How do I find a paper when I only remember the argument it made?

The reason this is hard is a mismatch between what you remember and what is indexed. You remember an argument, in your own paraphrase, stripped of the authors' vocabulary. Your reference manager indexes titles, authors, journals, and sometimes abstracts. The overlap between those two things is often zero, and no amount of persistence with keyword search will bridge it.

Full-text search is the first move that has a real chance, because the sentence you remember probably exists somewhere in the body of the paper even if the abstract never mentions it. If your library stores the PDFs, search inside them rather than across metadata. This alone resolves a large share of cases.

Before you type anything, break the remembered argument into parts. Bramer et al. describe a 15-step method that starts by naming the key concepts in the question, then listing synonyms, thesaurus terms, and variations for each one, and only then choosing syntax and a database (Bramer et al., 2018, p. 532). Do the same with your paraphrase. Write down the subject, the mechanism, the object, and the case or evidence, then give each part two or three alternative labels, including older terms the field has since dropped. They also recommend searching free text in titles and abstracts alongside indexed terms, because narrow topics are often described without the indexing term you expect (Bramer et al., 2018, p. 534).

When that fails, stop searching by content and start searching by context. Papers are read in clusters. Ask which chapter you were working on, roughly which month, and which other papers you read that week. Sorting your downloads folder or your library by date added and scanning twenty entries is faster than another hour of keyword attempts.

The third route is social. A distinctive claim, described in a sentence to someone in your field, is very often recognized immediately. This feels like an imposition and almost never is, since researchers enjoy the recall puzzle.

Which search moves work when keyword search fails?

Four. Full-text search rather than title and abstract. Citation chasing from a paper you know makes a similar argument. Searching your own downloads folder by date range. And asking a colleague, since a distinctive claim is often instantly recognizable to someone else in the field.

Citation chasing deserves more use than it gets. If you can name any paper that argues something adjacent, the paper you want is very likely in its reference list or among the works citing it. That reduces the search space from a literature to a few hundred entries you can scan.

Google Scholar is strongest once you are hunting a specific item rather than sweeping a topic. Across three environmental review case studies, Haddaway et al. found that article-specific Google Scholar searches located 61 of 84 grey-literature articles that hand-searching organisational websites had turned up, while none of those 84 appeared in the exported results of the broad searches, which covered 68 title-search records and a sample of 1,000 from 49,700 full-text hits (Haddaway et al., 2015, pp. 13-14). Search Scholar for the paper, with the distinctive phrase and a probable author, not for the subject area.

If the argument may have appeared in a report, thesis, or conference paper rather than a journal article, run a title search. Haddaway et al. measured 8 to 64% grey literature in Google Scholar title searches, averaging 40%, against 8 to 39% in full-text searches, averaging 19% (Haddaway et al., 2015, p. 7). Title search brings more screening noise, and it is also where less formally published work shows up. They recommend going through the first 200 to 300 results rather than stopping at 50 to 100, because grey literature was most abundant around page 35.

Once you have a candidate, switch from guessing to verification. Bramer et al. suggest checking whether a search actually retrieves references you already know are relevant, and confirming a match by combining a specific first author with a publication year and a page number (Bramer et al., 2018, p. 536). Then read the abstract, introduction, and conclusion to check that the author endorses the claim you remember rather than reporting it as a position under debate. The two sources qualify each other on scope. Google Scholar is effective for retrieving a known study, but Haddaway et al. found it missed 31.5% of relevant evidence in one case study and conclude that it should not stand alone for comprehensive searching (Haddaway et al., 2015, p. 15).

One thing that rarely works: asking a general-purpose chatbot to identify the paper. It will produce a plausible reference that does not exist, and you will lose an hour confirming that.

How do I stop this from happening again?

Record claims in a searchable place at the moment you read them, in the words you would use to search later. The reason the paper is unfindable is that your memory holds a paraphrase and your library holds only titles and metadata. Bridge that gap when the paper is in front of you.

The phrasing matters more than the completeness. Write the claim the way you would say it to a colleague, including the informal version, because that is what you will remember and therefore what you will type.

If you take nothing else from this, add one habit: every time you catch yourself thinking "this is the paper that shows X", write that sentence down with the citation. Those are precisely the papers you will hunt for later.