Blog

· By Thesisroom

Thesis Checker: What Each Kind of Check Looks For

Six kinds of thesis checker compared: language, similarity, references, statistics, structure and AI-writing scores, with what university guidance says and how to run each check by hand.

A black magnifying glass resting on a small open book on a wooden desk beside a larger book
Photo by Harshit Suryawanshi on Unsplash

Products sold as a thesis checker do quite different jobs. This guide covers six kinds of check: language, text similarity, references, reported statistics, structure, and scores for machine-generated writing. Pick the check that answers the worry you actually have, rather than the checker with the longest feature list.

For each kind, this guide sets out what the check looks for, what the published guidance says about it, and how to run the same check by hand.

Thesis checker types at a glance

Start from the problem you want to rule out, then pick the check that tests for it.

What you want to rule outKind of checkWhat it looks for
Typos and grammar slipsLanguage checkSpelling, punctuation and basic grammar
Unattributed copyingSimilarity checkText that matches other sources
Fake or wrong referencesReference checkWorks that cannot be found, and details that do not match the record
Wrong p-valuesStatistics checkP-values that do not fit their test statistic and degrees of freedom
Broken numbering and contentsStructure checkGaps in chapter, figure and table numbers, and contents entries with no matching heading
Questions about authorshipAI-writing scoreHow much of a text a detector labels as AI-written

The sections below explain each row.

1. Language checks: spelling, punctuation and grammar

A language checker flags spelling, punctuation and grammar errors. Your university may limit how you use one on assessed work.

The University of Birmingham’s guidance on using Grammarly responsibly allows fixing spelling, punctuation and basic grammar, and checking formatting and layout consistency. It does not allow rewriting or improving your arguments, reordering content to improve its structure or logic, or rewording to sound more academic.

Birmingham also says Grammarly now offers advanced features beyond basic grammar and style checks, and that students must not use these for assessed work. Read your own university’s policy before you accept a suggestion that changes more than a word. For more on where a grammar tool stops, see is Grammarly enough for a thesis.

Check it by hand: read the thesis aloud one chapter at a time, or change the font and line length so familiar pages look new.

2. Similarity checks: matching text, not a verdict

A similarity checker compares your text with other sources and highlights what matches. The match is where an examiner’s judgement starts.

De Montfort University’s Turnitin guide for internal examiners explains how this works for theses. Turnitin looks for any sequence of 7 words that already appears in other sources, and the score is the overall percentage of matching text. The guide states there is no threshold figure for the score that reliably indicates plagiarism or not.

The same guide tells examiners that the key is to look at how the highlighting is distributed. Scattered short matches are described as linguistic coincidence. Long runs over several sentences suggest copied text. It also says students cannot be guilty of plagiarising their own publications.

For what a similarity score means for a thesis, see what Turnitin similarity score is acceptable.

Check it by hand: for every passage you took from a source, confirm it is either in quotation marks with a citation or fully in your own words with a citation.

3. Reference checks: does each source exist and match

A reference checker looks up each entry in your bibliography to see whether the work exists and whether your details match the record. References copied from generated text are one reason to run it.

Walters and Wilder tested this in Scientific Reports in 2023. They asked ChatGPT-3.5 and ChatGPT-4 for short literature reviews on 42 topics and checked the 636 citations produced. They found 55% of the GPT-3.5 citations and 18% of the GPT-4 citations were fabricated.

Finding a work in the record is the first step. Read the part of the source you rely on before you cite it.

Check it by hand:

  1. If the entry has a DOI, open it through doi.org and compare the title, authors, year and journal with your entry.
  2. If it has no DOI, search the title in Crossref or your library catalogue.
  3. Mark any entry you cannot find, then fix it or remove it.

The full method, with a worked example, is in how to verify references.

4. Statistics checks: p-values against their test statistics

A statistics checker recalculates each p-value you report from its test statistic and degrees of freedom, and flags the ones that do not fit.

