1.1 What is Gemini?

Gemini is Google's family of large language models (LLMs), available as a chat assistant at gemini.google.com, as a mobile app, and inside Google Workspace (Docs, Gmail, Drive) for eligible accounts.

An LLM predicts plausible text based on patterns learned from vast amounts of data. It does not look facts up in a database by default. This is why it can write fluently and still be confidently wrong.

1.2 How a large language model generates text

Knowing what happens "under the hood" explains both why Gemini writes so fluently and why it sometimes invents things. Four steps explain it: text becomes tokens, the model makes a next-token prediction one token at a time, fluent output is not necessarily true, and it can only work with what fits in its context window.

Step 1: Text is broken into tokens

An LLM doesn't read words or letters. It reads tokens: common chunks of characters. Frequent words are usually a single token, while rarer words, other languages, numbers and codes are split into several pieces. Each coloured block below is one token.

Step 2: It predicts the next token, one at a time

Given the text so far, the model calculates a probability for every possible next token, picks one (usually a likely one, with some randomness), adds it to the text and repeats. Step through the demo below to watch a sentence being built.

Nothing in this process looks anything up or checks whether the sentence is true. The model chooses what is statistically likely to come next, based on patterns in its training data. Most of the time, likely text is also correct text. When it isn't, the output is still fluent and confident.

Step 3: Fluent is not the same as true

Here is how a fabricated citation happens. Asked for evidence, the model produces something with the shape of a reference: an author name common in the field, a recent year, a percentage, a journal-like title and a DOI-like string. Each token is plausible on its own; the combination may not exist.

Step 4: It only sees what fits in its context window

Everything the model can "see" at once (your instructions, uploaded files and the conversation so far) has to fit in its context window, which is measured in tokens. Anything outside it, including earlier chats, is not available unless you add it again.

Shown at roughly Free-plan scale: one 12,000-token paper already fills almost half the window.

How big is the window on each plan?

The window size is one of the biggest differences between plans (see the table in 1.4). The bars are drawn to scale:

• Free ~32,000 tokens · ~50 pages About one or two journal articles plus your conversation
• Google AI Plus ~128,000 tokens · ~200 pages About 6–8 articles, or one thesis chapter with its key sources
• Google AI Pro ~1,000,000 tokens · ~1,500 pages A full thesis draft plus dozens of papers in one chat
• Workspace for Education Higher than Free · ~1,500 pages with add-on Depends on your university's edition and whether it has the AI Pro for Education add-on

Indicative mid-2026 figures that change often. "Pages" assume English text; Bahasa Melayu and other languages use more tokens per word (Step 1), so fewer pages fit.

1.3 Getting access

1. Go to gemini.google.com and sign in with your Google account. If your university provides Google Workspace for Education (often your student or staff email), sign in with that account instead, because it comes with stronger data protection.
2. Review Settings → Activity to understand how your conversations are stored and whether they may be used to improve Google's services.
3. Start a new chat for each distinct task. Separate chats keep context clean and outputs focused.

1.4 Which Gemini are you using? Free, paid plans and Workspace for Education

The plans share the same interface but differ in capability, Gemini Notebook limits and, most importantly for researchers, how your data is handled. Prices below are for Malaysia.

* Student offers are for college and university students aged 18 and above, valid until 31 December 2026. Google AI Ultra starts from RM429.99/month. ** Gemini for Education is part of every Google Workspace for Education edition, but your university decides whether to switch it on and whether to buy the AI Pro for Education add-on. Users under 18 get a restricted experience. † With the Google AI Pro for Education add-on, purchased by the university. Prices and limits are indicative as of mid-2026 and change often. Check gemini.google/subscriptions and your university's IT department (e.g. its Pusat Teknologi Maklumat) for current details.

What this means for Gemini Notebook

• Free: enough for one thesis chapter or a small review, with up to about 50 papers per notebook.
• AI Plus (free for students until the end of 2026): about 100 sources per notebook, which is a good fit for most master's and PhD literature reviews.
• AI Pro: suits large or systematic reviews (hundreds of sources per notebook), and Notebooks also appear inside the Gemini app, so chat and notebook share the same sources.
• Workspace for Education: the safest place for your own drafts, transcripts or unpublished data, because sources and chats are not read by reviewers or used to train models. Gemini Notebook is included, with 5× higher limits if your university has the AI Pro for Education add-on. Ask your IT department whether it is switched on for your account.

1.5 Interface tour

Features and names change often and vary by plan and region. If something here doesn't match your screen, look for the equivalent option.

