Skip to content
SpeakingintermediateBands 6.0–8.0

See the same cue card answered at Band 6, 7 and 8

Produces three spoken-register answers to one cue card at ascending bands, then names the concrete moves that separate each level from the next.

The prompt
Cue card: [CARD TEXT]

Produce three answers to the same card, in spoken register (contractions,
false starts allowed, natural fillers, 200-230 words each):
- A Band 6 answer.
- A Band 7 answer.
- A Band 8+ answer.

Then a comparison table:
| Feature | Band 6 version | Band 7 version | Band 8 version |
covering: idea development, tense range, discourse markers, lexical precision,
self-correction, and how the answer handles the final "explain why" bullet.

Then: the three specific upgrades from 6→7, and the three from 7→8. Be concrete —
name structures and moves, not qualities like "more fluent".

Best on Claude · ChatGPT · Gemini.

What to fill in

TokenWhat to paste
[CARD TEXT]The full cue card, all four bullets included — the last one is what the comparison hinges on.

What you should get back

Three answers to the same card in genuine spoken register, a six-row feature comparison table, and two sets of three concrete upgrade moves.

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.

Band 6 Answer

"Okay, so I want to talk about a book I read recently. Uhm, it's called Project Hail Mary.

What it was about: It's a science fiction book, by Andy Weir. It's about an astronaut who wakes up on a spaceship, but he doesn't remember anything at all – like, who he is or what his mission is. He slowly starts to remember things, and he realizes he's on a very important mission to save Earth because the sun is getting dim. There's a big problem. And then, surprisingly, he meets an alien, an intelligent alien! They work together to try and fix the problem for both their planets. It was quite exciting.

When and where: I read it, uh, maybe two months ago. I read it mostly at home, in the evenings, before I went to sleep, you know, just winding down…"

Our verdict on this run

The spoken register came through better than expected — "Uhm", "like", "you know", and a genuinely conversational rhythm rather than the tidy paragraph this prompt usually returns at Band 6. The weakness is structural: it labelled the sections with the cue card's own bullets ("What it was about:", "When and where:"), which is a written convention nobody uses aloud and which would look mechanical if you copied the shape. The band ladder itself was clear, and the upgrade list at the end named concrete moves — extending one idea rather than adding a third, and precision over variety — rather than vague qualities.

How this prompt goes wrong

The Band 6 answer is written, not spoken. Models produce a tidy paragraph with no false starts, no contractions and no fillers, which makes all three versions sound like essays and hides the thing you are trying to see. Read the Band 6 version aloud: if it sounds like something you would write rather than say, ask again with "the Band 6 answer must contain at least two self-corrections and three contractions". The second risk is yours, not the model's — memorising the Band 8 version. Examiners are trained to spot memorised language and it is penalised.

Tips

  • Read the Band 8 version aloud once, then answer the card yourself without looking. Never memorise it.
  • The 6→7 upgrades are usually about developing one idea further; the 7→8 upgrades are usually about precision. That pattern holds across most cards.
  • Use a real card from our library so the follow-up questions you practise next actually match.

Related prompts

Common questions

Will memorising the Band 8 answer help?

No — it will hurt. Examiners are trained to identify memorised responses, and a rehearsed answer that does not quite fit the card you are given is conspicuous. Learn the moves, not the words.

Why does the Band 6 answer sound so polished?

Because models default to written register. That is this prompt's main failure mode; force fillers and self-corrections explicitly, or the ladder between the three bands becomes invisible.

How long should a Part 2 answer be?

You speak for up to two minutes, which is roughly 200–280 words at natural pace. Running short is a far more common problem than running long — the examiner will stop you.

Continue learning

    Need help?