A seminar paper gets marked in 20 to 40 minutes. Not because your supervisor is lazy, but because 25 papers sit alongside teaching, research and committee work. In that time nobody reads 18 pages word by word.
Marking is done by sampling, and the samples are not random.
You probably know the citation rules already. What gets explained less often is which five places get checked first and why. Once you know that, you can run the same pass over your own work in half an hour, in the same order.
In short: When a seminar paper is marked, the bibliography gets skimmed first, then two or three references are sampled and looked up in the original. The samples almost always land on the reference supporting the core claim, on the most recent source, and on titles that fit the research question unusually well. Running that same pass yourself takes about 30 minutes.
What actually happens during marking
The typical sequence looks like this:
- Bibliography first. Two minutes of overview: size, currency, proportion of citable publications, obvious gaps.
- Introduction and conclusion. Does the research question hold, is it answered at the end.
- Two or three reference samples at the central arguments.
- A pass through the main body, watching the line of argument and the density of support.
- Formalities, usually last and quickly.
The bibliography comes first because two minutes with it reveal more about a paper than ten pages of text. And the samples land almost always in the same place: at the claim your thesis rests on.
An uncomfortable rule follows: the reference you checked least is statistically the one that gets checked. Central claims get written from memory, because those are the ones you feel sure about.
Before you go through individual references, the rough pass over the whole bibliography is worth it: free check, no signup.
The five places that get checked first
1. The core claim and its reference
Every seminar paper has one sentence everything rests on. The reference to it gets read against the original, almost every time.
Check it yourself: open the page. Is there a sentence that supports your formulation? Not “fits the topic”, but supports.
2. The most recent source cited
Submitting in 2026 with 2015 as your newest source invites a question. The opposite too: a strikingly fresh source often gets looked up precisely because it is new.
3. The source with the most unusual title
A title that mirrors your own research question attracts attention. Generative models produce exactly such titles particularly readily, because they are optimised against the prompt.
The scale of the problem shows in a study by Rossiter: of 2,300 AI-generated references analysed, 71 per cent were faulty. Not as an edge case, as the majority. A source out of a chat log is statistically more likely wrong than right.
Jana, a politics student, cited an article for her seminar paper on municipal participation whose title contained her research question almost verbatim. Her supervisor read the title, thought “I do not know that one, I want to read it”, and searched for it. The journal existed, the volume existed, the article did not. Jana had taken it from an AI-generated reading list in October and never opened it.
4. Secondary citations
“As cited in” signals that nobody consulted the original. That is permitted when the original is genuinely unobtainable, but it invites sampling, because errors cluster there. When it is acceptable is covered under using secondary citations correctly.
5. Web sources without a stable address
A link with no DOI, no access date, no archived copy. If it is dead by marking time, the reference is worthless. How to prevent that is under avoiding link rot.
Fastest self-test: paste your bibliography into the quick check and see within about ten seconds which entries can be verified in catalogues and databases. Free check, no signup →
Citing in the text: three forms, three purposes
The formal rules are in your departmental guidance. What is rarely stated there: when each form is appropriate.
| Form | When | Typical error |
|---|---|---|
| Direct quotation | When the wording itself is the object | Too frequent; reads as shallow engagement |
| Paraphrase | The default in a seminar paper | Too close to the original, effectively plagiarism |
| Collective reference (“see also”) | State of research, overview | Does not cover a specific claim |
The rule of thumb for seminar papers: five to ten per cent direct quotation at most. More reads as a collection of material. The difference between the forms is explained in direct vs. indirect quotation.
The real danger sits with the paraphrase. Reworded too closely to the original, it is plagiarism even with a reference attached. How far you have to go is covered in paraphrasing without plagiarism.
The bibliography: what stands out in two minutes
| Signal | What it suggests |
|---|---|
| Only monographs, no articles | Research through the library catalogue, not through databases |
| Only articles, no foundational works | No theoretical grounding |
| Everything from one journal | Research through a single result list |
| Strikingly many titles from one year | Taken from somebody else’s reading list |
| Inconsistent formatting | Pasted together from different sources |
| Titles with no publisher or place | Metadata never checked |
None of these is an error in itself. Together they create an impression that shapes how the whole paper is read.
The counter-test takes five minutes: sort your bibliography once by year and once by publication type. If a cluster appears, you know which research route you over-used.
Which databases are worth using beyond Google Scholar is covered in the comparison of academic databases.