1.6 Strengths and limitations

2.1 The RTCFC framework

2.2 Weak vs. strong prompts

2.3 Iterate, don't restart

Treat Gemini as a conversation. After the first response, steer it:

• "Make it more concise."
• "Explain your reasoning for point 3."
• "That's too informal. Use a formal academic register."
• "Give me three alternatives."
• "What are the weaknesses of this argument?"
• "Ask me questions before you answer."

2.4 Prompt templates

Copy, adapt and save these templates. Text in [brackets] is for you to fill in.

Act as a tutor in [discipline]. Explain [concept] to a [level, e.g. master's student] in under 200 words. Then give one concrete example from [context] and one common misconception about it.

I am developing a research argument: [your argument in 2–3 sentences]. Do not agree with me by default. Ask me five probing questions, one at a time, that a sceptical examiner would ask. Wait for my answer before asking the next.

You are my academic writing assistant for a [PhD thesis / journal article] in [field]. Always: use British/American English [choose]; preserve my meaning and technical terms; explain each suggested change; never invent citations, data or quotations; flag any claim that needs a reference.

2.5 Gems: reusable assistants for academic work

A Gem is a saved Gemini assistant with your standing instructions, so you don't have to retype the same long prompt every session. Gems are available on all plans, including Free. Create one under Gems → New Gem (the left sidebar on the web).

Three Gems that work well for academic drafting and documentation (paste into the Instructions box):

You are a thesis structure coach for a [Master's / PhD] student in [field]. Use the attached thesis guidelines as the authority on structure and formatting.
When I share notes or a draft: (1) map it against the required chapter structure, (2) list missing or misplaced elements, (3) ask up to three questions that would strengthen the argument.
Never write full paragraphs for me. Never invent references, data or quotations. Use [British/American] English.

You turn my rough lab notes, field notes or analysis logs into clear, structured documentation.
For every input: organise it under the headings Aim, Materials/Data, Procedure, Analysis, Software and versions, and Deviations. Write in past tense and passive or "we" voice as I specify.
Flag with [MISSING] any detail a reviewer would need to replicate the work (sample sizes, parameters, versions, dates, ethics approval numbers). Do not fill gaps with guesses.

You check manuscripts against the attached author guidelines for [journal]. Report, as a table: requirement | where my draft meets or misses it | suggested fix.
Check word limits, section order, abstract structure, heading levels, reference style and declarations (funding, conflicts of interest, AI use, data availability). Do not change my wording.

3.1 Refining a topic and research questions

My broad research interest is [topic] in the context of [population/setting]. Suggest five focused, researchable questions. For each, state the likely methodology, one feasibility risk, and how it could contribute to the field. Then ask which direction I want to develop.

3.2 What Gemini can and cannot search

When Gemini "searches", it uses the open web through Google Search. It does not log in to your library, so it cannot query subscription databases such as Web of Science, Scopus, PsycINFO, EBSCOhost, ProQuest or IEEE Xplore, and it cannot read paywalled full texts. These platforms sit behind institutional authentication (campus IP, VPN or single sign-on) that only you have. The same applies to Deep Research: its reports draw on publicly reachable web pages.

✔ suitable · △ use with caution · ✘ not suitable

3.3 Generating search terms

Gemini can't run the database search for you, but it can help you design one that you then run yourself.

Help me build a Boolean search string for [Web of Science / Scopus] on [research question]. Identify the key concepts, list synonyms and related terms for each, and combine them with AND/OR and truncation (*), using the correct field tags for that database. Explain each choice so I can adjust it. I will run and refine the search myself.

3.4 Grounding your review with Gemini Notebook

Gemini Notebook (formerly NotebookLM) is Google's source-grounded research tool. Unlike an ordinary chat, a notebook answers only from the sources you add to it and links each statement to the passage it came from. Notebooks are also available inside the Gemini app, and they sync between the two. Availability, plan requirements and source limits vary and change, so check what your account offers.

Recommended workflow

Using only [paper title], summarise: (1) research aim, (2) method and sample, (3) key findings with numbers, (4) stated limitations, (5) relevance to my study on [topic]. Cite the passage for each point. If something is not in the paper, write "not reported".

Across all sources in this notebook, how is [key concept] defined and measured? Group the sources by approach, cite each one, and point out where they disagree or use incompatible definitions.

Based only on these sources, which aspects of [research question] are well covered, which are only mentioned briefly, and which are not addressed at all? Do not draw on outside knowledge. I will search the databases for anything missing.

3.5 Building a synthesis matrix

After summarising several papers you have read, ask Gemini to tabulate them so that patterns and gaps become visible:

Here is my synthesis matrix of [n] studies: [paste]. Identify (a) points of agreement, (b) contradictions, and (c) under-researched contexts, populations or methods. Base your answer only on the matrix and indicate which studies support each point.