Inconsistent p-values have been found often in published psychology papers. Nuijten and colleagues used the R package statcheck on papers from eight major psychology journals published from 1985 to 2013, reported in Behavior Research Methods. Half of the papers that used significance testing contained at least one p-value inconsistent with its test statistic and degrees of freedom. One in eight contained a grossly inconsistent p-value that may have affected the statistical conclusion. They suggest using statcheck to flag possible inconsistencies in your own manuscript.

Check it by hand: for each reported test, recompute the p-value from the test statistic and degrees of freedom in your statistics software, and compare it with the figure in your text.

5. Structure checks: numbering, contents and length

A structure checker tests the parts of a thesis that are pure arithmetic. It checks that chapter, figure and table numbers run without gaps, that the contents page matches the headings, and that the word count is within your limit.

The rules it checks against are your university’s. Newcastle University’s guidelines for the format of theses, for example, set the order of every part, an abstract of no more than 300 words and word limits by faculty. What each chapter should contain is covered in PhD thesis structure, chapter by chapter.

Check it by hand: search the PDF for “Figure” and “Table”, write each number down in order, and look for gaps. Then compare every contents entry with the heading on the page it names.

6. AI-writing scores: a signal, not proof

Some checkers report how much of a text they label as written by AI. Treat any such figure with care.

Vanderbilt University disabled Turnitin’s AI detection tool in August 2023. It noted that Turnitin claimed a 1% false positive rate, and estimated that, on its 2022 submission numbers, around 750 student papers could have been incorrectly labelled as partly written by AI. It also said AI detectors have been found more likely to label writing by non-native English speakers as AI-written.

If you are worried about being wrongly flagged, keep dated drafts, notes and version history. Our guide on how to prove you wrote your thesis explains what to keep.

Which checks to run before submission, in order

Run the checks that change the text first, and the ones that test the final file last.

  1. References: fix or remove every entry you cannot find, because a removed source changes your citations and text.
  2. Statistics: correct any p-value that does not fit its test statistic.
  3. Language: fix spelling, punctuation and grammar, within your university’s rules.
  4. Similarity: read the matches, not the score, and fix any passage that lacks quotation marks or a citation.
  5. Structure: check numbering, the contents page and the word count on the exported PDF, after every other change.

What Thesisroom does with this

Thesisroom covers three of the six kinds of check above, on an uploaded thesis or reference list. Citation Check looks up every entry in research databases such as Crossref and OpenAlex, and checks whether the DOI resolves and whether the paper has been retracted. The Statistics, References and Structure checks recalculate reported p-values, match in-text citations against the bibliography in both directions, and report numbering gaps and the word count against your limit. Grammar, similarity and AI-writing scores are not among its checks.

Frequently asked questions

What does a thesis checker check? It depends on the checker. The six kinds of check in this guide test language, text similarity, whether references exist, reported statistics, structure and numbering, or how much of a text looks AI-written. Choose by the problem you want to rule out.
Is a similarity score the same as a plagiarism score? No. De Montfort University’s guide for examiners says the score is the overall percentage of matching text, and that no threshold figure for it reliably indicates plagiarism. Examiners are told to look at how the highlighting is distributed.
Can I use Grammarly on my thesis? Check your own university’s policy. Birmingham, for example, allows spelling, punctuation and basic grammar fixes, but not rewriting arguments, restructuring content or using Grammarly’s advanced features for assessed work.
Are AI detectors reliable for a thesis? Not reliable enough to act as proof. Vanderbilt disabled Turnitin’s AI detector in 2023, citing false positives and evidence that detectors more often flag writing by non-native English speakers. Keep dated drafts as evidence of your process.
Which thesis check should I run first? Run reference and statistics checks first, because fixing them changes your text. Run language and similarity checks next, and check structure last, on the exported PDF you will submit.

Sources

How this guide was made: Thesisroom drafted this guide with the help of AI writing tools. Every fact, figure and source in this post was checked against its original source before publishing.