Analyse only coherence and cohesion, nothing else
Maps your controlling ideas, referencing chains and linker repeats — the criterion most learners cannot self-assess — with grammar and vocabulary explicitly off-limits.
Analyse ONLY coherence and cohesion in this essay: [MY ESSAY]
1. For each paragraph, state its single controlling idea in five words. If you
can't, say "no single controlling idea" — that's the finding.
2. Map the referencing chain: every pronoun and demonstrative ("this", "these",
"it") and what it points to. Flag any that are ambiguous.
3. List every linking device I used and count repeats. Flag mechanical over-use.
4. Identify where information order forces the reader to backtrack.
5. Show me the paragraph that would improve most from a different first sentence,
and give me three alternative topic sentences.
Say nothing about grammar or vocabulary.Best on Claude · ChatGPT · Gemini.
What to fill in
| Token | What to paste |
|---|---|
| [MY ESSAY] | Your full essay. This prompt needs the whole thing — paragraph order is what it is analysing. |
What you should get back
A five-word controlling idea per paragraph (or the admission that there is not one), a referencing map with ambiguous pronouns flagged, a linker frequency count, the points where a reader has to backtrack, and three candidate topic sentences.
What it actually returned
Tested on Gemini 2.5 Flash, API, August 2026, 2026-08-06. Trimmed — this is an excerpt of a longer response, not the whole thing.
1. Single Controlling Idea (5 words per paragraph): — Paragraph 1 (Introduction): Agree: public transport over roads. — Paragraph 2 (Body 1): Public transport benefits the environment. — Paragraph 3 (Body 2): New roads fail traffic solution. — Paragraph 4 (Conclusion): Public transport solves city problems. 2. Referencing Chain: Paragraph 2: — "this" → "pollution produced by cars" (implicitly, the act of producing pollution) — "their" → "cities like Copenhagen" Paragraph 3: — "more people" → (unspecified) "people" — "the road" → "a new road" — "This" → "when a new road is built, more people decide to drive, so the road becomes full again" — "they" → "roads"
Our verdict on this run
The referencing map is the strong part: it caught that "this" in body 1 points at something the sentence never actually names, and flagged "more people" as unspecified — both real cohesion faults a reader glides past. It also did exactly what the failure mode above predicts at point 1. Body 2 of the test essay carries two separate ideas (induced demand, and the cost and space roads consume), and rather than report "no single controlling idea", it smoothed them into "New roads fail traffic solution". That covered up the most useful finding available, which is why it is worth checking the five-word summaries yourself.
How this prompt goes wrong
It drifts back into grammar and vocabulary despite the instruction, usually from point 3 onward, and you end up with the same generic feedback the single-criterion focus was meant to avoid. The other failure is politeness at point 1: rather than say "no single controlling idea", the model invents one by summarising whatever the paragraph contains. That is the finding you most need, so check it yourself — if the five-word summary needs an "and" to work, the paragraph has two ideas and the model covered for you.
Tips
- Coherence is the criterion learners self-assess worst, because your own writing always feels logical to you.
- "No single controlling idea" is the most useful sentence this prompt can produce. Treat it as a finding, not a failure.
- Fix the backtracking points before you touch anything else — reordering information costs no words and moves the criterion fastest.
Related prompts
- Writing Task 2Test your introduction and conclusion against seven checksRuns a yes/no scorecard over the two most formulaic paragraphs in Task 2, quotes the evidence, then rewrites both in your own voice.
- Writing Task 2Force the model to quote your essay before it scores itMarks your Task 2 essay in four locked steps — evidence first, descriptors second, band last — so it cannot work backwards from a flattering number.
- Writing Task 2Make the model argue your essay is half a band worseAsks the model to build the strongest case against the band it just gave you, then arbitrate honestly between the two.
- Mock examinerFind out what an examiner notices in your first 50 wordsTests your essay the way it is actually read — fast — separating conspicuous errors from invisible ones and naming the feature that tips the band.
Common questions
Why exclude grammar and vocabulary?
Because a model asked about everything gives generic advice about everything. Restricting it to one criterion is what produces the referencing map and the backtracking analysis — output you will not get from a general "mark my essay" request.
What do I do with the referencing map?
Look only at the flagged ambiguities. Every "this" that could point at two things is a place the examiner has to re-read, and re-reading is exactly what the Coherence descriptors penalise.
Is a high linker count bad?
Repetition is. The descriptors penalise mechanical over-use, not linkers themselves — five "Moreover"s is a worse signal than none at all.