3.6 Workflow: from notebook to a short literature review

Once your papers are in a Gemini Notebook, you can turn them into a short, source-grounded literature review: a few paragraphs for an introduction, a proposal or the opening of a thesis chapter. You do the reading, questioning and arguing in the notebook, then bring the notebook into the Gemini app to draft, and finally check everything yourself.

Step 2: questions to ask in the notebook first

Ask these inside Gemini Notebook, one at a time, and save useful answers as notes. They also show you whether your sources are enough before you write anything.

1. What are the main themes across these sources on [topic]? Group the sources under each theme and cite them.
2. Where do the sources agree, and where do they contradict each other? Explain possible reasons (context, method, sample).
3. Which methods, populations and settings are used? Which are missing?
4. How has thinking on [topic] changed over time across these sources?
5. Based only on these sources, what questions remain unanswered?

Step 4: bring the notebook into the Gemini app

In the Gemini app, click + in the prompt box, choose NotebookLM (now Gemini Notebook; the label may differ on your account), then pick your notebook. On Workspace for Education accounts, your administrator must have Gemini Notebook switched on. If you can't see the option, open the notebook in Gemini's Notebooks section, or work directly in Gemini Notebook.

Step 5: the drafting prompt

Using ONLY the sources in the attached notebook "[notebook name]", write a short literature review (about [600–800] words) on [topic] for the [introduction / Chapter 2] of my [thesis / article].

Structure it around MY argument:
- [point 1]
- [point 2]
- [point 3]

Requirements:
1. Synthesise by theme, not one paragraph per paper. For each theme, say where studies agree, where they differ, and why (context, method, sample).
2. Cite every claim in APA 7 author–year format, using only notebook sources. Do not add outside sources, facts or statistics.
3. If a point needs evidence that is not in the notebook, write [NEED SOURCE] instead of filling the gap.
4. End with one paragraph on the gap my study addresses: [your gap in one sentence].
5. Use British English and a formal, cautious academic tone [or: follow my voice profile: paste it].

After the review, list every source you cited and the claim it supports, so I can check them.

For each citation in the review, quote the exact passage from the notebook source that supports it. Flag any sentence where the support is weak, indirect or missing, and any place where two sources were merged into one claim.

3.7 The verification routine

1. Exists? Find the source in a database or via its DOI at doi.org.
2. Accurate? Check the authors, year, title and journal against the publisher's page.
3. Supports the claim? Read the relevant section yourself. Summaries can distort findings.
4. Still valid? Check for retractions or corrections, for example via the Retraction Watch database.

4.1 The "you write, it scaffolds" approach

4.2 Outlining

I'm writing a [type: empirical article / review / thesis chapter] for [journal or audience]. My main argument: [1–2 sentences]. My key evidence: [bullets]. Propose an IMRaD-style outline with section headings, the purpose of each section, and the evidence that belongs there. Flag where my argument currently lacks support.

4.3 Paragraph structure

Strong academic paragraphs often follow a claim → evidence → analysis → link pattern.

Analyse the paragraph below. Identify its topic sentence, the evidence, the analysis and the link to the next idea. Tell me which element is weak or missing and why. Do not rewrite the paragraph.

[paste paragraph]

4.4 Abstracts

Here is my draft abstract for [journal], with a limit of [n] words. Check whether it clearly states background, aim, method, key results (with figures) and implication. Score each element out of 5, explain your scores, and suggest specific improvements. Keep my findings exactly as written.

[paste abstract]

4.5 Introductions: the CARS model

Swales' "Create a Research Space" model is a useful lens for critiquing an introduction:

1. Establish a territory: why the topic matters
2. Establish a niche: the gap, problem or question
3. Occupy the niche: your aims, approach and contribution

Map my introduction below onto Swales' CARS model (territory, niche, occupation). Quote the sentences that serve each move. Identify any move that is weak or missing, and suggest what content (not wording) I should add.

[paste introduction]

4.6 Results and Discussion

Results report what you found; the Discussion explains what it means. Gemini can help with structure, clear reporting and cautious wording. It is also where it does the most damage: it can quietly round, change or invent numbers, and it tends to overclaim. Your data and your analysis output are the only source of truth.

The five moves of a Discussion

Match your claims to your evidence

Using ONLY the table below, write a results paragraph for research question [RQ1: ...].
Report statistics in APA 7 format exactly as given. Do not calculate, round, re-order or add any numbers. Do not interpret the results; that belongs in the Discussion. If something in the table is unclear, ask me instead of guessing.

[paste your table or analysis output]

Compare my Results text with my tables below. List every number, direction (higher/lower) or significance statement in the text that does not exactly match the tables. Do not rewrite anything.

