later restores none of the understanding. A sentence you constructed reactivates the reasoning you did.

Give the writing its own pass. Carey, Steiner and Petri describe active reading as effortful reading undertaken with the intention to understand, and they set out a three-pass procedure: read once for general exposure without trying to critique everything, read a second time for comprehension, and read a third time while taking notes (Carey et al., 2020, p. 1). Notes written on the third pass are notes written by someone who already knows what the paper argues. When the paper leans on unfamiliar terminology, supplementary material, or an earlier study, they tell you to go and read those rather than treat the main text as self-sufficient (Carey et al., 2020, p. 5).

The other reason people forget is that they recorded the wrong unit. Notes made per paper answer "what is this paper about", which is a question you will never ask. At writing time you ask "what supports this claim", and a library of paper-shaped notes cannot answer that without rereading all of them.

So record claims rather than papers, and record why you cared. "Effect disappears after three years, contradicts my chapter two argument, page 412" is a note that works in six months. "Interesting longitudinal study of retention" is not.

Researchers who read for a living say the same thing about purpose. In Hubbard and Dunbar's survey of 88 respondents who gave reading advice, 6 told readers to start from a specific purpose or an explicit question, and 30 recommended critical reading, of whom 18 said to check whether the conclusions actually match the data and 5 said to interpret the data yourself (Hubbard and Dunbar, 2017, pp. 8-10). That gives you the fields worth recording: the paper's question, the result, what the data support as opposed to what the authors conclude, and one thing it leaves open for your project.

What should a literature note contain so it is still useful later?

The finding in your own words, the page it sits on, and a line saying what you thought of it. Three fields. Bibliographic data your reference manager already holds does not need repeating, and a summary of the whole paper is less useful than the one claim you will actually cite.

The judgment line is the field people leave out and miss most. Your assessment at the time of reading, when you had the whole paper in view, is information you cannot reconstruct later without rereading. Two or three words is enough: "small sample", "only direct test", "authors overstate this".

Record the page as the printed page of the journal or book, not the position in your PDF viewer. Those diverge as soon as front matter is included, and a page number that does not match the published article is worse than none.

Be honest about what this buys you. No study here tests whether a particular note format produces accurate recall months later. Hubbard and Dunbar surveyed 260 biological sciences students and researchers at a single UK research-intensive university, and they caution that the sample may not transfer to other settings (Hubbard and Dunbar, 2017, p. 1). What their data do show is that reading habits change with experience. Papers found independently were read more than once a week by 16% of second-year undergraduates, 41% of PhD students, and 63% of academics, and less experienced readers found the methods and results hardest and undervalued critical interpretation of results (Hubbard and Dunbar, 2017, pp. 7-8). The note fields above are worth adopting because they force the reading that experience otherwise takes years to produce, not because a trial proved they beat highlighting.

Why do highlights fail as a memory system?

Because highlighting records that a passage seemed important, not why. Six months later you get a yellow sentence with no context and have to reconstruct your own reasoning from scratch. Highlights also cannot be searched across a library or grouped by argument, which is what you need at writing time.

There is a second problem. Highlighting feels productive while requiring almost no thought, so it substitutes for the harder work of stating what the passage shows. A paper covered in yellow often signals a paper that was not really processed.

Highlights do have a real use, as a marker for a second pass. Highlight while reading, then spend five minutes afterwards turning the highlights into two or three written claims, and delete nothing. The highlights get you back to the passage, the claims are what you retrieve.

When you do go back, go back selectively. Sixty of Hubbard and Dunbar's 88 respondents recommended selective reading and 28 recommended reading sections in a set order, but there was no agreement on which order. Sixteen started from the abstract and 6 from the figures, some proposed abstract, then figures, then text, others proposed abstract, introduction, discussion, then figures, and 3 said to skip the methods entirely (Hubbard and Dunbar, 2017, p. 8). So do not adopt one fixed sequence. Reopen the sections that carry the claim you need, which is usually the figures, results, and the methods detail your own question depends on.

The other route back into a paper is your own work. Carey, Steiner and Petri suggest connecting an article to other papers and to your project, implementing one of its techniques, or testing or extending its hypothesis through a wider literature review, and they note this often sends you back to the article later with new criticisms (Carey et al., 2020, p. 6). Retrieval you do while comparing two studies or justifying a method choice holds better than retrieval you schedule for its own sake.