How Thesis-It uses AI

Thesis-It is a writing workspace with AI steps inside it, not a machine that writes a thesis. Every AI step works from your structure, your facts and your sources, on one section at a time, under your direction. The draft it produces is a proposal: nothing enters your thesis until you read it and accept it, and you can restore the version before it at any time. Your supervisor and your examiners remain the judges of the work, and you remain its author.

The instructions prohibit invented facts, numbers and citations. Missing facts should become marked gaps for you to fill, and inferences should be marked for you to confirm. Models can still make mistakes, and automated checks do not prove that a claim is true. You must check the draft against your research and sources. Every citation it writes must resolve to a source in your library or to a verified DOI, or it is flagged. It never judges or rewrites your university's format rules: those come from the structure you adopted, and the checks are plain rules the app runs the same way every time. Humanise writing is an optional rewrite with protected facts and references, an independent meaning check and your review before acceptance. Its changes remain AI assisted. No AI detection score, human authorship claim or guarantee of watermark removal is inferred from a rewrite. A reply it drafts to your ethics committee cannot claim a change your revised form does not make: such a sentence stays marked for you to confirm.

With the included AI, the app sends the text you chose, together with your facts and the section's brief, to the Google Gemini API, OpenAI API and Anthropic API through our server and shows you the answer as a draft to accept or reject. A subscription includes a monthly allowance on leading AI models, typically enough to draft one long chapter a month with every check. The app shows how much of the month's allowance is left, and each AI step shows its share before it runs. How far the allowance goes depends on how much text a step reads and writes, not on which AI service answers it. Included AI can pause for a while when the month's service budget runs short, and the app says so. With your own AI, you copy a prompt the app prepares, paste it into a chat service you already use, and paste the reply back. The app then checks that reply exactly as it checks its own drafts. You can also use your own Gemini API key, which stays on your device.

