In this guide
- The short answer
- The four UGC levels, and what each one costs you
- The four exclusions that change everything
- Why your raw score is not your real score
- Why your department's limit may be stricter
- Where AI detection fits in — and where it does not
- How to actually bring the number down
- Frequently asked questions
The short answer
Similarity up to 10% is treated as acceptable for a PhD thesis or dissertation under the University Grants Commission's academic integrity regulations. At that level there is no penalty and no revision required. Above 10%, penalties begin — and they escalate quickly.
But that sentence hides three things that matter far more than the number itself:
- The 10% applies after official exclusions are correctly applied — not to the raw figure at the top of your report.
- Your own university is free to enforce a stricter internal limit, and many do.
- Similarity and AI detection are separate measurements with separate rules. Passing one tells you nothing about the other.
The single most common mistake: a scholar sees 24% on a raw report, panics, and spends three days rewriting perfectly good paragraphs. Then a properly configured check comes back at 7% — because most of that 24% was the bibliography, the university's own cover-page template, and correctly quoted material. The rewriting was never the problem. The settings were.
The four UGC levels, and what each one costs you
The UGC's Promotion of Academic Integrity and Prevention of Plagiarism in Higher Educational Institutions Regulations, notified in 2018, define four bands of similarity. These regulations remain the governing framework for thesis submission across Indian higher education institutions.
| Level | Similarity | Consequence for a student |
|---|---|---|
| Level 0 | Up to 10% | Minor similarity. No penalty. Thesis proceeds. |
| Level 1 | Above 10% up to 40% | No marks or credits. Revised script required within a limited window — commonly six months. |
| Level 2 | Above 40% up to 60% | No marks or credits. Revised script required after a longer waiting period, typically one year. |
| Level 3 | Above 60% | Registration for the programme is liable to be cancelled. |
Two things are worth sitting with here. First, the jump from Level 0 to Level 1 is not a warning — it is a lost submission cycle. Second, the gap between 10% and 40% is enormous, which tells you the regulations were written to catch serious copying, not to punish a scholar whose literature review shares standard phrasing with the field.
Institutions also typically have an Academic Integrity Panel that reviews cases rather than applying penalties mechanically. If your similarity is genuinely explainable, that explanation has a place to go — but it is a great deal easier to make the case before submission than after.
The four exclusions that change everything
This is the part almost nobody is told, and it is worth more to you than any rewriting advice on this page. The regulations explicitly exclude certain categories of text from the similarity calculation:
- Quoted work reproduced with permission or proper attribution. If it is in quotation marks and correctly cited, it should not be counting against you.
- References, bibliography, table of contents, preface and acknowledgements. In a thesis with 200 references, this alone can be several percentage points — sometimes double digits.
- Generic terms, laws, standard symbols and standard equations. You cannot paraphrase Ohm's law, and you should not be penalised for stating it correctly.
- Small matches below a set word threshold. Most institutional configurations exclude very short strings, because a four-word phrase matching a random web page is noise, not plagiarism.
The catch: these exclusions are settings. They do not apply themselves. A report generated with default settings and no exclusions will show you a number that your university's own configured check would never produce. That is why two reports on the identical document can differ by fifteen points or more.
Practical step: before you order any check, ask your department two questions — which tool they use, and which exclusions their configuration applies. If they tell you "bibliography and quotes excluded, small matches under 14 words excluded", you now know exactly how to have your own check run so the number you see matches the number they will see.
Why your raw score is not your real score
Here is a realistic breakdown of a 24% raw similarity index on an engineering thesis:
| Source of the match | Contribution | Genuine problem? |
|---|---|---|
| Reference list and bibliography | 9% | No — excludable |
| University cover page, declaration, certificate format | 4% | No — mandated template |
| Correctly quoted and cited passages | 3% | No — excludable |
| Standard definitions and equations in the field | 3% | No — unavoidable |
| Paraphrased literature review, lightly reworded | 4% | Yes — fix this |
| Methodology text reused from your own earlier paper | 1% | Yes — self-plagiarism, cite it |
Nineteen of those twenty-four points are structural. Five are real. And those five are fixable in an afternoon — but only if you can see the source-by-source breakdown that tells you which is which. A report that gives you a percentage and nothing else is close to useless for this purpose.
Self-plagiarism is real, and scholars get caught by it constantly
Reusing your own methodology section from a published paper counts as similarity. It will match, it will be flagged, and "but I wrote it" is not automatically a defence. The fix is straightforward — cite your earlier work like any other source, or rewrite the section — but you have to know it is there.
