Ch 1–4
Foundations
What an LLM actually is (and why "AI" is the wrong word), supervision as the skill you already have, and classifying your data before it goes anywhere near an AI tool.
A plain language reference guide for academics
Understanding & Using AI as a Scientist
Your institution has an AI policy and your students are already using it, but nobody has told you what you're actually allowed to do.
Get the guide · A$49Instant download. Not useful? Email me within 5 days for a refund.
The problem
Twice in five years, our jobs have been remade.
First in 2020, when the pandemic forced every academic to rebuild teaching and research from a kitchen table. Then in late 2022, when ChatGPT landed and it suddenly felt like our students had the answer sheet to every assessment we'd spent a decade perfecting.
Now there's a policy, a framework, and a training module, and still nobody has answered the questions you actually have. Can my research data go in this thing? Do I have to disclose it? What do I do about the obviously-AI student email? Why is everyone else getting useful output while mine reads like customer service?
We are not behind because we are not capable. We are behind because the rules of our jobs keep changing and no one has taken the time to catch us up.
What this is
Understanding & Using AI as a Scientist is a 25-chapter plain language reference guide for academics working out where AI fits in their research and teaching, or who just need enough knowledge to guide their students. It assumes no technical background. It covers what these tools actually are, how to decide what data can safely go near them, a four-check loop for verifying what comes back, disclosure templates for manuscripts and grants, and how to handle AI in your teaching, your peer review, your ethics application, and your students' work. It shows you how to make AI output sound like you, instead of the garbled nonsense you're used to.
It's for the scientist who feels left behind by the AI revolution, or who's been deeply dissatisfied by an LLM while everyone else sings its praises.
What's inside
Every question you actually arrive with, in the order you arrive with them.
Ch 5–7
Trust & accountability
What actually goes wrong (hallucination, sycophancy, scope creep), the four-check loop for catching it, and what disclosure really requires when your name is on the work.
Ch 8
Get set up
The full walkthrough: setting up Claude safely, training off, folder structure, teaching it your voice, connecting it to the literature, and three starter tasks where a bad answer costs you nothing.
Ch 9–13
Research & integrity
Participants who never consented to this, the assumptions hidden in the model, open science documentation, writing the paper, and where copyright risk actually converts to yours.
Ch 14–16
Peer service
Reviewing papers without breaching confidentiality, drafting decision letters as an editor, and the blunt chapter on why grant review is a hard no.
Ch 17–21
Teaching
Taking admin off your plate, making the actual teaching better, designing assessment AI can't shortcut, teaching AI literacy, and the essay on what to do about the obviously-AI student email.
Ch 22–24
Supervision & duty
The conversation to have with your PhD students, getting AI use past your ethics board, and an honest accounting of what your queries actually burn.
Ch 25 + back matter
Reference
The prompting vocabulary (why it ignores half of what you ask), plus a full glossary and prompt tables for behavioural and social scientists.
What you take out of it
Six things you'll still be using in six months.
-
The four-check loopSource · Scope · Voice · Ownership. Every output, every time.
-
Green / amber / red data classificationWith a one-page decision tree, so the call is made before the file is opened.
-
The 11-point de-identification checklistIncluding the trip-wire most people miss.
-
Disclosure templatesFor manuscripts, grants, and theses, matched to venue.
-
The session logThe ten-field paper trail that satisfies the Australian Code and your future self.
-
A five-day fast-trackFor the reader who wants to give this a real shot this week.
How it's built
Every worked success is paired with a worked failure.
Every chapter has the same shape: an entry checklist, a by-the-end box, the content, and one gate question you should be able to answer before moving on.
The literature review that produced a real synthesis sits next to the one that fabricated citations. The analysis that caught a problem sits next to the one that smoothed a problem over. The success shows you what good looks like. The failure shows you what the same tool produces when supervision slips. You need both.
Who this is for
Two lists. Read both.
