What to hand somebody who asks about AI
What to hand somebody who asks about AI
Free, ungated material for subject guides and reference desks — no signup, no lead capture
You are being asked about AI by everyone. You need practical answers. However? Most of what a search returns is written by companies selling something.
This page is a byproduct of our work. It is practical, it is current, and it is free.
Everything here is free to read, free to link, and free to forward.
No paywall, no freewall. Free in the way human beings mean free — no email address to hand over, no download to unlock, no course, no trial, and no follow-up sequence if you click something.
We are a doctoral coaching practice, so we do sell something — but not on any page we would ask you to link, and not to anybody who arrives from one. This is part of us 'giving back.'
Link any of it. Nobody will know, and nobody will follow up.
Back to The Ghost in the Machine
One of nine pages on AI and academic honesty. Browse all 62.
Related links
The most useful thing we publish is a log of what these tools actually did — and do, every day — to real text.
Not might do. Not tend to do. Dated cases from our own work, each with the passage before and the passage after, and the model named as high-end and frontier-tier.
Nine so far, and they are nine distinct failures rather than nine instances of one.
A real degree reported as invented and deleted — by a model that had researched the degree itself minutes earlier. Two precise sentences — the critical pivot of a peer-reviewed study — merged as a duplication. A specific figure replaced with a vague one. Four insertions lost in a single session. A list copied to a second page that came back two items short.
Those are different mechanisms with different causes, and a reader who has learned to watch for one will not catch the others. That is why the list is a list, and why it is worth putting the whole of it in front of somebody rather than the headline.
What every case has in common: the output read perfectly well. No error, no flag, no gap on the page. The prose closes over the loss and the writer reads straight past it.
This is why it matters at a reference desk. Somebody who has been told AI hallucinates will check their citations. Nobody checks whether a paragraph they wrote three weeks ago still says what it said.
The Ghost in the Machine — the nine, with the cases behind them.
And it is a work in progress by definition.
We add to it when something happens. The list was seven cases a fortnight ago; two were added on 21 August, and one of the original seven turned out to be two separate failures rather than one.
So it will be longer next time you look, and we would rather say that than present a fixed number as though the problem were finished.
And it will not be fixed by a better model. Our founder is a computer scientist, and his position is that these faults are largely inherent to how statistically based systems work — not defects awaiting a patch. Plan accordingly.
If you or somebody at your institution has a case worth adding — dated, with both versions of the text — we would like to see it. We will credit it or not, as you prefer.
What to put in a subject guide
Six pages, and you can link any of them without linking us.
What AI actually did to real text — the case log, nine cases and growing. For anybody who suspects a tool damaged their work, or who does not yet know that is possible. It also carries a checklist of the nine failure modes, which is the part worth putting in a guide.
Wrongly accused of AI use — what a detector score means, what the research shows, and what to do in the first hour. It carries a visible review date — the page is checked on a fixed cadence because the research moves — and seven peer-reviewed and institutional sources.
Written for supervisors, not students — what a Chair is actually responsible for, and how to raise AI without accusing anyone. Free to read and free to forward. We have not found another practice publishing for this audience.
Backups, and why syncing is not backing up — the least glamorous page we have.
Is this journal real? — a checking guide for predatory and cloned journals, with the manual method and an easy to use prompt for those who want one. Free, and there is a printable version — the format most use at a reference desk.
And AI and institutional policy, if the question has reached whoever writes the rules where you are.
Six things you can send or print and hand over
Nothing to unlock, and no form in front of any of them.
What lives in your laptop bag — printable checklist for power, adapters, and conference travel. The reference-desk question that is not about writing.
Help Needed Checklist — sixty-three questions that tell a writer what kind of help they actually need, sorted into four levels. The one to hand somebody who says they are stuck but cannot say how.
Is this journal real? — the journal checking guide, manual method first, with an AI prompt for those who want one.
The Ghost in the Machine — the nine failure modes, with tick boxes. Valid as of August 2026. Hand it with the case log, or on its own.
Book writing decision tree — for anybody asking whether their thesis should become a book. Purpose, audience, budget, publishing, and the questions nobody asks until it is late.
Clarity exercise — fillable. For a writer who cannot say in one sentence what their study is about — which is more of them than you would expect.
One test that needs no software
We call it the Kindergarten Counting Problem.
AI-assisted prose announces three reasons and gives four. It says both of these and lists three. It refers to the four criteria above when there are five.
Fluent, confident, and wrong about something no human writer gets wrong past a second draft.
It takes a minute to check, it needs nothing installed, and it is a useful tool for any reference desk.
The full set of tells, including what they do and do not prove.
Before you link anything
You should not take our word for any of this, and we have tried to make checking quick.
Our publication record — actual peer-reviewed work, not a bio.
References rather than testimonials — real past clients you can telephone.
And when we say we have not found something, we mean we looked. If you find a counterexample to any claim on this site, we would like to know — tell us and we will correct it and say so.
What this is not
Not a general writing resource. Purdue OWL is better at that and always will be. We link them ourselves.
Not a substitute for your institution's own policy. Ours is the outside view; the rules your users must follow are the ones their institution sets.
And not neutral about AI. We use these tools daily, we say so, and we publish the many faults they have and their consequences to text. A page claiming AI is uniformly dangerous would be as unreliable as one claiming it is safe.
We are not asking you to.
But if somebody's situation is beyond a subject guide — a returned dissertation, an accusation with a hearing attached, a writer who has stopped working — here is what we do, and much of it is free.
And if you would rather simply link the material and never hear from us again, that is entirely fine. It is why none of it is gated.
One of nine pages on AI and academic honesty. Browse all 62.
