Self-assessment
IELTS Speaking Feedback Without a Tutor: How to Mark Yourself Honestly
Most advice about practising without a tutor assumes the problem is access. It usually isn't. The problem is that self-assessment has a blind spot built into it — you are using the same brain to produce the speech and to judge it. This guide is about working around that blind spot rather than pretending it doesn't exist.
The honest limits of marking your own speaking
The four IELTS Speaking criteria are not equally self-assessable. Two of them leave written evidence you can inspect at your own pace. Two of them depend on hearing yourself as a stranger would, which is precisely the thing you cannot do. Being clear about which is which saves you from months of confident, useless self-marking.
| Criterion | Can you judge it alone? | Why |
|---|---|---|
| Lexical Resource | Yes, reliably | Vocabulary range is visible in a transcript. You can literally count how many times you reused the same word. |
| Grammatical Range & Accuracy | Yes, mostly | Errors and structural variety are inspectable on the page. You'll miss errors you don't know are errors, but you'll catch most. |
| Fluency & Coherence | Only if you count | Your memory of an answer smooths over your own hesitations. Counting events works; "did that sound fluent?" does not. |
| Pronunciation | Rarely | You hear your own accent as neutral by definition. This is the criterion where outside input is worth the most. |
The practical consequence: spend your self-review time on the top three rows, and get outside input — human or automated — specifically for the bottom one. Splitting your attention evenly across all four wastes most of it.
Count events, don't judge impressions
This is the single idea that makes solo practice work. "That answer felt smoother" is not data — it tracks your mood and your familiarity with the topic more than your delivery. Three countable events, on the other hand, correlate directly with how a Fluency & Coherence band is awarded:
- Filler sounds — um, uh, er, ah. Count every one. Track them per minute rather than per answer, so answers of different lengths stay comparable.
- Silences longer than about 1.5 seconds. Brief pauses for thought are normal and unpenalised. It's the long ones, repeated, that signal your fluency isn't automatic yet.
- Restarts. Every time you abandon a sentence and begin again ("I went — I mean, I was going…"). This one surprises people most, because restarts feel like self-correction, which feels virtuous.
Three numbers, written down after each session. They move slowly, they don't care how you felt, and they're the closest thing to an objective examiner you have when you're working alone.
A sixty-second transcript beats a full one
Transcribing every practice answer is the advice everybody gives and almost nobody sustains past the first week. Transcribe one minute instead — the weakest minute of your session — and write it exactly as spoken, including every filler and false start. The recurring patterns show up in sixty seconds; you don't need fifteen minutes of evidence to find them.
Raw transcript — one minute, unedited
"So, um, the place I want to talk about is, uh, it's a café near my, near my university. I go there, um, quite often because it's — the atmosphere is very, uh, relaxing. And the coffee is good. I mean the coffee is, um, quite good."
Counted rather than judged, that minute reads: six fillers, one long silence, two restarts ("near my, near my"; "it's — the atmosphere"), and one abandoned intensifier ("good" downgraded to "quite good"). None of that is audible to the person who said it — they remember having described a café. All of it is obvious on the page.
Notice what the transcript also reveals about a criterion you can self-assess: "good" appears twice in three clauses, and the whole excerpt uses one tense. That's a Lexical Resource and a Grammatical Range observation, from the same sixty seconds.
Calibrating: is your self-score actually honest?
Self-assessment drifts in a predictable direction — people who are anxious about the test mark themselves down, and people who are comfortable in English mark themselves up. Two cheap calibration checks:
- Re-mark an old recording cold. Score an answer from a month ago without looking at what you gave it at the time. A gap of more than half a band means your marking isn't stable enough to trust week to week.
- Mark someone else first. Score a stranger's practice answer from anywhere online, then score your own in the same sitting. Most people are visibly stricter with the stranger — the size of that gap is roughly your personal bias.
Where solo assessment runs out
Three situations mean you've hit the ceiling of what you can do alone, and should get outside input:
- Your counted numbers stop moving for three straight weeks despite consistent practice.
- Your self-assessed band and an actual test result differ by more than half a band.
- You suspect Pronunciation is your limiting criterion — the one you fundamentally can't hear from inside.
"Outside input" doesn't have to mean a paid tutor. A language partner briefed to comment on one specific criterion, a study group that trades one honest note per answer, or automated scoring that reports the four criteria separately will all break the blind spot. What won't help is more of the same solo listening that got you here.
Practising this in BandLift
BandLift automates the counting and covers the criterion you can't self-assess. It runs a full timed 3-part mock, then reports every filler, every pause over 1.5 seconds, and every restart with timestamps — the three numbers above, measured rather than estimated. Pronunciation is scored from phoneme-level analysis, not self-judgment. And the Adaptive Improvement Model rewrites your own Part 2 answer a band higher, keeping your ideas, so the comparison is against your own language rather than a stranger's model answer.
It scores honestly by design. If your performance is a 6.0, it says 6.0 — which is the only version of feedback that's useful when you're marking yourself.
Download BandLift on the App StoreFrequently asked questions
Why does my practice feel good but my score never moves?
Because you're almost certainly assessing the wrong signal. Asking "did that sound okay?" measures your comfort with your own voice, not your delivery. Counting concrete events — fillers, pauses over 1.5 seconds, restarts — gives you a number that moves independently of how the session felt.
Which criteria can I genuinely assess on my own?
Lexical Resource and Grammatical Range & Accuracy, reliably — both leave written evidence in a transcript. Fluency & Coherence, but only if you count rather than judge. Pronunciation is the hardest, because you can't hear your own accent from outside your own head.
How much should I transcribe?
Sixty seconds. Full transcription of every answer is unsustainable and most people abandon it within a week. One carefully transcribed minute — every filler and restart included — surfaces the same recurring patterns for a fraction of the effort.
When should I stop self-assessing and get outside help?
When your counted numbers stall for three consecutive weeks despite consistent practice, or when your self-assessed band and an actual result differ by more than half a band. Both signal a blind spot you can't see from inside your own assessment.