The seminar-paper-specific error: references without a position
In a first-year essay, reporting the literature is still acceptable. In a seminar paper at honours level it is not. What is expected is a position of your own, supported rather than asserted.
In practice the error looks like this: three paragraphs of literature summary, each neatly referenced, followed by one sentence beginning “It follows that …” with no reference, because it is your own conclusion.
That is correct as it stands, but it has to be made visible. Signal phrases help:
- “The literature emphasises X (Smith 2021; Jones 2023). What it overlooks is that …”
- “Against this reading stands the fact that …”
- “For the present research question it follows that …”
That way the person marking can see at a glance where the literature ends and your contribution begins. Without those markers, a well-referenced paper reads as a summary and gets marked accordingly.
Tobias, an education student, submitted a carefully researched seminar paper with 24 sources. The feedback: “solid command of the material, but where do you stand?” He had stated his position in four places, just without signal phrases, buried between summaries. On the second attempt he changed nothing substantive, only the markers. One grade better.
Before you submit: check the references attached to the places where you take a position. Those are the ones that get read against the original. Check your bibliography now →
Fix the citation style once, not three times
The most expensive avoidable error in seminar papers has nothing to do with content: switching style mid-way.
The typical course of events: you start with APA because you know it from last term. In week three you discover this department wants footnotes. You switch, but not everywhere, and the bibliography ends up carrying two formats.
That costs twice. It is immediately visible in the two-minute bibliography scan, and it costs you an hour of visual checking you would rather have spent on the main body.
Three moves before you set your first citation:
- Find the departmental guidance, not the university’s. Requirements are almost always set at department or chair level.
- Set the style in your reference manager, do not carry it in your head. In Zotero under
Edit → Settings → Cite → Styles. - If there is no requirement, pick a common style and hold it consistently. Consistency counts for more than the choice itself.
Which style is usual in which field is covered in the comparison of citation styles. The two most common have their own guides: APA 7 and the Harvard style. Whether your field expects footnotes or parenthetical references is settled in footnotes vs. parenthetical citations.
Special cases in seminars
Classmates’ presentation slides. Not citable. If a publication appears on them, cite the publication.
Unpublished seminar papers by others. Generally not citable, and risky: if it came out of the same department, the person marking knows it.
Grey literature. Reports, working papers and government publications are often the most current sources on a topic. Citable, but with particular care over the details: see citing grey literature.
Non-English sources. Cite in the original, with a translation in brackets or a footnote. Details under translating and citing foreign-language sources.
AI-generated text suggestions. These belong in the declaration of authorship, not in the bibliography. What goes where is covered under AI disclosure.
How many references per page
A question with no official answer and an expectation anyway. What reads as unremarkable in seminar papers:
| Part of the paper | Reference density | Reason |
|---|---|---|
| Introduction | low | The research question is yours, not the literature’s |
| Theory / state of research | high, 2 to 4 per page | This is where you report |
| Analysis / your own part | medium | References support, they do not carry alone |
| Conclusion | very low to none | New references in a conclusion read as an afterthought |
Two deviations stand out: a theory page with no references at all, and a conclusion full of new sources. Both read as a break in the argument.
More important than the number is the distribution. Five references on one page and none on the next three suggests a section drawn from a single source.
Your own pass, in the same order
Go through your paper in the order it gets marked:
- Skim the bibliography: size, years, publication types, formatting consistent
- Find the core claim and look up its reference in the original
- Check the most recent and the most unusual source for existence
- Go through every secondary citation: is the original genuinely unobtainable
- Click every link, add access dates, archive the critical pages
- Reconcile text and bibliography in both directions
- Mark your own position with signal phrases
- Estimate the share of direct quotation, cut above ten per cent
The whole thing runs 30 to 45 minutes. Most of it is existence checking, and that can be done for the entire bibliography at once.
What to take away
- The bibliography gets read first. It shapes how the whole paper is read.
- Samples land at the core claim. That is exactly where people cite from memory.
- Unusually well-fitting titles get looked up. And they are the typical output of AI research.
- Direct quotation sparingly. Five to ten per cent is enough.
- Mark your position. Without signal phrases the best research reads as a summary.
A seminar paper rarely fails on missing knowledge. It fails on references that do not hold when looked up, and on a position that is present in the text but invisible. Both cost you under an hour before you submit.
Start with the part that runs automatically: paste your bibliography into the quick check and see which entries can be verified. Free, no signup.