A dissertation runs 20,000 to 100,000 words, takes two to five years to draft, and lands at a viva where examiners cross-question every chapter. TextSight is the pre-Turnitin and pre-iThenticate calibration that sits between your draft and your supervisor: sentence-level highlights per chapter, methodology safe in Light mode, discussion restored to your habitual hedging, 90-day audit history for the handoff. Not a detector workaround. A way to defend the prose you submitted as your own.
A dissertation defended at viva on an AI rewriter rewrite you cannot explain is worse than no AI rewriter at all. TextSight is built around a different goal: pre-Turnitin chapter-by-chapter calibration so you can hand a clean, defendable thesis to your supervisor and examiners.
If you used ChatGPT for outlining, literature brainstorming, or polishing a draft, that is the realistic 2026 default and most institutions accept it with disclosure. What none accept is undisclosed AI-generated substantive prose surfacing at viva. The integrity question is not whether you used the tool. It is whether the argument is yours and whether you can defend the writing orally.
The TextSight workflow is built around that question. You paste each chapter section, the classifier flags sentences that read template, you rewrite them in your own register, and the AI rewriter is there for the connective prose only. Methodology gets a hand-rewrite. Discussion gets your habitual hedges back. The 90-day Pro history doubles as evidence of pre-submission screening when a supervisor or examiner asks how you prepared the thesis.
A dissertation is not one document with one AI score. Each chapter has its own register, paraphrase density, and false-positive baseline. Read the score in context of the chapter type, not as a single number across the whole thesis.
Lit review chapters are dense paraphrase of other scholarship in your own words. The form itself overlaps with patterns AI classifiers learned to flag. Expect raw scores in the 55 to 70 band even when entirely your own writing. The right move is to read sentence highlights and re-group citations by argument rather than chase a higher total. Three to five citations per claim, not one per sentence.
Methodology reads identically across thousands of theses in a discipline because it describes a standard procedure. That uniformity is a strength scholarly and a weakness against classifiers. Scores land in the 60 to 75 band. The fix is not auto-rewrite. It is a hand-rewrite that surfaces the decision behind each procedural step: which alternative you considered, why you rejected it, what your supervisor pushed back on.
Short paragraphs that wrap tables and figures use stock phrasings by convention ("Table 3 reports the descriptive statistics"). They trigger easily and that is fine. Scan only the framing prose, leave the tables themselves alone, and watch for one section drifting noticeably from another.
Discussion is where genuine synthesis happens and where ChatGPT drafting is most likely to have crept in if you used it. Healthy scores run 70 plus. If discussion lands lower than your literature review, treat that as a real signal worth investigating, not a calibration quirk. Restore your habitual hedges from earlier drafts.
Both swing widely depending on whether you wrote scaffolding paragraphs or argument-led prose. The classic AI tell is the opening recap of the thesis question at the head of every chapter, and the "this chapter has demonstrated" closer at the foot of each. Drop both. Lead with the specific argument, end on the unresolved tension that motivates the next chapter.
The thesis authenticity phase usually runs one to three months. Pro at $19.99 standard, $14.99 a month on yearly, and $13.99 a month with verified institutional email. Yearly is the default. Full details on the pricing page.
Billed $89.88/year — Save $30
Billed $179.88/year — Save $60
Billed $359.88/year — Save $120
.edu, .ac.uk, .ac.in, .edu.au, .edu.sg, .edu.ph, .edu.vn, and .ac.ke emails get Pro at $13.99/mo. Applied automatically at signup. View full pricing →
TextSight is the pre-submission scan that runs while you edit, so you have sentence-level context before your institution's Turnitin or iThenticate check decides anything. Both still dominate university integrity workflows worldwide. TextSight sits between your draft and that check, not in place of it.
A dissertation chapter runs 6,000 to 15,000 words. Pro accepts 10,000 characters per scan, roughly 1,600 words or six pages. Split at section breaks. A 12,000-word chapter is eight blocks. Free tier covers 5,000 characters and is fine for the abstract and chapter intro only.
Chapter intros carry the highest AI risk because ChatGPT defaults to scaffolding paragraphs that restate the thesis. Target Authenticity above 75. If the intro recaps the thesis question, compress it into one sentence and lead with the argument the chapter actually lands.
Light for methodology, results, and any chapter heavy on technical terms, statistics, or instrument names. Balanced for literature review, introduction, and discussion. Maximum is risky; reserve it for paragraphs you are wholesale rewriting and never use it on procedural content.
For each flagged methodology paragraph, write the decision behind the procedure in your own words. Name the alternative you considered. State why you rejected it. This is what protects you at viva: not an AI rewriter rewrite you memorised, but the reasoning you already had.
Pull two or three early-draft chapters from before AI use and read your habitual hedges. "It appears that," "may indicate," "broadly consistent with." Restore that register in discussion. Vary paragraph openings. Add the one comparison to prior work that surprised you.
Re-scan the chapter block by block. Confirm each section sits above 70 Authenticity Score. Pro keeps 90 days of scan history with timestamps, character counts, and per-block scores: a defensible record of pre-submission screening for supervisor or examiner.