TEXT: [paste]
TABLES: [paste]

My key findings are: [1–3 bullet points]. My study design is [design, sample, setting].
The prior studies I want to compare with are: [list, or attach my Gemini Notebook].
Propose a Discussion outline using five moves: restate the key finding, compare with prior studies (only those I listed), possible explanations, limitations specific to my design, and implications. Do not add any sources. Mark where I need evidence with [ADD SOURCE].

My study is a [design] with [sample]. Review my Discussion below and list every claim that is stronger than this design supports (for example causal language from correlational data, or generalising beyond my sample). For each, suggest a more accurate wording and explain why.

[paste Discussion]

4.7 Canvas: draft and edit side by side

Canvas opens a document editor next to the chat. Instead of copying text back and forth, you and Gemini work on the same document: highlight any passage and ask for a change, adjust length or tone with one click, and export the result to Google Docs. Turn it on by choosing Canvas from the tools menu in the prompt box (or set it as a Gem's default tool).

Quick documentation tasks that suit Canvas

In Canvas, turn my rough supervision meeting notes below into formal minutes with these headings: Attendees, Date, Decisions, Action items (owner, deadline), Next meeting. Keep every decision exactly as written; don't add anything that isn't in my notes. Mark anything unclear as [CHECK].

[paste notes]

In Canvas, draft a README for my research dataset using this information: [project title, collection dates, instruments, file list, variables, licence]. Include sections for Overview, File structure, Variables (as a table), Methods summary, Licence and How to cite. Mark any section where you lack information as [TO COMPLETE].

A four-step Canvas routine

1. Draft: generate the skeleton in Canvas from your notes.
2. Refine: highlight passages and ask for targeted changes, never "rewrite everything".
3. Complete: fill every [MISSING], [CHECK] and [TO COMPLETE] marker yourself.
4. Export: send it to Google Docs, where version history keeps a record of what was AI-drafted and what you changed. This is useful when you write your AI-use disclosure (Module 6).

4.8 Research diagrams with Nano Banana

Nano Banana is the nickname for Gemini's image-generation models. Besides pictures, they can produce diagrams with readable labels: conceptual frameworks, research-design flowcharts and process overviews. You can also turn a photo of a whiteboard sketch into a clean draft. Choose Create image from the tools menu in the prompt box, or simply ask for a diagram.

Where AI-generated diagrams fit

Anatomy of a good diagram prompt

Image models follow layout instructions literally, so spell out every element. Each colour below is one part of the prompt:

Create a clean vertical flowchart of my research design with these stages, in order: [Stage 1: ...], [Stage 2: ...], [Stage 3: ...], [Stage 4: ...]. Show [Phase A] and [Phase B] as two shaded groups. Use rounded boxes, one-way arrows, a white background and a flat academic style with sans-serif labels. Use exactly these labels and add no other text or numbers.

Redraw the attached hand-drawn sketch as a clean, professional diagram. Keep exactly the same boxes, labels and arrows. Do not add, remove or rename anything. If a label is unreadable, write [?] instead of guessing. White background, flat academic style.

Keep everything else identical, but: (1) change the arrow from "[A]" so it points to "[B]", (2) rename "[old label]" to "[new label]", (3) translate all labels into Bahasa Melayu for my presentation version.

Worked example: a draw.io diagram from Canvas

draw.io (also called diagrams.net) is a free diagram editor that runs in your browser and saves to Google Drive. Behind the scenes, every draw.io diagram is saved as text code, so Gemini can write that code for you from a plain description. You paste it into draw.io, and then you can move and edit every box, label and arrow with your mouse.

You don't need to know any code. Describe your diagram in everyday words, the way you would explain it to a colleague. Copy the prompt below, open Canvas in Gemini, and replace the example boxes and arrows with your own.

I want to make a diagram in draw.io (also called diagrams.net), but I'm not technical. Please create it for me as draw.io code that I can copy and paste.

What the diagram is: the conceptual framework for my study.

Boxes (use these words exactly):
- On the left, stacked: "Perceived usefulness", "Perceived ease of use", "Institutional support"
- In the middle: "Attitude toward AI" (light blue)
- On the right: "Intention to use AI in academic writing" (light green)
- Above, in a dashed yellow oval: "AI literacy"

Arrows:
- Each box on the left points to the middle box
- The middle box points to the box on the right
- "AI literacy" has a dashed arrow pointing at the arrow between the middle and right boxes

Look: left to right, neatly spaced, simple academic style, no extra text or titles.

Please give me ONE code block only, starting with <mxfile> and ending with </mxfile>, so I can copy it in one go. After the code, tell me in 3 short steps how to open it in draw.io.