Why your department's limit may be stricter
The UGC figure is a floor for penalties, not a ceiling on institutional standards. Universities routinely set tighter internal norms, and individual guides set tighter ones still. In practice we regularly see:
- PhD theses: commonly capped at 10%, sometimes 5% in competitive departments
- M.Tech / M.Phil dissertations: usually 10–15%
- PG and UG project reports: often 15–25%, with wide variation between colleges
- Journal submissions: publisher-dependent, frequently under 15% and sometimes under 10%
Ask. Get it in writing if you can — an email from your guide or a line in the departmental handbook. "I thought the limit was 10%" is a poor position to be in when the handbook says 8%.
Where AI detection fits in — and where it does not
The 2018 regulations predate generative AI entirely. They set no numerical threshold for AI-generated content, because in 2018 there was nothing to threshold. That gap has been filled — inconsistently — by individual universities writing their own rules.
What this means for you is practical rather than philosophical: your AI score is governed by your institution, not by the UGC, and the only reliable way to find out the limit is to ask your department. Some have set explicit caps. Some review flagged work case by case. Some have no policy at all yet.
What is consistent everywhere is that similarity and AI are measured separately and reported separately. A thesis can return 4% similarity and 60% AI, or 30% similarity and 0% AI. Clearing one does not clear the other, and a report that only gives you one number is only telling you half the story.
It is also worth knowing that AI detectors flag patterns, not dishonesty. Uniform sentence length, formulaic transitions and heavily standardised terminology all raise the score — which is why non-native English writers and scholars in template-heavy disciplines get flagged more often regardless of who wrote the words. We wrote a full guide on what to do when your own writing gets flagged as AI.
How to actually bring the number down
In rough order of how much they help per hour spent:
1. Fix the settings before you fix the text
Confirm exclusions are applied to bibliography, quotations and small matches. This is free and frequently accounts for more than half the raw score.
2. Read the source list, not the percentage
Work down the ranked list of sources. The top three or four matches almost always account for most of the fixable similarity. Everything below 1% is usually noise.
3. Rewrite from the idea, not from the sentence
Swapping synonyms leaves the sentence structure intact, and structure is exactly what matching algorithms detect. Close the source, write down what the point actually was in one line, then write your own sentence from that line. It is slower and it works.
4. Turn long quotes into analysis
A 60-word block quote contributes 60 words of similarity and demonstrates nothing about your thinking. Quote the eight words that matter, then spend three sentences explaining why they matter. Your examiner will prefer it too.
5. Cite your own earlier work
If you reused text from your own published paper, cite it properly. This converts a self-plagiarism flag into an excludable, correctly attributed quotation.
6. Do not touch the template
Your cover page, declaration and certificate are mandated formats. Rewriting them to reduce similarity will get your submission rejected for a different reason. Exclude them; do not edit them.
The order matters more than the effort. Almost every scholar we work with starts at step 3 and never does steps 1 and 2. Doing them in order routinely turns a three-day panic into a two-hour job.
Frequently asked questions
How much plagiarism is allowed in a PhD thesis in India?
Up to 10% similarity is Level 0 under the UGC regulations and carries no penalty. Above 10% and up to 40% is Level 1 and requires a revised submission. Confirm your own university's internal limit, which may be lower.
Does a 20% similarity score mean I have plagiarised?
No. Similarity measures text overlap, not misconduct. Correctly cited quotations, your reference list, standard definitions and institutional templates all generate matches. What matters is what the matched text is — which is why the source-by-source breakdown is the part of the report you should actually be reading.
Can I check my thesis myself before submitting?
Yes, and you should. Checking your own work is preparation, not evasion — your university will run the identical check on you eventually. The only thing to be careful about is ensuring the check does not store your paper in a student repository, because your own official check would then match your thesis against your own earlier copy and return a falsely enormous score.
Is 0% similarity possible?
Effectively no, and it would be a warning sign if it happened. Any real thesis cites sources, uses standard terminology and follows an institutional template. A 0% result usually means the check was misconfigured, not that the document is unusually original.
What happens if I cross the limit after submission?
It goes to your institution's academic integrity panel, which reviews the case rather than applying penalties automatically. The outcome depends heavily on what the matched text actually is. This is precisely why finding out before submission is worth the small cost of a check.
Find out your real number, not your raw one.
Send us your thesis and we will run it with exclusions configured properly — then tell you plainly which matches are harmless and which ones you need to fix.
Note: This guide summarises the UGC's 2018 academic integrity regulations for general information and is not legal or institutional advice. Regulations and institutional policies can be amended, and individual universities may enforce stricter limits. Always confirm the current requirements with your own department before submission.