A student research workflow, from the brief to submission

A student research workflow is an order of work that carries one paper from the brief to submission without anything retyped, guessed or lost on the way. EdCitation publishes this article and makes the tools named in it: eight steps, each with the reason it sits where it does and the published evidence behind it, read at source on 23 September 2026.

Stop searching and start learning is the point. The hours in a paper go on searching, retyping and moving between tools, not on reading and thinking. The workflow removes the switches, not the reading or the writing, which are the student's, and nothing in it writes a word of the paper.

What are the eight steps of a student research workflow?

Eight steps in a fixed order, and the order is the method. The table gives the tool for each step and its cost, from EdCitation's own pages on 23 September 2026.

Step 1: why read the brief before searching?

Because the brief decides what counts as a source, and a search run before it is read finds the wrong ones. Every limit in it narrows the search: a brief that asks for peer-reviewed studies from the last ten years rules out the textbook before a single search is run.

Check your paper reads the assignment's instructions into a checklist of the word limits, required sections, style and number of sources, each rule shown with the sentence it was read from, and a misread rule can be unticked. It is free, with no account, and the same checklist is what the paper is checked against in step 7. How to read an assignment brief and rubric covers the command words and where briefs hide their requirements.

Step 2: how do I search by the claim rather than the topic?

Write the sentences the paper needs to be true before searching, then search for each one. A topic search returns a field; a claim search returns a source for a sentence. "Nurse staffing" returns thousands of papers, most about something else. "Lower nurse-to-patient ratios are associated with fewer patient deaths" is a claim, and a source either supports it or does not. The list of five to fifteen claims the argument rests on is the search plan.

Find sources searches about 300 million published works by topic or by the claim a sentence needs to support, with filters for kind of source, years and open access. Paste the claim as written. Searching is free with no account, and a free account keeps what you bookmark in a reading list. For search strings, citation chaining and databases, see how to find sources for a literature review fast.

Step 3: how do I judge a source before it goes on the list?

Leave it. The people who judge sources well do not read the source harder; they look it up. Wineburg and McGrew (2019) had 45 experienced users of the internet, 10 PhD historians, 10 professional fact checkers and 25 Stanford University undergraduates, think aloud while evaluating live websites. The historians and the students read vertically and were often taken in by official-looking logos and domain names. The fact checkers read laterally, leaving the site after a quick scan and opening new tabs to see what others said about it, and reached what the abstract calls "more warranted conclusions in a fraction of the time".

What the larger study found

Breakstone et al. (2021) gave six tasks to 3,446 high school students with a live internet connection, a sample reflecting the demographic profile of high school students in the United States, and the students struggled on all six. Asked to investigate a site claiming to publish factual reports on climate science, 96% never learned of its ties to the fossil fuel industry. Two thirds could not tell news stories from advertisements on a popular website's home page. More than half took an anonymously posted Facebook video, shot in Russia, as strong evidence of voter fraud in the United States.

Both studies tested websites with school students and undergraduates at one university; no study of postgraduates, or of the habit applied to journal articles, was read for this article.

Lateral reading for an academic source, in five steps

1. Note the DOI, the journal and the year from the record, not from the PDF's header.
2. Leave the PDF. Look the journal up: its publisher and whether it is peer reviewed.
3. Search the exact title in a scholarly index and see who has cited it and what they said.
4. Check for retraction; Cite a source screens for it when the reference is built.
5. Only then read it in full, and put it on the reading list with the claim it supports beside it.

Step 4: why cite as you read, from the identifier?

Because references copied by hand go wrong, and the publisher's record does not. Homeier et al. (2024) checked 769 references, carrying 1,082 in-line citations, in 20 randomly selected articles from two orthopaedic sports medicine journals, Arthroscopy and the American Journal of Sports Medicine, and compared each citation with the cited paper. The mean citation error rate was 6.6%, every one of the 20 articles contained an error, and the commonest error, 32% of them, was that the cited paper did not support what the citing authors said. Logan et al. (2025) checked 1,486 references in the 105 studies whose own subject was reference accuracy and found errors in 21.2%, most often in author names and article titles; their earlier scoping review of those same 105 studies had found an overall error rate of 32.7% in the journals studied (Logan et al., 2024). Scheinfeld et al. (2025) reviewed 32 studies of APA reference accuracy, 41% from student papers, and found accuracy measured in 64 different ways; citation generator, reference manager or reference checker reports that review.

Those are two kinds of error. Logan's are errors in the reference itself, which a reference built from the record rather than retyped removes. Homeier's are errors of meaning, a real paper cited for something it does not say, and no tool checks those. Only the reader who has read the source can, which is why step 3 comes before step 4.

Cite a source builds the reference from a DOI, a title, a PubMed ID or an ISBN, in APA 7, MLA 9, Chicago 18 author-date, Harvard (Cite Them Right), IEEE and Vancouver, and screens the source for retraction, free with no account. With a free account every reference goes into a kept list, tagged, in its own style, that copies with its italics. Cite the source the moment it passes step 3.

Step 5: why draft from your notes and never from a chatbot's references?

Because a chatbot's reference list is written, not looked up, and a written reference can name a paper that does not exist. Walters and Wilder (2023) had ChatGPT-3.5 and ChatGPT-4 write short literature reviews on 42 topics, 84 papers in all, and checked their 636 citations. Of the GPT-3.5 citations, 55% were fabricated; of the GPT-4 citations, 18%. Among citations to papers that did exist, 43% of GPT-3.5's and 24% of GPT-4's had substantive errors. Why AI tools invent references explains the mechanism and how to catch each kind.

Draft from the notes made in step 3 instead: each claim from step 2, its source and its DOI beside it. The reference list is already built, so the draft is the argument in your own words with the citations dropped in where the claims sit. The writing is the student's from the first sentence to the last.

