The faculty AI Ladder, devoted to research. Fifteen exercises, at your own pace.
You already have the judgment — you built it over a career of reviewing, revising, and being reviewed. This ladder points that judgment at AI. Fifteen short exercises across five rungs, each rung setting up the next, all run on your own working manuscripts.
Every exercise is drawn from how working scholars actually use AI — and from where it actually fails them. Reviewer simulations catch circular arguments before Reviewer 2 does; citation audits catch the one wrong issue number in a hundred references; and the occasional “nobody has theorized this” claim turns out to have been theorized in 2015. The ladder exists so you find these things out cheaply.
Why this order? Because the failure mode for faculty is not naivety — it’s over-trust of fluent output on exactly the work where the stakes are highest. So the climb starts loose, then tightens: by Rung 5 you are verifying the AI the way you’d verify a first-year’s literature review.
Any AI chatbot works — ChatGPT, Claude, Gemini, Copilot. Free versions are fine. Switching between them is fine too — Rung 4 requires it.
Notes are optional. If you want to remember what happened, a line after each exercise is plenty: what surprised you, and what it got wrong. Nobody collects them; they are only there for the write-up at the end.
The starters are a floor, not a ceiling. Every exercise gives you something to paste so you never face a blank box. Swap in your own material — a starter you changed is worth more than one you pasted clean.
Each rung ends with a self-check. If the answer is no, repeat the rung’s last exercise with new material before moving on. Nobody sees your answer.
| # | Exercise | The question to ask |
|---|---|---|
| Rung 1 — PlayNothing on the line. Your ideas, low stakes. | ||
| 01 | Explain your research to a smart neighbor Paste to start: “Explain my research to a smart neighbor with no academic background: [paste your abstract].” | How well did it capture your research — and was what it dropped the jargon, or the contribution? |
| 02 | Pitch your idea to a skeptical reviewer Paste to start: “You’re a skeptical senior scholar in my field. I’ll pitch a research idea — push back hard, no encouragement.” | Did the idea hold up under real pushback, or crumble on contact? |
| 03 | Write the abstract of the paper you haven’t written yet Paste to start: “Here’s a research direction I’m toying with: [describe it]. Write the abstract of the finished paper — including the finding. Then write a second abstract where the finding came out the other way.” | Would either version be worth publishing? If the null result is as interesting as the finding, the question is a good one. |
| Before you climb on: could you talk to an AI about your own research agenda without a script? | ||
| Rung 2 — DirectYou steer. The answer depends on you. | ||
| 04 | Make it attack your hypothesis Paste to start: “My hypothesis: [state it]. Write the strongest case that it’s wrong — the version that would worry someone who holds it. No both-sides summary.” Then: “Which side is right? Commit to one.” | Which side actually won — and did the attack find something your literature review missed? |
| 05 | Reframe the same question, watch the field change Paste to start: “What does the literature say about [your question]?” Then: “Same question, framed as a [different field’s] problem.” | What did the frame cause — and has your usual framing been hiding a neighboring conversation? |
| 06 | Have it quiz you on your own literature Paste to start: “Quiz me on the literature of [your area], one question at a time. Go harder where I miss. Stop when you find a paper or debate I should know but don’t.” | Where did your command of your own field actually run out — and what did you do about it? |
| Before you climb on: could you say what your framing changed and what it didn’t? | ||
| Rung 3 — ApplyA real manuscript of your own. This is where it earns its keep. | ||
| 07 | Run a reviewer simulation before Reviewer 2 does Paste to start: “Act as three reviewers for [target journal]: one sympathetic, one methods-focused, one hostile. Review this draft — specific page-level objections only: [attach or paste].” | Which objections are real — and which are a fluent reviewer performing rigor? |
| 08 | Audit your argument’s logic Paste to start: “Map the core argument of this section as premises and conclusions (P1, P2 → C). Flag circularity, hidden assumptions, and any place the conclusion doesn’t follow: [paste].” | Is the gap it found real? A paper can sound right without being right — this is the exercise that tells them apart. |
| 09 | Compress the paper and see what survives Paste to start: “Compress this manuscript to one page, then to half a page, keeping the core argument intact: [attach].” Then: “List what each cut lost.” | Was what got cut the scaffolding, or the contribution? If the contribution vanished at one page, the paper has a problem. |
| Before you climb on: could you point to a change in your manuscript that came from these exercises — and one suggestion you rejected? | ||
| Rung 4 — InterrogateExamine the tool itself. The call stays yours. | ||
| 10 | Two AIs review the same manuscript Paste to start: give the same draft and reviewer prompt to two different chatbots. Then: “Another AI reviewed this paper and said: [paste]. Where do you disagree, and which disagreement matters?” | Which catches are real, which are style — and which model would you trust on what? |
| 11 | Make it argue your construct away Paste to start: “Argue that my construct [name it] is just [its nearest neighbor] renamed — the strongest reduction case a hostile reviewer could make.” Then: “Now name the test that would separate them.” | Does your construct survive its nearest neighbor — and was the case against it stronger than you expected? |
| 12 | Mark the line: its venue recommendation, your call Paste to start: “Given this abstract, recommend three target journals with reasons, labeled RECOMMENDATION. Then list every consideration only I can weigh — career stage, school lists, review speed — labeled MINE.” | Exactly where did the AI stop and you start? This is the point of the rung. |
| Before you climb on: could you name a disagreement between two AIs about your paper that matters, and make the call yourself? | ||
| Rung 5 — VerifyIndependence. The rung that protects your name on the byline. | ||
| 13 | Verify citations one by one against the record Paste to start: “For each reference in this list, verify against the published record that the authors, year, journal, volume, issue, and pages are exactly right. Check each one individually — no sampling: [paste references].” | Did the AI actually check, or perform checking? Spot-verify two of its “verified” entries yourself. |
| 14 | Trace your load-bearing claim Paste to start: “My paper claims: [your ‘no one has shown / theorized X’ claim]. Search hard for the paper that falsifies this. Finding one is a success, not a failure.” | Does your gap claim survive contact with the literature? A claim that dies here dies cheaply; at review it dies expensively. |
| 15 | Check the checker: the clean-room re-run Paste to start: open a new chat with no prior context. “Review this manuscript cold: [attach].” Compare against the review you got in your long working conversation. | How much had your working conversation shaped what the AI told you — and what does an unprimed reader actually see? |
| Before you climb on: could you tell a co-author when to trust this and when not to? That is the top of the ladder. | ||
No schedule, no project tie-in. Nothing here depends on a particular paper — where an exercise needs material, use whatever manuscript, data, or half-idea you happen to be carrying. The order is a suggestion; it exists only because each rung assumes the one below it.
From explaining your abstract to a neighbor to running clean-room checks on the checker. That distance is the lesson. Start wherever you are — and welcome to the Ladder.
One thing is worth writing, once, at the end: a short account of your climb. One to two pages. What you tried, which exercises surprised you and how, and what you would now hand to AI in your research without a second thought and what you would not let it near. Name the exercises. Bring it to the college’s faculty AI conversation — a person reads it and replies.
If your write-up shows the climb, guide a colleague up. The ladder spreads one climber at a time — and the college is building its AI cadre from exactly the faculty who have climbed it.