How to Clean Up a Half-Finished Draft (Import It, Fix the Citations, Finish It)

To fix a messy draft, work in one honest order: rebuild the structure first, then verify every citation, then fill the gaps, then polish the prose. Most half-finished papers fail in the same shape, a skeleton that does not quite hold and a reference list nobody has checked, and trying to fix everything at once is why the cleanup stalls. Sort the structure before you touch a sentence, because moving a paragraph is cheap and rewriting one you are about to delete is wasted. Then make every citation real and every claim supported. The polishing that feels like progress is the last step, not the first, and doing it earlier is how a messy draft stays messy.

This is the situation almost every draft lands in: you have a few thousand words, some of them written by you and some by a chatbot, the structure is roughly there but not quite, and the citations are a mix of real, wrong, and never checked. It is not a blank page, which is good, but it is not finished either, and the gap between the two feels larger than it is. The way across is to stop seeing one big mess and start seeing four jobs in order. Let's take them one at a time.

Why the order matters

The instinct with a rough draft is to open it at the top and start fixing whatever catches your eye: a clumsy sentence here, a missing citation there, a heading that sounds off. That feels like work, but it is the slowest possible route, because the small fixes you make early get undone by the big decisions you make late. Polish a paragraph, then realise the whole section belongs somewhere else, or does not belong at all, and the polishing is gone with it.

So work from the largest unit to the smallest. Structure is the largest: it decides what stays, what moves, and what goes, and every later decision depends on it. Citations come next, because a section you have decided to keep needs sources you can trust before you build on it. Then the gaps, the thin spots where the argument is asserted but not supported. Polish is the smallest unit, the sentences themselves, and it only makes sense once you know which sentences are staying. Each pass protects the work of the next. That is the whole logic, and it is the same order a strong paper is built in the first place, which we walk through in how to write a research paper.

Step 1. Fix the structure first

Read the draft once for the skeleton alone, not the sentences. You are answering one question: does the order of the sections make an argument, or is it just a pile of things you knew? Map it out, the headings in sequence, with a one-line note under each saying what that section is supposed to do. The moment you write those notes, the problems show themselves. Two sections arguing the same point. A results section before the reader knows the method. A heading with nothing under it but a placeholder. A paragraph that wandered into the wrong section months ago and never left.

Then fix the skeleton. Reorder sections so each one sets up the next and the whole thing builds toward your point instead of listing facts. Merge the two that overlap. Cut the section that does not earn its place, even if it is well written, because well-written and irrelevant is still irrelevant. Mark the gaps where a section is missing entirely so you know to come back to them. This is the pass where the draft stops being a heap and becomes a shape. Everything after it is filling in that shape, which is far easier than fighting it.

Do this before anything else for one practical reason: structural changes are the expensive kind. If you verify a citation in a section you later delete, you verified it for nothing. If you polish a paragraph you later move and rewrite, same waste. Get the skeleton settled and you only do the careful work once, on the parts that are actually staying.

Step 2. Verify every citation

Now the references, and this is where the real risk in a draft lives, especially if any of it came from a chatbot. A general AI tool predicts citations the same way it predicts any other text, so it produces references that look flawless, a believable author, the right journal, a correctly formatted DOI, that point at papers which were never written. The prose around them holds together while the citations quietly do not. We cover that specific failure in used ChatGPT for your essay? fix the citations; here it is one step in the larger cleanup.

For each reference, check two things in order. First, does the source exist. Paste the exact title in quotation marks into Google Scholar; a real paper turns up as the top hit, and a fabricated one returns nothing. If the entry has a DOI, paste it after https://doi.org/ in your browser; a real one loads the article's page, a made-up one returns a "DOI Not Found" error. A perfectly formatted DOI that leads nowhere is one of the most reliable tells, because the format is exactly the part a model gets right.

Second, and this is the step people skip, does the source actually support the sentence it is attached to. Open the paper and read the relevant part. A real source cited for a claim it never makes is still wrong, and a careful marker reads it as a misuse, not an honest slip. Both failures, the invented paper and the misattributed one, are caught the same way, by opening the source and confirming it backs your point. The full routine, with the signals that should stop you on an entry, is in how to check if a citation is real.

Every reference then lands in one of three buckets. Keep it if the paper is real, the details match, and it supports the claim. Fix the metadata if the paper is real and supports your point but a detail is off, correcting it against the publisher's page rather than what the draft says. Replace it if the paper does not exist or does not back the claim, by deleting it and finding a real source you have actually read. Do not swap a fake for another plausible-looking reference you have not opened; that keeps the exact risk you were removing, just in a tidier wrapper.

Step 3. Fill the gaps

With the structure settled and the citations honest, the holes are now visible: the sections you marked as missing, the claims that have no source behind them, the places where the argument asserts something and just moves on. This is where a half-finished draft is most obviously half-finished, and it is the step that actually advances the paper rather than tidying it.

Work claim by claim, not citation by citation. For each unsupported sentence, ask what would genuinely back it, find a real paper that does, read enough of it to be sure, and attach that. Sometimes the claim survives with a solid source behind it. Sometimes you discover the point was overstated and needs softening, which is useful to learn now rather than in feedback. And if you genuinely cannot find any source for a claim, that is information too: the claim itself may not hold, and the honest fix is to change it, not to prop it up with a reference that does not quite fit. Source first, then the sentence; when the source comes before the claim, there is nothing to fabricate.

Step 4. Polish last

Only now do you touch the sentences, because only now do you know which ones are staying. Polishing is the satisfying part, the part that feels like writing, which is exactly why it is so tempting to do first and so wasteful when you do. A paragraph you tighten and then delete is effort spent on nothing. Save it for the end, when the structure is fixed, the citations are real, and the gaps are filled.

Then read for the prose alone. Tighten bloated phrasing, break run-on sentences, cut filler and hedging, and make sure each paragraph flows into the next so the draft reads as one continuous piece rather than a stitch of passages from different sittings, which a reassembled draft often is. Read a section aloud; your ear catches clumsy rhythm your eye skims past. Proofread for spelling, grammar, and consistent formatting last of all, ideally a day later with rested eyes. The draft that arrived as a mess now reads like one paper, because you fixed it in the order that let each pass hold.

How importing the draft into CiteOwl does the heavy lifting

Every step above you can do by hand, and the order is the same whatever tool you use. Where CiteOwl helps is that you can hand it the draft you already have instead of rebuilding it in a new document. You import the file you started in, a PDF, a Word document, or a LaTeX project including a zip, and CiteOwl reads it and rebuilds it for you.

Rebuilding means a few concrete things. It reconstructs the structure, turning your headings and paragraphs into real sections you can reorder and edit, so the skeleton you would otherwise rebuild by hand is already there. It pulls figures and equations out of the original and brings them across as real elements rather than flattened images. And it re-links the citations against real databases, matching each reference to a source it can actually find and flagging the ones that do not resolve, which is the verification pass from Step 2 done as the draft comes in. You start the cleanup with the structure mapped and the shaky citations already surfaced, instead of hunting for them.

From there it helps you fix and finish the draft section by section, and this is the part that keeps you in control: every change it makes arrives as a reviewable diff, the old text and the new side by side, that you accept or reject. Nothing is rewritten silently behind your back. When it proposes a structural change, a corrected citation, or a filled gap, you see exactly what it did and decide whether it stays. The honest version is that importing does not finish the paper for you; it does the mechanical reconstruction and the first verification pass, then puts you in front of a draft that is ready to fix, with the changes laid out so the judgement stays yours.