How do I draft when I keep stopping to look up citations?

The interruption cost is the real problem, and it is larger than it feels. Writing an argument means holding several things at once: the claim, the next two sentences, the objection you are pre-empting, the paragraph's destination. Opening a reference manager and searching for a paper drops all of it, and the thirty seconds you spent looking becomes five minutes of getting back.

Do that eight times in an hour and you have written two paragraphs. The same hour with placeholders produces a page and a half, and the lookups are then done in a batch where the context cost is paid once.

Protecting the drafting block is the other half of this. Peterson and colleagues recommend writing daily or three to four times a week in a protected block of 15 to 30 minutes, set as a specific goal such as 8:30 to 9:30 on weekdays, and they warn that constant interruptions disrupt thinking and eat the time you had for writing (Peterson et al., 2018, p. 2). Their fixes work the same way a placeholder does: postpone small tasks, decline nonurgent requests, turn off email and internet access. A citation lookup is a small task, and it belongs on the postponed list with the rest of them.

The placeholder has to carry enough to identify the source later. A bare marker leaves you staring at a sentence with no idea which of four papers you meant. Ten words is enough: the author if you remember it, otherwise the finding and the study type, which is almost always enough to find it in your own notes.

There is a second benefit that surprises people. A draft full of placeholders makes visible how much of your argument rests on evidence and how much is your own reasoning. Sections with no placeholders at all are either your original contribution or claims you have not noticed need support.

Deferring the lookup also improves the citation you eventually make. Penders argues that a citation identifies where an idea, method, question, or critique came from, and lets the reader judge whether you used the source appropriately, so he asks you to cite transparently rather than neutrally, saying in the prose whether a study establishes a finding, supplies context, or challenges an assumption (Penders, 2018, pp. 2-3). You cannot do that at the moment the sentence stalls, because you do not yet know the final shape of the claim. Zhang makes the related point that drafting is not a transcription step, and that writing a paper can expose weaknesses in the design, methods, or interpretation that send you back to revise the work itself (Zhang, 2014, p. 1). A claim that changes during revision does not need a source found early.

What placeholder system actually works?

A distinctive marker plus enough detail to find the source later. Something like @@CITE Fenton longitudinal retention, three-year effect@@ rather than a bare cite tag. The marker must be searchable and must never appear in normal prose, so a character pair you would not otherwise type.

A doubled at sign works in every editor, survives conversion between formats, and is easy to count, which lets you track how many are outstanding. A highlight colour works too, as long as it survives export.

Whatever you choose, use one form for the whole thesis. Three placeholder conventions invented over three years means the final search misses some, and a submitted chapter containing a live placeholder is a memorable kind of embarrassment.

When do I go back and fill the citations in?

At the end of a section, not at the end of a sentence and not at the end of the thesis. Section-level filling keeps the context fresh enough that you remember which paper you meant, and it is small enough to finish in one sitting, which is what stops a backlog forming.

Filling is a different mode from writing and benefits from being scheduled differently. It is mechanical, it tolerates interruption, and it can be done on the kind of low-energy afternoon when drafting is impossible.

Work the list by section rather than in the order the questions occurred to you. Penders advises prioritising citations separately for each section and each publication, since reference limits can be severe, as low as three for a Nature correspondence item, and dropping references at random is the wrong response (Penders, 2018, pp. 3-4). In an introduction, start with review articles for broad context. In a discussion, prioritise empirical papers, which give the detailed account of what each study contributed. Penders also suggests limiting a specific claim to a single citation where you must, preferring reviewed over unreviewed material, and moving methodological references to supplementary information (Penders, 2018, p. 4). That cuts the size of the lookup list, but it does not remove the obligation to verify the claims your argument actually rests on. Peterson and colleagues add the tracking habit: keep the writing on a calendar, expect productivity to fluctuate, and name the specific obstacle when progress stops (Peterson et al., 2018, pp. 4-5).

Before submission, search for the marker one final time across every file including appendices. This takes one minute and is the cheapest insurance available at that stage.