Pre-scan your lab reports, design specs, technical memos, ethics statements, and capstone reports before Turnitin or your advisor sees them. Sentence-level highlights show exactly which lines read AI, with perplexity and burstiness signals tuned for templated methods sections, recurring units and symbols, and the LaTeX-heavy register engineering courses are taught to write in. FERPA-aware, no training on student work. Free to try. No card.
For engineering undergrads and grad students writing 8 to 14 deliverables a semester across mechanical, electrical, computer, civil, chemical, and biomedical courses, plus a senior capstone and a thesis chapter or two. The realistic 2026 default is draft fast, scan before submission, fix the specific sentences that read AI.
Engineering students carry a writing load that mixes templated lab and design genres with prose-heavy ethics and rationale sections. Detectors over-flag the templated register that ABET-accredited programs explicitly teach, which means false positives hit hardest in the methods paragraphs you were trained to produce. Pre-scanning is the cheapest insurance against a wrongful integrity review on a lab report you actually wrote yourself.
Four to ten pages of structured methods, results, and discussion. Free tier covers a single lab scan up to 5,000 characters. Pro at $19.99 a month, or $14.99 a month on yearly, unlocks 10,000 character pastes for longer ME or ChemE reports and unlimited scans for the weeks you have two labs due back-to-back.
Senior capstones at MIT, Stanford, Caltech, CMU, and Georgia Tech often involve real client or sponsor deliverables: 30 to 60 page design reports, feasibility studies, or system architecture documents. The 90-day Pro history matters when a capstone advisor asks about a draft section you submitted three weeks ago. PDF export keeps a defensible record of what you scanned and when.
EE, CompE, and ME courses run design specs and technical documentation across two to four students. Templated section structure makes mixed-author drafts read uneven to any detector. Scanning section-by-section instead of the merged document tells you which teammate's paragraphs are pulling the score down without surfacing it as a public accusation.
MIT, Stanford, Caltech, Carnegie Mellon, Georgia Tech, UIUC, Michigan, UT Austin, Purdue, Cornell, and UC Berkeley all run Turnitin's AI check across core engineering coursework. IIT Bombay, IIT Madras, IIT Delhi, IIT Kanpur, IIT Kharagpur, and the IISc programs follow the same pattern. ETH Zurich, EPFL, TU Delft, Imperial College London, Cambridge, and Oxford engineering science round out the global list. Most of these programs have a pre-Turnitin culture among students: draft normally, scan with TextSight before submission, edit the flagged sentences, then submit through Canvas, Blackboard, Brightspace, Gradescope, or the school's preferred LMS. The .edu discount applies on Pro for verified MIT, Stanford, CMU, Michigan, UT, Purdue, Cornell, Berkeley, IIT, ETH, EPFL, TU Delft, and Imperial emails.
Six genres cover most of the writing an engineering student submits across an undergrad and into grad school. Each has its own false-positive profile, and TextSight is calibrated for all six.
The most common engineering genre and the one most often over-flagged. The methods section rewards templated phrasing, fixed reagent concentrations, and prescribed apparatus vocabulary. Aim for an Authenticity Score above 70 on the discussion section, where original analysis lives. Scattered yellow flags inside a tight methods paragraph usually reflect the genre, not AI use.
Trade-off analysis and justification of design choices for courses like mechanical design, electrical systems, software architecture, and structural engineering. The narrative-heavy format trips detectors less than methods sections but more than ethics statements. Scan the full rationale as one document, because flow between alternatives matters here.
Short engineering memos summarising a test result, a design review, or a status update. Numeric content sits outside the classifier; only your narrative explanation gets scored. Common false positives come from textbook-style phrasing around tolerances, stress concentrations, and signal-to-noise ratios. Rewrite definitions in your own voice.
Mechanical, electrical, civil, and chemical engineering capstones ask for 30 to 60 page reports with problem statement, design alternatives, analysis, prototype, and ethics. Scan the problem statement and design rationale sections most carefully, because those are the prose-heavy parts where AI residue accumulates and where reviewers focus first.
ABET-required ethics paragraphs are the single most over-flagged section in a capstone or design report. The structure follows a prescribed format around stakeholder impact, environmental responsibility, and professional conduct, and that structure overlaps almost completely with how ChatGPT writes ethics paragraphs. Add one specific anecdote about your design choices to break the templated rhythm.
The senior or grad deliverable. Often a thesis chapter or a journal-style methodology for an IEEE or ACM workshop paper. Multiple draft cycles, advisor reviews, and a final defense. The 90-day history on Pro is built for this rhythm: scan after each revision, watch the score climb, keep the PDF receipts.
Verified institutional emails get the .edu discount on Pro automatically at signup. Pro is $19.99 a month standard, $14.99 a month on yearly, and $13.99 a month with .edu verification. 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, and .edu.au emails get Pro at $13.99/mo. The discount applies automatically at signup. View full pricing →
Templated methods, standardised formulas, and repetitive units and symbols all push perplexity down. The classifier reads that as AI-shaped even when you wrote every word.
Every engineering course teaches the same methods structure: apparatus, procedure, materials, calibration, data collection. The vocabulary is prescribed and the sentence rhythm is uniform on purpose, so a second researcher can reproduce your experiment. That uniformity reads as low burstiness to any classifier. Scattered yellows inside a methods paragraph are the genre, not AI.
