Write it up
Seven sections written at different times over several months, in slightly different voices, none of which have ever been read one after another. And a dataset you downloaded a while ago, whose version you had better have written down at the time.
- One report that reads as though one person wrote it in one go, under the limit, fully referenced, and ready to hand over.
No new marks are available here. Several are still available to lose.
Going over the limit, burying evidence in an appendix nobody reads, or referencing so loosely that a source cannot be traced. All three are avoidable in an afternoon, and all three cost more than they should.
You have been carrying a figure on every page since step 1. Here they are together. Three of the seven are ours, the research question, the data section and the analysis, and follow from what each criterion's evidence actually is. The other four follow published guidance for this format, adopted whole rather than cherry-picked.
| Section | Crit | Marks | Words |
|---|---|---|---|
| Research question | A | 50 | |
| Environmental issue and background | A | 4 | 500 |
| Strategy, perspectives and tensions | B | 4 | 500 |
| Method | C | 4 | 500 |
| Treatment of data, prose only | D | 6 | 200 |
| Analysis and conclusion | E | 6 | 550 |
| Evaluation | F | 6 | 600 |
| Total | 30 | 2,900 |
Note Criterion D (Treatment of data): six marks for 200 words, while everything else costs between 92 and 138 words per mark. That is not an oversight. It is because the tables and graphs that earn those six marks are not counted at all.
Four things belong at the top of the first page: the title, your personal code, the codes of anyone you collaborated with on methodology, and the word count. There is no requirement for a cover page or a contents page.
Structure is your decision. The IB offers no template and asks only that the report be clear, concise and focused. Both prose and numbered steps are acceptable for the method.
And those six are the whole of what the IB leaves out of the 3,000. Everything else you type counts, including the numbered steps of a method. The list says nothing either way about figure captions, so keep them to the line or two the guide models rather than hiding analysis in them.
Everything below is how we suggest you actually do it.
You get one draft read, and it will say "develop this"
Your teacher may comment on one full draft: general remarks on strengths and weaknesses, annotations, advice. What they may not do is correct it. The improving is yours.
This is the whole reason to aim at 2,900 rather than 3,000. Feedback on an ESS investigation almost always asks for something to be developed rather than trimmed, and at the ceiling the only way to act on it is to delete something good. A hundred words in hand is the difference between improving the report and rearranging it.
Hand over something complete rather than something polished. A finished-but-rough evaluation gets you useful feedback; a beautiful background and three missing sections gets you comments on the background. You only get one go at this, so spend it on the parts you are least sure about.
Using AI, and saying so
The IB does not ban AI tools, and pretending otherwise helps nobody. What it asks is narrow, and worth knowing exactly rather than by rumour: the work has to be yours, and anything you take from a tool is cited like any other source.
The test being applied is whether you used AI to help you learn, or to appear to have done something you did not do. On that test a tool can explain a statistical test you have not met before, suggest search terms, or argue with a plan you have already written. It cannot choose your question, decide what your data means, or write your evaluation, those are the things actually being marked, and handing them over is handing over the marks with them.
Authenticity. The work is your own. A tool is not an author, and it cannot be listed as one.
Transparency. If you copy, paraphrase or modify anything a tool produced, text, an image, a translation, a chunk of analysis, name the tool in the body of the work at the point you used it, and add it to your bibliography. The guidance sits in an appendix to the IB's Academic integrity policy, updated in February 2026. Your school will have its own position on top of that; where the two differ, follow the stricter one.
Ask a chatbot for sources and it will sometimes invent them, complete with plausible authors, journal titles and page numbers. Criterion A (Research question and inquiry) wants citations detailed enough to be traceable, so a fabricated reference is a lost mark and an academic integrity finding at the same time.
The rule that prevents it: every source in your bibliography is one you have actually opened. If you cannot find it by searching its title, it may not exist.
Before you sign it off
You will confirm this is your own work, and authenticity can be checked against your proposal, your draft, your references and your usual writing. Where you worked with others on methodology, the writing must still be entirely yours. And the same piece of work cannot be submitted for both this and your extended essay.
Eight steps, one environmental issue, and a report that argues something about a real place.
The part worth keeping, long after the marks come back, is the bit in the middle: that a decision about a hillside somewhere involves people who want incompatible things, and that measuring the ground carefully can inform that argument without settling it. That is the whole subject, in 2,900 words.
Referencing a dataset
The IB does not prescribe a style; your school does. At the International School of Geneva - Campus des Nations it is Chicago footnotes, used across every IA and the extended essay, so you learn one system and reuse it everywhere.
Whatever style you use, the requirement is traceability. On this route that word does more work than usual, because your evidence is a file that can be revised, moved or replaced. A reference that gets somebody to the publisher is not enough; it has to get them to the version you used.
If the download offers a DOI, use it instead of the URL. It is a permanent address for one specific version, and it survives the publisher reorganising their website, which a URL does not.
Note the second card. The data and the document explaining how the data was made are two sources, and citing both is what shows a reader you went and read the methodology rather than only pressing download.
The last read-through
Read the whole thing once, in order, out loud if you can bear it. Seven sections written over several months do not automatically agree with each other, and the disagreements are invisible while you are inside any one of them.
On this route there is one specific thing to hunt for: a claim that is bigger than your rows. It creeps in through ordinary words. A study of thirty monitoring stations becomes a study of a river; a study of a river becomes a study of a lake; a study of one indicator becomes a study of water quality.
An early draft of the Geneva study called itself an investigation of Lake Geneva. Every monitoring station in it was on a tributary, and the one station actually on the lake had been dropped because it read zero for nearly thirty years.
The word had reached the title, the research question and the conclusion before anybody noticed, and each place it appeared was a claim the data could not support. Nobody had lied. One noun had been doing work it was never entitled to do.
Search your finished report for the name of the biggest place you mention, then check every hit against the rows you actually analysed. If your data is thirty stations on six streams, that is what your title, your question and your conclusion are allowed to say.
It can respond to prose you have already written: where a paragraph is unclear, where you have said the same thing twice, whether a sentence means what you intended.
It cannot write or rewrite it for you. And whatever you did use it for gets named at the point you used it and listed in your bibliography, which is what the section above is about.
Ready to submit
0 of 11The last checklist in the process. Once you submit the final version it cannot be retracted, so it is worth going through slowly.
The Geneva rivers investigation used on the secondary-data route of this guide is the author’s own analysis of a published cantonal dataset. Every figure quoted comes from the Canton of Geneva’s open data, downloaded on 10 September 2026; the duplication, the missing year and the varying campaign counts described in the guide are as found on that date and may since have been corrected. The selection rules, the analysis and the conclusions are the author’s and not the canton’s.