Supervisors and committee members sign off on the work and want you to defend successfully. Bring AI detection into the relationship proactively rather than reactively, so the first time a flag comes up you have already done the audit.
Tell your supervisor that you plan to run a pre-submission AI scan on each chapter, that you are aware lit review and methodology produce real false positives, and ask whether your institution's Turnitin or iThenticate configuration is known to be lenient or strict on academic register. Most supervisors have already seen the false-positive problem with previous candidates and will appreciate the heads-up.
When you send a chapter for feedback, attach the TextSight sentence-level highlight summary as a supporting document. Your supervisor does not need to act on it; they need to know you have done the audit. If the institutional check later flags the chapter, your advisor already has context for the conversation.
Before formal submission, run a full re-scan across every chapter and export the per-chapter PDFs. The 90-day Pro history holds the timestamps and scores. Hand the package to your committee alongside the manuscript so the integrity audit lands with the thesis itself, not as a reaction after a flag.
If a committee member raises an AI-detection result, treat it as a substantive question. Walk through the specific flagged paragraph, explain the disciplinary register or legitimate paraphrase that triggered the flag, and reference your earlier TextSight scan. Sentence-level evidence beats a one-line institutional summary every time.
A ChatGPT-drafted methodology paragraph from a sociology dissertation, followed by a hand-rewrite that adds the decisions an examiner will ask about. Sample sizes, instrument names, and ethics references are preserved exactly.
"This study employed semi-structured interviews to gather qualitative data from participants. A purposive sampling strategy was utilised to recruit 24 participants who met the inclusion criteria. Interviews were audio-recorded and transcribed verbatim. Thematic analysis was conducted using Braun and Clarke's (2006) six-phase framework. Ethics approval was obtained from the institutional review board (Ref: 2024-HRE-0418). Data saturation was reached after 22 interviews."
"I chose semi-structured interviews over focus groups because the topic (workplace disclosure) carries enough stigma that participants were unlikely to speak openly in front of peers. Purposive recruitment landed 24 participants through two professional networks; I stopped at 22 once the theme map stabilised. Transcripts were coded against Braun and Clarke's (2006) framework, though I broke from their order and ran phase three before phase two because early codes clustered unevenly. Ethics approval is logged at Ref 2024-HRE-0418."
What changed: swapped passive voice for first-person reasoning. Named the alternative method (focus groups) and the reason for rejecting it. Disclosed a departure from the cited framework, which is the kind of detail examiners probe in viva. Sample size, instrument, citation, and ethics reference are unchanged. Score moved 65 points and the paragraph is now defendable orally.
Doctoral candidates split roughly into LaTeX users in STEM disciplines and Word or Google Docs users in humanities and social sciences. TextSight reads the prose regardless of the source, but the paste-flow matters because the classifier reads natural language, not markup.
There is no native Overleaf plugin. The clean workflow is to compile the chapter, then copy body text from the rendered PDF or the Overleaf preview pane and paste that into the scan window. Raw LaTeX commands and math environments distort scores if pasted directly. The compile-then-paste round-trip takes about thirty seconds per block.
Paste the prose and leave the footnote markers in place if you wish. TextSight treats footnote bodies as part of the section if you paste them together; usually it is cleaner to scan body prose first and footnotes separately if you want a focused read.
Select the section, copy, paste into the TextSight scan window. The clipboard transfer strips Docs formatting so the classifier sees clean prose. The 10,000 character cap on Pro and 5,000 on free is the same regardless of source application.
Drag a DOCX, PDF, or TXT into the scan window if you want to preserve formatting context. Pro accepts files up to 10,000 characters per scan and returns the same sentence-level result as paste-in. Useful when chapter formatting matters and you do not want copy-paste to clip footnotes or section breaks.
A doctoral thesis takes weeks to rewrite. Pro retains every scan for 90 days with timestamp, character count, Authenticity Score, and the sentence-level highlight map. PDF export keeps a permanent per-chapter record for the institutional file.
Every scan you run on the same chapter is grouped under the chapter name in your Pro history. You can see how a methodology section moved from a 19 Authenticity Score on the first scan to an 84 after rewrite, with every intermediate scan dated. For a candidate iterating through eight chapters over two months, this is roughly 75 to 150 scans with full traceability.
Each scan exports to PDF with timestamp, score, sentence-level highlights, and the source text snippet. Save one PDF per chapter into your dissertation working folder. If your committee or institutional integrity office later asks how you screened the thesis before submission, the per-chapter PDF set is the answer.
If an examiner asks about a paragraph that the institutional check raised, your TextSight log shows the same paragraph with its sentence-level reasoning, the date you scanned it, and the rewrite you applied. Sentence-level evidence with timestamps is a stronger position than a verbal claim that you screened the manuscript.
More for doctoral writers.
Journal-bound shorter format with the same section-by-section calibration logic.
Research paper workflow →Detection-side workflow for capstone, master's, and PhD chapter drafts.
Thesis detector →The pre-scan workflow that catches Turnitin flags before your supervisor does.
Read the guide →The standalone AI rewriter with Light, Balanced, and Maximum modes.
Open the AI rewriter →Free to try. No card. Institutional Pro at $13.99/mo for verified .edu, .ac.uk, .ac.in, and .edu.au emails.