The narrative around standardised formulas like Bernoulli, Navier-Stokes, Ohm, Kirchhoff, and the ideal gas law tends to repeat textbook phrasing because the equations themselves are not yours to invent. Rewriting the explanation in your own voice is the cheapest fix. Reuse the equation, replace the sentence around it.
kPa, N/m squared, A/m, dB/Hz, and the SI prefix chain all push token frequency up and perplexity down. A paragraph dense with units reads predictable to the classifier even when the underlying analysis is original. The fix is not to remove units, just to vary the prose around them.
Papers written in LaTeX or Overleaf often have a high equation-to-prose ratio. Equations are stripped at scan time and never scored, but the side effect is a lower effective word count. Below 200 words of pure prose between equations, the score gets less reliable. Scan the full prose body as one document, not section by section.
Most engineering undergrad papers do not go straight to IEEE or ACM venues. The realistic pipeline is undergrad conferences, capstone showcases, and design expos that started adding AI-content review in 2025.
ASEE Annual Conference and Exposition runs the largest undergrad engineering education track in North America. SEFI European Society for Engineering Education runs the equivalent in Europe. Your school's Capstone Day, Senior Design Expo, or end-of-semester Engineering Showcase is the in-person venue. Each of these added a basic AI-content screen at submission in 2025, with review at the round between regional and national for the larger venues.
Reviewers started getting submissions that were obviously templated. Some teams were running the project prompt through ChatGPT, lightly editing the methodology and ethics sections, and submitting. The review boards responded by adding a basic AI screen, particularly on the ethics statement and design rationale where templated prose stands out most.
For a multi-round venue, scan your team's paper and poster abstract before each round. Aim for an Authenticity Score above 75 on the design rationale and above 70 on the ethics statement. Below 65 on either means rewrite before submission, particularly the executive summary and recommendation sections that reviewers read first.
If the conference committee asks about AI use, the Pro tier exports a PDF showing the input text, the Authenticity Score, the sentence-level flags, the timestamp, and the classifier version. That is the format a conference appeals process actually wants to see.
Engineering papers written in LaTeX have a lower effective word count once equations are stripped. Here is how to scan them so the score is still trustworthy.
Equations, inline math, and display math are stripped from the prose before classification. The classifier never sees the math. Only the surrounding sentences get scored. This is the right behaviour for engineering text, because the equations themselves are not yours to vary.
A 5,000 character paste with heavy equations might only contain 400 words of actual prose. Below 200 prose words, the classifier has less signal to work with and the score becomes noisier. Scan the full prose body as one document so the classifier has the full context, not section by section.
If you paste from the rendered PDF, ligatures and math escape characters can confuse the parser. Paste the source body from Overleaf or your local TeX editor instead. The classifier handles raw LaTeX commands cleanly, and the prose detection works the same way.
Grad students writing equation-heavy thesis chapters should scan chapter by chapter, not the full thesis. Each chapter is enough prose for a stable score, and the 90-day Pro history lets you compare chapter scores across revisions.
A single percentage is not a fix path. The TextSight result panel shows which sentences reacted and why, so you can edit the specific lines instead of rewriting a whole lab report.
Every sentence in your lab report is colour-coded by its own AI-likeness score. Red sentences clustered in one paragraph are a stronger signal than scattered yellows. Scattered yellows in an otherwise structured methods section often just mean you were taught to write in the templated register. You read the pattern, not just the headline number.
Above sentence highlights, paragraph-level cards show which sections of your report are pulling the score down. For a 30-page capstone draft, this is faster than reading every red sentence. Identify the two or three paragraphs that need rewriting, then drill into sentences inside those sections only.
Perplexity is how predictable your word choices are to a language model. Low perplexity reads AI-like. The score is shown per-sentence on Pro, which is the diagnostic context you need to decide whether a flag is real AI residue or just an unusually well-rehearsed methods description.
Burstiness is how much your sentence length and structure vary across the document. ChatGPT defaults to uniform medium-length sentences. Real student writing has bursty rhythm: one short sentence, one long, one fragment. Low burstiness across an entire lab report is the classic AI fingerprint, and it shows up most often in methods and ethics paragraphs.
Student submissions are protected by FERPA in the US, by GDPR in the EU and the UK, and by local equivalents elsewhere. TextSight is designed to honour those rules out of the box, not as a paid setting you have to find.
Lab reports, capstone drafts, design specs, and technical memos you submit for scanning are never used to train the classifier or any other model. This is a contract clause, not a configuration toggle. It applies on the free tier the same way it applies on Pro and Business.
The free tier needs no email, no account, no identity. For students worried about an unpublished capstone or thesis chapter leaking, this matters. You can scan a sensitive sponsor-deliverable draft without TextSight ever knowing who you are.
Scan history is private to your account. We do not share scan data with universities, professors, capstone advisors, ABET, IEEE, Turnitin, or any third party. Your lab reports and capstone drafts are not part of any institutional record, and your professor cannot pull them.
Any scan can be deleted from your history. On Pro you can delete individual records. Data retention is bound to your settings, and a standard DPA is available on Business and Enterprise tiers for capstone teams and engineering project clubs.
The full student landing page with the false-positive defence and the academic tone preset.
For students →The broader undergrad workflow across all majors, with .edu Pro pricing.
For college students →The pre-scan workflow that catches Turnitin flags before your professor does.
Read the guide →Free, Starter, Pro, Business. Yearly billing saves 25%. .edu discount on Pro at signup.
See pricing →Free to try. No card. .edu Pro at $13.99/mo for verified institutional emails.