Step 2: what Canvas gives you back

<mxfile>
  <diagram name="Conceptual framework">
    <mxGraphModel dx="800" dy="400" grid="1" gridSize="10" page="1" pageWidth="827" pageHeight="300">
      <root>
        <mxCell id="0"/>
        <mxCell id="1" parent="0"/>
        <mxCell id="pu" value="Perceived usefulness" style="rounded=1;whiteSpace=wrap;html=1;" vertex="1" parent="1">
          <mxGeometry x="40" y="40" width="180" height="50" as="geometry"/>
        </mxCell>
        <mxCell id="peou" value="Perceived ease of use" style="rounded=1;whiteSpace=wrap;html=1;" vertex="1" parent="1">
          <mxGeometry x="40" y="130" width="180" height="50" as="geometry"/>
        </mxCell>
        <mxCell id="is" value="Institutional support" style="rounded=1;whiteSpace=wrap;html=1;" vertex="1" parent="1">
          <mxGeometry x="40" y="220" width="180" height="50" as="geometry"/>
        </mxCell>
        <mxCell id="att" value="Attitude toward AI" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#dae8fc;strokeColor=#6c8ebf;" vertex="1" parent="1">
          <mxGeometry x="320" y="130" width="170" height="50" as="geometry"/>
        </mxCell>
        <mxCell id="int" value="Intention to use AI in academic writing" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#d5e8d4;strokeColor=#82b366;" vertex="1" parent="1">
          <mxGeometry x="580" y="120" width="190" height="70" as="geometry"/>
        </mxCell>
        <mxCell id="lit" value="AI literacy" style="ellipse;whiteSpace=wrap;html=1;dashed=1;fillColor=#fff2cc;strokeColor=#d6b656;" vertex="1" parent="1">
          <mxGeometry x="475" y="20" width="120" height="45" as="geometry"/>
        </mxCell>
        <mxCell id="e1" style="endArrow=block;html=1;" edge="1" parent="1" source="pu" target="att">
          <mxGeometry relative="1" as="geometry"/>
        </mxCell>
        <mxCell id="e2" style="endArrow=block;html=1;" edge="1" parent="1" source="peou" target="att">
          <mxGeometry relative="1" as="geometry"/>
        </mxCell>
        <mxCell id="e3" style="endArrow=block;html=1;" edge="1" parent="1" source="is" target="att">
          <mxGeometry relative="1" as="geometry"/>
        </mxCell>
        <mxCell id="e4" style="endArrow=block;html=1;" edge="1" parent="1" source="att" target="int">
          <mxGeometry relative="1" as="geometry"/>
        </mxCell>
        <mxCell id="e5" style="endArrow=block;dashed=1;html=1;" edge="1" parent="1" source="lit">
          <mxGeometry relative="1" as="geometry">
            <mxPoint x="535" y="155" as="targetPoint"/>
          </mxGeometry>
        </mxCell>
      </root>
    </mxGraphModel>
  </diagram>
</mxfile>

In Canvas, write Mermaid flowchart code (left to right) for my conceptual framework: [list of constructs] with these relationships: [A → B, B → C, D moderates B → C]. Use exactly my labels. Then explain how to export it as SVG or PNG.

A figure you build and edit from code is your own drawing, but Gemini still helped create it. Record it in your AI-use log and disclose it if your faculty or journal requires (see 6.2).

Before you use an AI diagram

• Every label is spelled correctly and matches the terms in my text
• Every arrow matches my theory or method, and nothing extra was added
• It contains no numbers, data or results that I have not verified
• I have checked the rules of my faculty, conference or journal for AI-generated images
• I have added a disclosure in the caption or acknowledgements

Figure 2. Conceptual framework of the study. Diagram drafted with Google Gemini (Nano Banana Pro) from the author's sketch and verified and edited by the author.

Images made with Gemini carry an invisible SynthID watermark that identifies them as AI-generated, and some plans also add a visible watermark. Don't try to remove either. Disclose instead.

5.1 Layered editing

Edit in passes, from big-picture to fine detail. One focus per prompt gives better results than "fix everything".

5.2 Editing prompts

Proofread the text below for grammar, spelling and punctuation only. Do not change meaning, terminology or style. Present your changes in a table: original | revised | reason.

[paste text]

Reduce the following text by about 20% without losing any argument, data or citation. Show the shortened version, then list what you removed and why.

[paste text]

Identify informal phrases, overstated claims and unhedged generalisations in the text below. For each, suggest a more precise academic alternative and explain the difference.

[paste text]

5.3 Simulated peer review

Act as a rigorous but fair peer reviewer for [journal]. Review the manuscript section below. Give: (1) a summary of its contribution, (2) major concerns, (3) minor concerns, and (4) questions for the authors. Be specific and refer to sentences directly.

[paste section]

Only paste your own unpublished work. Never upload a manuscript you are reviewing for a journal. That breaches confidentiality (see Module 6).

