The criteria examiners rate your Writing and Speaking against are not secret. They are public PDFs, free, hosted by IELTS itself:
- Writing: ielts.org/cdn/Guides/ielts-writing-band-descriptors.pdf
- Speaking: ielts.org/cdn/ielts-guides/ielts-speaking-band-descriptors.pdf
Yet most prep advice is written as if these documents did not exist — vague tips about "using linking words" and "sounding natural" instead of the actual sentences an examiner matches your performance against. If you are chasing a specific band, the single highest-value reading you can do is the row for that band, in those PDFs, in the current wording. This page explains what the documents are, how to read one row properly, and the trap that catches people who quote them.
The failure
The failure is second-hand descriptors. Test takers read a blog's paraphrase of the criteria instead of the criteria, and end up optimising for phrases IELTS never wrote — or worse, phrases IELTS used to write. The famous example: "addresses all parts of the task". That is the pre-2023 Writing wording. The current PDF, marked "Updated May 2023. Please visit IELTS.org for updates", says something different — "The main parts of the prompt are addressed" — and the difference changes what the sentence asks of your essay. A guide still teaching "all parts" is teaching a retired document.
What the documents are
For Writing and Speaking there is no answer key. An examiner rates your performance on four criteria per skill, using a table where each band is a row of descriptions.
Writing — the official scoring page lists the criteria as:
"Task achievement (for task 1) and task response (for task 2), Coherence and cohesion, Lexical resource, Grammatical range and accuracy."
"The criteria are weighted equally and the score on the task is the average."
Speaking — the same page lists:
"Fluency and coherence, Lexical resource, Grammatical range and accuracy, Pronunciation"
"Each of the criteria carry equal weighting and the overall average gives the IELTS score for Speaking."
Eight criteria in total, four per skill, each worth a quarter of that skill's score. The Writing PDF also carries a header note worth knowing:
"A script must fully fit the positive features of the descriptor at a particular level. Bolded text indicates negative features that will limit a rating."
The version trap
Two official Speaking descriptor PDFs circulate, and they are worded differently. The current one, hosted on ielts.org's CDN, describes Pronunciation Band 6 as "Uses a range of phonological features, but control is variable" and says at Band 8 that "Accent has minimal effect on intelligibility". An older "public version", still live on a Cambridge server (assets.cambridgeenglish.org/webinars/ielts-speaking-band-descriptors.pdf), says "uses a range of pronunciation features with mixed control" and, at Band 8, "L1 accent has minimal effect on intelligibility".
Never mix wording from the two versions — half the descriptor quotes you meet online are stitched from both. Treat the ielts.org CDN PDF as the one to read; the differences are not substantive, but a quote should come from one document, whole.
How to read a band row: Writing Task Response, 6 vs 7
Here is Band 6 Task Response, current wording, taken apart phrase by phrase:
"The main parts of the prompt are addressed (though some may be more fully covered than others). An appropriate format is used. A position is presented that is directly relevant to the prompt, although the conclusions drawn may be unclear, unjustified or repetitive. Main ideas are relevant, but some may be insufficiently developed or may lack clarity, while some supporting arguments and evidence may be less relevant or inadequate."
- "The main parts of the prompt are addressed" — the prompt has parts. A question asking you to discuss both views and give your opinion has three. Band 6 requires that a reader can find each main part answered somewhere in the essay — even unevenly ("some may be more fully covered than others" is tolerated at 6).
- "A position is presented that is directly relevant" — you have an opinion and it matches the question asked, but the conclusions it leads to "may be unclear, unjustified or repetitive". A stated view with wobbly follow-through still fits Band 6.
- "some may be insufficiently developed" — ideas can be named without being built. At 6, an undeveloped but relevant idea is priced in.
Now Band 7:
"The main parts of the prompt are appropriately addressed. A clear and developed position is presented. Main ideas are extended and supported but there may be a tendency to over-generalise or there may be a lack of focus and precision in supporting ideas/material."
Read the deltas: "addressed" becomes "appropriately addressed" — the unevenness allowance is gone. "A position" becomes "a clear and developed position" — stating a view is no longer enough; the essay has to build it. "Main ideas are relevant" becomes "main ideas are extended and supported" — every main idea needs development, and what 7 still forgives is only over-generalising and imprecise support. The jump from 6 to 7 on this criterion is not better English — it is fuller answering: even coverage, a developed position, every idea extended.
The same reading works for Speaking Fluency and Coherence. Band 6:
"Able to keep going and demonstrates a willingness to produce long turns. Coherence may be lost at times as a result of hesitation, repetition and/or self-correction."
Band 7:
"Able to keep going and readily produce long turns without noticeable effort. Some hesitation, repetition and/or self-correction may occur… However, these will not affect coherence."
Both bands hesitate. The difference is what the hesitation costs: at 6 it sometimes breaks the thread of what you are saying; at 7 it does not, and long turns come "without noticeable effort". Hesitating is not the failure — losing the listener while you hesitate is.
What the descriptors are not
They describe performances — they do not generate one. A row tells an examiner what a Band 7 essay looks like once it exists; it is not a checklist you can tick your way through, and inserting descriptor vocabulary ("cohesive devices", "less common lexis") into an answer earns nothing. Note also that in both Speaking versions, Band 7 Pronunciation is defined only by reference to its neighbours — "all the positive features of Band 6, and some, but not all, of the positive features of Band 8" — so anyone selling you a standalone Band 7 pronunciation description wrote it themselves.
And there are no descriptors at all for Listening and Reading. Those papers are marked from answer keys — you get a raw score out of 40, converted to a band, and IELTS publishes indicative average-marks tables for those two skills only.
What actually costs you
- Preparing from a paraphrase — or the pre-2023 wording — instead of the current PDFs.
- Quoting descriptor lines stitched from the two Speaking versions.
- Treating a band row as a recipe rather than a description of a finished performance.
- Polishing one criterion while ignoring another — all four are weighted equally, and the score is the average.
- Never testing yourself against the rows: you cannot see "insufficiently developed" in your own essay without a marked attempt.
Try one
Read the two rows either side of your target band, then write something and hold it against them.
[Try a marked task →] — one full task, marked, no card required.