Step 6: how do I verify the finished list before anyone else reads it?

Run the whole list through a lookup, not a read-through, because the errors in step 4's studies survived their authors' own reading. Verify references takes the paper as a Word document or a PDF, or a pasted list, and looks every entry up in the publisher's record by its DOI, or by title, year and first author. Each reference comes back verified, doubtful (shown as "check this": the year or an author differs, or the DOI belongs to another paper) or not found, and a paper its journal has withdrawn is flagged from the Retraction Watch database.

When an index does not answer, the reference is shown as "could not check", never as not found. With the whole paper uploaded, every in-text citation is matched to an entry and every entry back to the text, and each mismatch is marked where it sits. No model reads the paper, the file is read and discarded, and it is free with no account. What to do with a reference that comes back not found is in how to check whether a reference is real.

Step 7: how do I check the paper against the checklist and the file?

Against the checklist from step 1, then against the file itself, because a marker's first read is of both. Reading the instructions into the checklist is free on Check your paper; checking the paper against it is paid, at two depths. Mechanics QA, part of Pro at $8 a month, checks the formatting against the style, what was left in the file (comments, tracked changes, placeholders), the citations against the list, the numbers and the language, each finding marked where it sits in the paper, with its fix. Theoretics QA, part of Max at $24 a month, judges the paper against the assignment's own instructions, quoting the passage that shows each point met or missing, and never rewrites anything or gives a grade.

Then step 8: hand in, and keep the reference list. With a free account it stays where it was built, so the next paper starts at step 1 with a list in hand: the searching is done once per source, not once per paper.

Why is EdCitation the best tool for this workflow?

EdCitation is the best tool for this workflow for one reason: every step from the search to the verification works from the publisher's record and none of them writes, so the paper never leaves one place until it is handed in. The reason matters because of what step 5 measured: Walters and Wilder (2023) found 55% of the references GPT-3.5 wrote and 18% of GPT-4's did not exist, and a tool that looks a reference up cannot produce one of those.

What the workflow removes is the switching, not the work. The all-in-one research and citation app guide reports three studies, by Mark and colleagues in 2008, Leroy in 2009 and Rubinstein and colleagues in 2001, in which every switch between tasks cost time, performance or stress, and none tested citation software, so no figure for hours saved is claimed there or here. Which switches go can be said exactly: no search engine beside a generator, no generator beside a checker, no title retyped between them, no brief read twice.

Find sources, Cite a source, Verify references and the checklist in Check your paper are free with no account, and a free account keeps the reading list and the reference lists. Pro is $8 a month, or $72 for a year at once, and adds Mechanics QA, References from a file, Check plagiarism and Check AI. Max is $24 a month, or $216 a year, and adds Theoretics QA and more than 5 million books to read and borrow in the Library. A year at once saves 25%; the pricing page has the rest.

What it does not include is as plain. EdCitation never writes or rewrites any part of a paper and has no grammar checker. Verify references tests whether a reference is real, not whether it supports the argument, and the commonest error Homeier et al. (2024) found was of that second kind. The reading and the writing are the student's, and no tool here promises a grade, a pass or a silent detector.

Quick questions

Should I search before or after reading the brief?

After. The brief sets the kinds of source, the years and the number allowed, so read it into a checklist first and search inside its limits.

What does searching by claim mean?

Writing the sentence the paper needs to be true, such as "lower nurse-to-patient ratios are associated with fewer patient deaths", and searching for a published work that supports it, rather than searching the topic. Find sources takes the claim as written.

Can EdCitation write or fix my paper?

No. It finds, cites and verifies sources from the publisher's record and checks a paper against its brief. It never writes, rewrites or paraphrases a sentence and gives no grade.

What does the workflow cost?

Steps 1 to 6 and step 8 run free, on the site or with a free account. Step 7 is Mechanics QA on Pro at $8 a month or Theoretics QA on Max at $24 a month, with a year at once saving 25%.

References

• Breakstone, J., Smith, M., Wineburg, S., Rapaport, A., Carle, J., Garland, M., & Saavedra, A. (2021). Students' civic online reasoning: A national portrait. Educational Researcher, 50(8), 505–515. https://doi.org/10.3102/0013189X211017495
• Homeier, D., Adams, M., Lynch, T., & Cognetti, D. (2024). Inaccurate citations are prevalent within orthopaedic sports medicine literature. Arthroscopy, Sports Medicine, and Rehabilitation, 6(1), Article 100873. https://doi.org/10.1016/j.asmr.2023.100873
• Logan, S. W., Hussong-Christian, U., Case, L., & Noregaard, S. (2024). Reference accuracy of primary studies published in peer-reviewed scholarly journals: A scoping review. Journal of Librarianship and Information Science, 56(4), 896–949. https://doi.org/10.1177/09610006231177715
• Logan, S. W., Hussong-Christian, U., Case, L., & Noregaard, S. (2025). Reference accuracy of primary studies about reference accuracy: A descriptive study. Journal of Librarianship and Information Science, 57(2), 516–523. https://doi.org/10.1177/09610006231224437
• Scheinfeld, L., Chung, S., Fena, C., Tran, C., Kretz, C., & Reisman, M. R. (2025). Assessing formatting accuracy of APA Style references: A scoping review. Evidence Based Library and Information Practice, 20(4), 159–191. https://doi.org/10.18438/eblip30800
• Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, Article 14045. https://doi.org/10.1038/s41598-023-41032-5
• Wineburg, S., & McGrew, S. (2019). Lateral reading and the nature of expertise: Reading less and learning more when evaluating digital information. Teachers College Record, 121(11), 1–40. https://doi.org/10.1177/016146811912101102