5.4 Paraphrasing: handle with care

5.5 Writing in your own voice

Left to its defaults, Gemini writes in a generic "AI register": long, evenly balanced sentences, stock phrases ("delve into", "pivotal role", "in today's rapidly evolving landscape") and confident claims with few hedges. Examiners and reviewers notice, and it doesn't sound like you. You can fix this by giving Gemini a voice profile: a description of how you write, drawn from your own past writing.

Your voice fingerprint

A voice profile records measurable features of your style, not your content. Here is an example profile for a researcher in education:

Generic vs. personalised: the same notes, two drafts

• Red: generic AI phrasing. Delete it or say it plainly.
• Amber: overstated claim. Hedge it to match your evidence.
• Green: your habits: specific context, your hedging, your structure.

The personalised draft still comes from your notes and findings. The profile changes how it sounds, not what it claims. "Author, 2025" is a placeholder: add real citations yourself.

Six steps to a draft in your own voice

Worked example: drafting from your own past paper

Aisyah has a published paper she wrote as first author. She uses it as the voice sample for a new paper. Here is the full conversation in Gemini:

She then runs the voice check (below), rewrites the flagged sentences herself, checks every source, and saves the profile in a Gem for the rest of the paper.

Attached is a paper I wrote as [first/sole] author. I drafted the [sections] myself.
Step 1: Analyse only my WRITING STYLE in those sections (not the content) and summarise it as a voice profile of max 150 words, quoting short examples. Stop and show me the profile before going further.
Step 2: When I approve it, draft [section/paragraph, ~n words] for my NEW paper using ONLY the notes I give you. Follow the profile; do not reuse sentences, ideas or references from the attached paper; mark gaps as [ADD SOURCE] or [EXPAND].

Prompt templates

Below are [3] samples of my own academic writing. Analyse my STYLE, not the content. Describe:
(1) average sentence length and variation, (2) formality and vocabulary, (3) how I hedge claims, (4) active/passive voice and use of "I"/"we", (5) how I open and close paragraphs, (6) how I integrate citations, (7) my typical transition words, (8) spelling variant and punctuation habits, (9) phrases I never use.
Write it as a concise "voice profile" (max 200 words) that another writer could follow. Quote short examples from my samples.

[Sample 1]
[Sample 2]
[Sample 3]

Using my voice profile below, turn my notes into a draft [section/paragraph] of about [n] words.
Rules: use only the ideas, findings and sources in my notes; keep my terminology; do not add claims, statistics or citations; mark gaps as [ADD SOURCE] or [EXPAND]; avoid these words: delve, pivotal, tapestry, landscape, multifaceted, invaluable, revolutionise.

VOICE PROFILE: [paste]
MY NOTES: [paste]

Compare the draft below with my voice profile. List every sentence that does not sound like me (wrong length, wrong register, generic phrasing, over- or under-hedging) and say why. Do not rewrite them. I will revise them myself.

VOICE PROFILE: [paste]
DRAFT: [paste]

My notes are in Bahasa Melayu. Draft the paragraph in English, following my voice profile. Keep these terms in the original language, in italics, with a short gloss the first time: [list, e.g. sekolah kebangsaan, guru cemerlang]. Do not translate quotations from participants; keep them in Malay and give an English translation in brackets.

VOICE PROFILE: [paste]
NOTES: [paste]

5.6 When the output isn't what you expected

Even a good prompt sometimes gets a poor answer: too generic, the wrong tone, a changed meaning, or ignored instructions. Don't start over in frustration, and don't accept it either. Name the problem, then fix that specific problem.

Diagnose it: symptom, likely cause, quick fix

The recovery routine

Work down the steps. Most problems are solved by step 2 or 3.

1. 1

   Name the problem

   Say exactly what is wrong, using the table above: "the tone is too informal", not "this is bad".

2. 2

   Give specific feedback

   Quote the problem sentence, say what you want instead, and say what must not change.

3. 3

   Show an example

   Paste a sentence or paragraph that is "like this". One good example beats ten adjectives.

4. 4

   Break the task down

   Ask for one thing at a time: structure first, then wording, then length.

5. 5

   Reset

   Use Edit prompt on your original message, or start a new chat with a clean summary. Long chats drift.

6. 6

   Change the tool

   Switch to the more capable model, use Gemini Notebook for source-based answers, or Canvas for targeted edits.

7. 7

   Stop and write it yourself

   If three rounds haven't worked, the fastest route is usually your own pen. Use Gemini to check the result instead.

Worked example: fixing a "tighten this" request

The fix took one round because the feedback named each problem (step 1), said exactly what to keep (step 2), and asked for a change table so the result was easy to check.

