Most students treat AI as a single category with a single risk level-either it’s allowed or it isn’t. That framing collapses a meaningful distinction. IB English students now move between chatbots that comment on drafts, automated rubric-scoring platforms, vocabulary aids, and close-reading tools, all sitting within the broader set of English revision resources available for Paper 1, the IO, and the HL Essay. The preparation task you point a tool at-planning, drafting, revising, or checking criteria alignment-determines both its benefit and its risk. Judge each by what part of the assessed thinking it touches, not by the logo or the hype.
The rule that keeps AI useful rather than compromising is simple enough to state: diagnose, don’t draft. Before sending a prompt, ask whether you’re asking the tool to create interpretation or to evaluate interpretation you’ve already written. Only the second is integrity-safe. A diagnostic prompt focuses on existing work-“Here is my paragraph; which claim is least supported, and what evidence from this text would I need to add?” A generative prompt asks the tool to supply ideas or structure outright-“Write a commentary for this text,” or “Give me three insightful points I can use.” If the output contains a new interpretive claim you didn’t already hold or can’t justify aloud, treat it as off-limits to adopt; at most, treat it as a question to test against the text. That rule is clear in the abstract. The harder question is how to apply it when you’ve already written something under time pressure and specific criterion gaps are staring back at you.
Paper 1 Commentary-The Five-Step AI-Feedback Drill
Unseen commentary practice can accumulate hours without building real skill when the only feedback loop is your own re-reading. You improve what you already notice; blind spots persist. The fix is structured feedback, and the AI-assisted drill makes that accessible without a teacher on demand. Take an unseen passage, write a full timed commentary without tools, then paste one paragraph into an AI assistant and ask it to separate supported claims from unsupported ones-marking which sentences are anchored in precise textual references and which are just assertions. That diagnosis shows where the claim-evidence chain holds and where it breaks.
In a second pass, target the technique-versus-effect problem that Paper 1 criteria consistently penalize. Ask the tool to mark where you’ve named a device and where you’ve actually explained how the choice shapes meaning or reader response, then rewrite one weaker paragraph yourself-keeping your interpretation, tightening the logic and the evidence. A four-week randomized controlled study with 259 university students found that AI feedback on drafts improved organization and content development beyond a control condition, which supports using structured, feedback-only prompts rather than letting the tool produce the analysis.
The drill isn’t complete until you’ve checked for transfer. Take a different unseen text and write another timed commentary with no AI support, then compare the two pieces. Look for clearer topic sentences, stronger claim-to-quote links, and more explicit comments on effect appearing in the second script without any prompting. If those gains show up only when the AI is on screen, the tool is doing the thinking; if they show up in the independent script, the feedback loop is working. What this transfer test can’t account for is the one IB task where you don’t control the conditions at all-no second draft, no revision cycle, no chance to return.
Individual Oral-A Three-Stage Preparation Workflow
Written tasks have a forgiving rhythm: draft, revise, iterate. The IO has none of that. Once you’re speaking, the argument is live, the structure is committed, and whatever scaffolding shaped it-including any AI-assisted framing-has to hold under questioning without a prompt in sight. That pressure calls for a three-stage workflow where AI’s role is strictly bounded. First, brainstorm global issues independently, then use AI only to test whether each framing gives enough room for genuine analytical comparison between the two texts. Second, run a pairing check by asking where the texts create real contrast rather than collapsing into similarity, using that as a check against your own notes. Third, draft your opening statement independently and use AI only to flag whether it states a debatable argument about the global issue or simply describes content. The tool diagnoses; you choose the argument and the wording.