This guide is for you if
- You feel behind, and you're tired of feeling behind
- You tried an LLM, got garbage, and quietly gave up
- You supervise students who are already using it, and you need to be ahead of them
- You have no technical background. None is assumed
This guide will not
- Tell you whether to use AI. That's your decision and your institution's
- Certify any tool as safe for any data category
- Replace your ethics office, your DMP, or your integrity office
- Teach you to code, or chase every model release
About the author
I'm Dr Kristyn Sommer, developmental scientist, autism advocate, and science communicator to a community of 500,000+. A year before writing this guide I walked back onto campus after two maternity leaves feeling completely lost. Even Copilot felt beyond my reach.
I caught myself up in a flurry of necessity and the kind of deep, intense focus that only an autistic special interest can produce, and then I wrote down everything I wished someone had told me, in the order I wished they'd told me.
The honest bits
- This guide was written with Claude's assistance (drafting, flagging inconsistencies, generating templates), with every output reviewed, edited, and approved by me. My voice, my standards, my responsibility. It's the supervision practice the guide teaches, applied to the guide itself. Full disclosure inside.
- I teach through Claude because it's my daily tool and lived experience beats secondhand advice. The principles transfer to ChatGPT, Copilot, and whatever arrives next.
- It reflects my practice as at May 2026. It deliberately teaches durable habits rather than chasing model releases.
- It is not institutional policy and not legal advice. Always check your institution's current AI policy.
This guide keeps changing
I'm always updating what I know, and I'd rather hear that something is wrong than leave it sitting there being wrong. Feedback and constructive criticism are genuinely welcome, and most of what changes between versions comes from readers. There's a two-minute form for it, and updates are free once you've bought it.
Licence and limits
What you're buying, and what it isn't.
The licence
One reader. You. Print it, annotate it, keep it on every device you own, and quote it in a talk with attribution. All fine, and all intended.
What it doesn't cover is uploading it to a shared drive, a course site, or a repository, or passing the file on. If your department or lab wants it, that's good news and there's a team licence. Email me at hello@drkristynsommer.com.au and it costs less than buying it one at a time.
What this is not
It isn't institutional policy, legal advice, ethics approval, or data-management advice. It doesn't replace your ethics committee, your data management plan, your research integrity office, or your institution's AI policy.
It reflects my own practice as at August 2026, and it's provided for education and professional development. Where it tells you to check something with your institution, that instruction is the advice. Check it.
Questions
I'm not in the behavioural or social sciences. Is it still relevant?
The worked examples lean on my field (developmental psychology), but the frameworks (data classification, the four-check loop, disclosure, assessment design) are discipline-agnostic. If you have research data, students, and a name to protect, it applies.
Do I need any technical background?
No. The guide starts before the jargon does. There's a glossary for every term, defined once, in plain language.
I use ChatGPT or Copilot, not Claude. Will it still work for me?
Mostly, yes. One chapter (the setup walkthrough) is Claude-specific. The other 24 are about judgement, data, disclosure, teaching, and supervision. Platform-agnostic by design.
Will it be out of date in six months?
The tool screenshots will age; the practices won't. The guide explicitly refuses to track model releases. It teaches the habits that survive them.
What exactly do I get?
A 196-page PDF, delivered by email link immediately after purchase. Read it on anything; print it if you're a print person. It's designed for both.
Can my department or lab buy it for staff?
Email me at hello@drkristynsommer.com.au for team licences.
I'm a PhD or grad student and A$49 is a lot.
It is, on a stipend. Email me at hello@drkristynsommer.com.au and ask, and I'll send you a code for 80% off. You don't have to prove anything or explain yourself. I'd rather you had it.
What if it's not useful to me?
Email me within 5 days and I'll refund you. I'd rather have the feedback than the A$49.
The catch-up
Stop feeling behind.
The rules changed and nobody caught you up. This is the catch-up: 25 chapters, in plain language, from someone who was exactly where you are.
Get the guide · A$49Instant PDF download. 5-day refund, just email me.