That's not what I needed. Problems: (1) [quote the issue and say what's wrong], (2) [...].
Redo it. Keep exactly: [claims / numbers / citations / terms that must not change].
Change only: [what you want changed].
Before you answer, restate my rules in one line each, then show the revised version.

Here is a paragraph I wrote that has the style I want:
[paste your example]
Rewrite the text below to match that example's sentence length, hedging and tone. Do not change any claims or citations.
[paste text]

Summarise this conversation as a short brief I can paste into a new chat: the task, the final agreed rules, the format I want, and anything I said I did NOT want. Bullet points only, no commentary.

5.7 Responding to reviewer comments

A good response letter is clear, polite and easy for the editor to check. Gemini is useful for organising long reviews and finding the right tone, but every decision (what to accept, what to change, where you disagree) is yours. Revise the manuscript first, then write the responses.

The response table editors expect

Illustrative example; "Smith (2023)" is fictional.

Below are the reviewer comments on my manuscript. Split them into numbered single points (R1.1, R1.2, R2.1 …), keeping each reviewer's exact words. Label each as major or minor, and by type: clarification, additional analysis, literature, structure or language. Do not draft any responses yet.

[paste reviews, with names removed]

For each comment below I have given my decision and the change I made. Draft a concise, polite response for each, as a table: comment | response | change in manuscript (page, line).
Rules: use only the changes I list; never promise or describe a change I did not make; thank each reviewer once at the start, not in every row; no flattery; British English.

R1.1 — decision: [agree / partly / disagree] — change: [what you did, page, line]
R1.2 — …

I disagree with comment [R2.3] because [your reasons and evidence]. Draft a respectful response that acknowledges the reviewer's point, explains my reasoning with this evidence, and states clearly that I have not made the change. Do not concede the point, and do not add arguments I did not give.

Review my response letter below. Flag (1) any defensive, dismissive or over-apologetic wording, (2) any comment that has not been answered, and (3) any response that mentions a change without a page and line reference. Do not rewrite the letter.

[paste letter]

6.1 Core principles

6.2 Universiti Malaya's AI policy and guidelines

On 15 March 2026, Universiti Malaya released two documents that govern AI use in teaching, learning and research: the Universiti Malaya Academic Policy on Artificial Intelligence (AI) and the Universiti Malaya Guidelines on the Ethical Use of Artificial Intelligence (AI). They apply to all students and academic staff, across coursework, assessments, research projects, final-year projects, dissertations, theses, and online or blended learning.

"AI should support learning and creativity rather than replace students' own thinking, analysis, and academic effort."

Universiti Malaya Academic Policy on Artificial Intelligence (AI)

The five levels of AI use

For every assessment, your lecturer sets a permitted level of AI use and tells you at the start of the semester. The level, not your own judgement, decides what is allowed. The right-hand column maps each level to what you have practised in this workshop.

1. Level 0

   No AI use

   "AI tools are not permitted."

   No Gemini at all, not even for grammar.

2. Level 1

   Minimal use

   "Basic tools such as grammar or spelling checkers may be allowed."

   Proofreading only: the "track-changes style edit" in 5.2. No rewriting.

3. Level 2

   Limited use

   "AI may assist with idea generation or concept clarification."

   Concept explainer and Socratic partner (2.4), research question refiner (3.1), search terms (3.3).

4. Level 3

   Open use

   "AI may support learning and content development with proper declaration."

   Gemini Notebook (3.4), outlining and drafting support (Module 4), Gems, Canvas and voice profiles (2.5, 4.7, 5.5), all declared.

5. Level 4

   Integrated use

   "AI may be required as part of the assessment activity."

   Tasks where using and critiquing Gemini is the assignment, e.g. evaluating its output.

Theses, dissertations and research reports

Declaring AI use: theses and journal papers

UM's guidelines ask you to (1) declare the AI tools used, (2) explain how each tool was used, and (3) provide logs or evidence if requested. The easiest way to meet the third requirement is to keep a simple log as you work:

Declaration of the Use of Artificial Intelligence (AI)

In accordance with the Universiti Malaya Academic Policy on Artificial Intelligence (AI), I declare that I used the following AI tools in preparing this [thesis/dissertation/research report], with the knowledge and approval of my supervisor:

- Google Gemini (Gemini Notebook): to summarise and compare literature that I had located and read, in Chapter 2.
- Google Gemini (Canvas): to check grammar and clarity in Chapters 1–5.
- Google Gemini (Canvas): to generate draw.io code for Figure 3.1, which I edited and verified in draw.io.

The research questions, data, analysis, interpretation and conclusions are my own. I reviewed and verified all AI-assisted content, take full responsibility for this work, and can provide a log of my AI use on request.