A 2025 systematic review of 12 studies on generative-chat feedback found improvements in structure, cohesion, language, and students’ self-regulation, while also noting that feedback can be unreliable without clear pedagogical guidance. That second point matters most for the IO. Any analytical structure you can’t sustain under live questioning is a liability, not an argument-and no amount of AI feedback will protect you once it collapses. So before seeking more feedback on your opening, run a quick oral check. Say your opening aloud. Then, without tools, answer each of the following in roughly sixty seconds. What exactly are you claiming about the global issue in Text A? Which specific textual choice proves it, and what effect does it create? What is the comparative tension-not just a parallel-that Text B introduces? And what counter-reading would weaken your argument, and why are you rejecting it? If you can’t move cleanly through those four, treat the opening as over-scaffolded, strip out any AI-shaped phrasing, and rebuild from the texts. That kind of revision is available in preparation. It isn’t once the submission is on the page and under external assessment.
HL Essay-Navigating the Highest-Stakes Integrity Line
The HL Essay carries the highest integrity stakes, and the real test isn’t school policy language-it’s whether you can reconstruct your reasoning aloud, tracing each move from passage to claim without glancing back at anything the AI wrote. If you can’t, that section isn’t independently authored, whatever it looks like on the page. From that principle, the line between permitted and prohibited use is practically clear: testing whether a proposed line of inquiry is focused enough, getting feedback on whether a thesis is specific and arguable, and sanity-checking citation formatting are all diagnostic uses. Asking the tool to draft analysis paragraphs, or to supply interpretations and claim sequences you then adopt, is not. Rewrite any section that fails the reconstruction test from your own notes and textual evidence before going further.
Current IB expectations treat the HL Essay as independently authored work, and schools can reasonably question scripts that depart sharply from a student’s established voice or that the student cannot orally justify. Used within those limits, AI feedback can function as one more set of eyes on a draft. A quasi-experimental study with 60 English as a Foreign Language undergraduates found no statistically significant difference in writing gains between AI feedback and teacher feedback conditions (Cohen’s d = 0.10), leading the authors to describe AI as a scalable complement when carefully scaffolded and ethically deployed. For HL work, that means using an assistant to flag gaps in coherence or thesis precision, then making every substantive analytical move yourself.
Protecting Independence and Staying School-Aligned
Improvement on a practiced draft doesn’t automatically carry into examination conditions. Without some form of measurement loop, there’s no reliable way to distinguish genuine skill development from AI-assisted masking-where feedback refines the specific draft you’ve worked on while leaving the underlying analytical capacity untouched. The following cadence provides that check.
- Log each AI-assisted session in four fields: task (Paper 1 / IO / HL); what you produced before AI (one line); what the AI flagged (1-2 short diagnostics); what you changed (1 bullet, in your own words).
- Within 48 hours, do a no-AI transfer test: Paper 1-one paragraph on a new extract; IO-a 60-90 second opening from your outline; HL-one paragraph tracing claim → evidence → effect.
- Score transfer 0-2: 0 = same weakness; 1 = some improvement but still needs prompting; 2 = clear improvement without prompting.
- If you record two “0” scores in a week, restrict AI to post-attempt feedback and focus prompts on criteria/evidence checks only.
- Whenever you cannot explain a revised claim aloud without echoing the AI’s phrasing, rewrite that section from scratch using only your notes and the text.
Because school rules differ and are evolving, confirm your approach matches local academic honesty guidance. Check your school’s current policy for references to AI or digital assistance, and ask your IB English teacher if a specific use case is unclear. Keep a brief record of which tools you used and at which preparation stages so that, if you’re ever asked, you can describe your process accurately and confidently.
Using AI to Strengthen Your Own Thinking
Across Paper 1, the IO, and the HL Essay, what determines whether AI tools help or hurt is not how advanced the tool is-it’s the discipline you bring to how you prompt it. Write first. Use AI to diagnose. Own every interpretive claim you submit or speak. That sequence isn’t a compliance strategy; it’s the only configuration in which AI feedback builds thinking that holds under examination conditions, where the tool isn’t in the room. Each time you open a chatbot, the question worth asking isn’t “Can I use this?” It’s “Will I still be able to think this way when the screen is blank?” That’s the only question the exam will actually ask.