Key takeaways
- Business school AI policies vary far more than students assume, and most turn on disclosure rather than an outright ban.
- Frameworks like SWOT and PESTLE exist to make analysis repeatable — applying them correctly produces exactly the predictability detectors flag.
- The real risk in business coursework is not phrasing but fabricated figures and citations, which models produce in a confident register.
- Group projects create genuine authorship ambiguity; agree an edit owner and keep per-contributor drafts.
- A detector score is not a mark and is not evidence of misconduct. Check what your institution says about using one as sole evidence.
What business school policies actually require
Students tend to assume there is a single rule. There is not. Business schools have landed in noticeably different places, and the differences are usually about disclosure rather than permission.
Broadly, module handbooks tend to fall into one of a few positions: AI prohibited for this assessment; AI permitted for specified stages such as brainstorming or proofreading; AI permitted with a declaration of what was used and how; or no stated position at all, which is the most dangerous case because it leaves the judgement with whoever marks your work.
The disclosure rule is usually stricter than the usage rule
Most students who get into trouble were allowed to use AI and did not declare it. Read the assessment brief rather than the general university policy — module-level rules override the institutional default, and they are where the specifics live.
Policies also change between intakes. A rule you learned in first year may not be the rule for this module. Check the brief each time; it takes a minute and it is the single cheapest thing on this page.
Why business writing gets flagged
AI detectors score two properties: perplexity, how surprising each word is given what came before, and burstiness, how much sentence structure varies. Low on both reads as machine-generated.
Business education is, quite deliberately, a training programme in writing that is low on both.
| What you were taught | Why | Effect on a detector score |
|---|---|---|
| Apply SWOT, PESTLE, Porter's Five Forces | Frameworks make analysis repeatable and comparable | Repeatable analysis produces repeatable sentences |
| Follow the standard report skeleton | Markers and executives can navigate it quickly | Predictable section-by-section phrasing |
| Write concise executive summaries | Senior readers will not read past page one | Plain, efficient prose has low variation |
| Hedge conclusions and cite sources | Recommendations must be defensible | Cautious phrasing typical of model output |
| Use consistent professional register | Business communication convention | Removes the idiosyncrasy that reads as human |
A case analysis is the clearest example. Situation, problem, options, evaluation, recommendation — you were given that shape because it works. Filling it in competently produces text that is statistically unsurprising from the first line to the last.
Group projects and mixed authorship
Group submissions create a problem that individual assignments do not. Four people write four sections in four styles, someone merges them, and the combined document has visible seams. A marker may read that as inconsistency; a detector may score the smoothed-over result as machine-assembled.
There is also a genuine authorship question hiding here. If one member used AI and did not tell the others, everyone's name is on the submission. That is worth an explicit conversation at the start rather than a discovery at the end.
What actually helps:
- Draft in one shared document with revision history, rather than emailing files around. The record then exists by default and shows who wrote what.
- Agree one person owns the final edit pass, so the document has a consistent voice on purpose rather than by accident.
- Agree the group's AI position at the first meeting, and make sure it matches the assessment brief.
- Keep each contributor's own drafts until the mark is confirmed.
Where AI genuinely fails in business coursework
The detector conversation distracts from a larger problem. In business subjects, the serious failure mode is not clumsy phrasing — it is confident, specific, invented fact.
- Fabricated figures. Market sizes, growth rates, margins and market shares are produced in exactly the same authoritative tone as accurate ones. A model has no way to signal that it is guessing, and a plausible-looking number is worse than an obviously wrong one because nobody checks it.
- Invented citations. References to consultancy reports, journal articles and industry surveys that do not exist, often with real publisher names and plausible years attached. This is the fastest way to turn a phrasing query into an integrity finding.
- Stale company facts. Leadership, ownership, product lines and strategy change. Any company-specific claim needs checking against a current source, not against what a model recalls.
- Financial reasoning that looks right. Models are noticeably weaker at multi-step quantitative reasoning than their fluency suggests. A well-formatted NPV or ratio analysis can be internally inconsistent while reading perfectly.
The practical rule
Every figure, date, company fact and reference gets checked against a source you personally opened. If you cannot source it, cut it or reframe it as a general statement. This is worth more than any amount of rewriting.
If your work is flagged
- Preserve the revision history. Export or screenshot it before copying the text into a new document. Incremental drafting over days is difficult to fabricate and is the strongest evidence available to you.
- Assemble your working material. Notes, the case pack, spreadsheets, the sources you cited, group chat where you divided the work. For quantitative assignments your workings are especially persuasive — a spreadsheet with your own errors and corrections in it is hard to fake.
- Offer to discuss the analysis. Volunteer to explain why you chose that framework, what the numbers imply, and why you rejected the alternatives. Someone who did the analysis can do this straightforwardly.
- Ask what the concern rests on. Request the tool, the score, and whether anything beyond the score is relied on. A detector percentage is not a mark and most institutions state it is not sufficient evidence of misconduct on its own.
- Check the brief you were actually given. If the assessment permitted AI for certain stages and you followed that, say so and show your declaration. This is where students who disclosed properly are in a completely different position from those who did not.
- The full false-positives guide — what detector scores measure and the complete evidence checklist
A defensible workflow
The aim is a process you could describe openly to a marker without it counting against you. If you would not want to explain a step, that is the step to change.
- Read the brief for the AI rule first. Before you start, not after. Note whether disclosure is required and in what form.
- Do the analysis yourself. Framework selection, the numbers, and the recommendation are the assessed thinking. This is the part that cannot be delegated without it becoming someone else's work.
- Draft in a tool with revision history. Google Docs, Word or Notion. The record then exists whether or not you ever need it.
- Verify every checkable claim. Figures, dates, company facts, references — against sources you opened. Nothing goes in unsourced.
- Edit for structure and specificity. Cut padding, vary sentence length, and add the concrete detail that only your analysis produced. Our editing guide covers the full pass in order.
- Declare what you used, if required. A short, accurate statement. Under-declaring is the common failure; over-declaring costs nothing.
Checking a section's detector score before submission is reasonable as a rough signal that the prose is reading generically. Treat it as a prompt to add specificity, not as a number to optimise.
Paste a section you wrote to see rewrite options. English only; the model is optimized for English.
Sources
- Google Search's guidance about AI-generated content — Google Search Central, 2023
- Liang et al., 'GPT detectors are biased against non-native English writers' — Patterns (Cell Press), 2023
- Understanding false positives in AI writing detection — Turnitin