For a journal paper, use the publisher's wording and location: many ask for a short statement in the Methods, the Acknowledgements or a dedicated declaration before the references. Name the tool, the purpose and your responsibility.

During the preparation of this work, the author(s) used Google Gemini to [improve language and readability / generate an initial outline / suggest search terms]. After using this tool, the author(s) reviewed and edited the content as needed and take full responsibility for the content of the publication.

This is a workshop summary, not the official text. Quotations are taken from UM's published summaries of the two documents. Always read the full policy and guidelines on the UM policy page and the UM ethical-use guidelines page, and follow your lecturer's stated AI level and any faculty-specific thesis rules.

6.3 Traffic-light guide

This is a general guide. At UM, the AI level your lecturer sets for each assessment (6.2), your supervisor's instructions and the target journal's author guidelines always take precedence. Check them before you start.

6.4 Data privacy checklist

• I have removed names and identifiers from any participant data
• I am not uploading material under an NDA, embargo or ethics restriction
• I am not uploading another researcher's unpublished work
• I know whether my account's chats may be reviewed or used for training
• I am using my institution's approved account where one exists

6.5 About AI detectors

AI-text detectors are unreliable. They produce false positives, especially for non-native English writers, and false negatives. Don't rely on them to "prove" your work is human-written, and don't use "humanising" tools to disguise AI text. The best protection is transparent use and a record of your own process: drafts, notes and version history.

6.6 Supervision in the AI age

UM requires students to tell their supervisor before using AI in a thesis, and supervisors to approve it (6.2). In practice, that works best as an ongoing conversation rather than a single permission. This section is for supervisors and students together: agree expectations early, know what good AI use looks like in a literature review, and build trust through transparency.

Establish AI expectations early

At the start of the project, and again at each milestone (proposal, data collection, writing, submission), the supervisor and student should discuss five questions:

1. 1. Allowed

   Where can AI be used?

   e.g. grammar and clarity, search terms, Gemini Notebook summaries of papers the student has already read

2. 2. Limited

   Where should AI use have boundaries?

   e.g. drafting only from the student's own notes; figures only as editable draw.io code; nothing beyond the assessment's UM AI level

3. 3. Excluded

   What must the student do independently?

   e.g. research questions, data analysis and interpretation, the argument of the literature review, the conclusions

4. 4. Disclosure

   When and how should AI use be declared?

   e.g. AI-use log shared before each meeting; thesis declaration and journal disclosure as in 6.2

5. 5. Verification

   How will understanding and ownership be demonstrated?

   e.g. explaining a chapter's argument without notes, walking through key sources, drafts and version history at milestones

AI-use agreement: [student name], [degree], supervised by [supervisor name]
Project: [title]                     Date agreed: [date]     Review at: [next milestone]

1. ALLOWED: AI may be used for: [ ... ]
2. LIMITED: AI may be used only with these boundaries: [ ... ]
3. EXCLUDED: the student will do these independently: [ ... ]
4. DISCLOSURE: AI use is recorded in an AI-use log, shared [before each meeting / monthly], and declared in [thesis declaration / journal disclosure].
5. VERIFICATION: understanding and ownership will be shown by: [ ... ]

Accounts: research material is used only in [UM Workspace for Education account].
Applicable UM AI level for assessed work: [0–4].
Signed: [student]   [supervisor]

Literature reviews in the AI era

The literature review is where AI shortcuts are most tempting and most visible. Use this to set the standard, and Module 3 (3.4–3.7) for the techniques.

"Discover" means search terms and related ideas: the sources themselves still come from scholarly databases (3.2).

Maintaining trust through transparency

Outcome: TransparencyTrustResponsible AI use

Supervision tips for the AI age

7.1 Key lessons from Modules 1–6

7.2 Your Gemini toolkit

Not sure which tool to use? Start here.

7.3 The workflow on one line

7.4 Five questions to ask before every use

• Allowed? What AI level applies to this task, and does my supervisor know?
• Safe? Is this material allowed on the account I'm using?
• Mine? Do the ideas, analysis and argument still come from me?
• True? Can I verify every fact, number and citation against a real source?
• Recorded? Have I logged this use so I can declare it?

7.5 Keep learning on your own

A four-week practice plan

How to work things out yourself

Useful links

• Gemini Apps Help: official how-to guides for every feature
• Gemini Notebook (NotebookLM) Help: sources, notebooks and limits
• Gemini for Education: what your Workspace for Education account includes
• UM Policy on the Use of AI and UM Ethical Use of AI guidelines
• COPE (Committee on Publication Ethics): publication ethics guidance, including AI and authorship
• Your target journal's Guide for Authors: search for "AI", "generative" or "artificial intelligence"