When you run an included-AI step, its text (the section, your recorded facts and relevant structure and source material) is sent to the Google Gemini API, OpenAI API and Anthropic API. When a section is drafted from your chapter plan and it reports or discusses your results (your results, discussion, implications, recommendations, limitations or conclusion), that material includes the results you kept from your analyses, and a section that discusses them also includes the accepted text of your results sections. The draft is asked to cite each kept result by its marker instead of retyping its number. When you ask Thesis-It AI to suggest research questions, objectives or hypotheses, it is sent your research idea, your problem statement and what you recorded about your research. When it reads your chapters for your research questions, objectives and hypotheses, it is sent up to 10,000 words of them, with the research questions, objectives and hypotheses you recorded, and whether you are writing a proposal. When it suggests where your sections belong among your studies and papers, it is sent passages of your chapters, your research questions, the contributions you recorded, the studies you recorded and the sections you already placed, and the titles of the results you kept, never a data file or its rows. When it writes your question for Google Scholar Labs, it is sent what you wrote about your research and your three answers about how you will study it, and nothing else you recorded. When it finds the research gaps in a search you ran, it is sent the topic of your search, your thesis language and the results pages you pasted or the papers you ticked, with the titles, authors and notes of other authors' papers they show, and nothing else you recorded. When you search in Thesis-It, or tick a gap, Thesis-It sends your topic, or the papers' titles and DOIs, to Crossref, and asks OpenAlex for their abstracts, which are not AI services. When it proposes your title, questions and methodology, it is sent what you wrote about your research and your answers about how you will study it, the gaps you kept with the words they rest on, the titles and the first part of the abstracts of the verified papers behind them, your answers to its earlier questions, the title, research questions, objectives, hypotheses, framework and methodology you already have, your writing stage and your thesis language. When it proposes hypotheses and a framework for the questions you kept, it is sent the title, research questions, objectives, methodology, hypotheses and framework you have, what you wrote about your research and your answers about how you will study it, the gaps you kept with the words they rest on, the titles and the first part of the abstracts of the verified papers behind them, and your thesis language. When it plans your analyses, it is sent your research questions and hypotheses (your objectives, where you recorded neither), the study design you recorded and the words of any extra analysis you add, never a data file, its rows or the names of its variables. It plans new questions by itself only once you turn that on in Analysis, and until then only when you press a button for it. We store draft packages, candidate replies, check results, the results pages you paste (for 30 days, so you can come back to them), the questions it writes for Google Scholar Labs and your analysis plan (the analysis chosen for each question or extra analysis, and the words it rests on) so you can resume and review them across devices. We also retain usage counts and the answering provider and model where reported. Your template, guideline or model thesis file is never sent anywhere: it is read in your browser, and only what the app takes from it is stored. When you choose to read one with your own AI, you attach the file in your own AI chat, under your own account, and Thesis-It sends nothing: the app checks the reply you paste against the text it read in your browser. For a template or a guideline that is its extracted text. For each model thesis, up to three, it is also its writing style, measured as numbers such as how long its sentences and paragraphs run, and a keyed fingerprint of its wording: a one-way code for each run of eight words, computed with a key our server derives for your thesis and never stores, so someone who reads our database cannot read its words back from the codes. When you ask your own AI how your model thesis argues, you attach the model thesis in your own AI chat in the same way, and Thesis-It sends nothing. Its findings describe the model's patterns: what each section does, how its review groups earlier studies, how it justifies its method and how its discussion relates its findings to earlier studies. The app checks each one against the text it read in your browser, refuses any that repeats eight of the model's words or holds a number, a year, a citation or a name you have not recorded about your research, and keeps the rest with your AI's short reasons as suggestions until you accept them or leave them out. A finding you leave out, or leave waiting, is deleted after thirty days. The ones you accept guide your drafts in Thesis-It's own words while you follow that model thesis, and are deleted thirty days after you stop following it. Your AI's reasons are never put in a prompt. The model's findings, data and wording stay the model's, and you are responsible for your rights to the files you add. When you draft, with the included AI or your own, or when you polish a section, the prompt carries your writing rules: how your university's template or guideline asks you to refer to yourself, the conventions your chapters already keep (how they refer to you, and British or American spelling), the rules measured from the model thesis each chapter follows, never its text, the rules you set yourself on the Writing style tab, and Thesis-It's own defaults. The included AI's drafts, the replies you paste from your own AI and Polish's suggestions are checked on our server against the fingerprint of each model thesis you follow, for passages that may repeat one of them. When you draft your ethics form's answers, or your replies to your committee, with the included AI, the form's questions are sent to the Google Gemini API, OpenAI API and Anthropic API with what they are answered from: what you recorded about your research and your study's facts, your ethics checklist, the accepted text of your method sections, the details you gave, the sentences kept from your committee's guidelines, your earlier answers and, for a reply, your committee's comments. The numbers your browser holds, and any identity card, mobile or passport number, are taken out first. The form file itself is never sent. AI requests run when you start a step, when you keep the research gaps you picked (which starts proposing your title, questions and methodology), when you reach hypotheses and framework and no hypothesis answers the questions you kept (which starts proposing them), when you enable automatic checks and open a new draft for review, or, for your analysis plan, when you turn on planning by itself in Analysis. Your provider preference in Settings restricts which disclosed providers may receive included-AI requests. It is a preference under the AI-use and privacy agreement, not a separate consent process.

Accepted AI drafts keep their origin and recorded provider, model or student-named tool. The app keeps the version before an accepted draft. The Finish screen suggests an AI-use statement with tool names from these records where available. Review it, add any other tools and describe your own verification before accepting it into your thesis. Legacy records and work outside the app may not identify a tool. Use it, together with your supervisor's guidance, to declare your use of AI honestly.

Malaysian universities publish guidance on the use of generative AI in postgraduate work, for example UiTM's 2024 guideline on the responsible use of AI (section 6.3), which asks for use that is declared, supervised and does not replace the student's own contribution. Thesis-It is built for that posture. Your own university's, faculty's and supervisor's rules govern your thesis: read them before you use any AI step, and if they forbid a use, do not use it here either. To stop every AI step for a thesis, turn AI off for it under Settings → AI settings.

Ask support@thesisit.app about anything on this page. The Privacy Policy at https://thesisit.app/legal/privacy names the providers involved and where they process data.