Hi, I’m Harold, founder of Toeflair. Between 2019 and 2022 I took the TOEFL six times, going from 98 to 106 — and this isn’t an essay about how to prep for the test. It’s about how that whole stretch eventually pushed me into building a TOEFL prep product of my own.
Three years, six attempts, back to square one
Let me show you what those six attempts looked like: 98, 99, 94, 96, 98, 106. Dropping from 99 to 94 on the third one really threw me. My brain immediately went looking for excuses — bad sleep, weird question set, the test was just harder that day, the universe had it in for me. I cycled through every single one of them.
Then on the fifth attempt I came back with another 98. The exact same score I’d started with. That’s when I had to admit it wasn’t a luck thing. I’d been practicing wrong from the start.
The section breakdown was the real giveaway. For the first five attempts, Listening hovered between 22 and 23. Never broke 24. That single section was what kept my total stuck under 100. It wasn’t until the sixth attempt, when Listening jumped all the way to 27, that the total finally cleared 100. Speaking sat at 22-24 in the early years, nothing dramatic. Writing was the rollercoaster — I got 28 on the 99 attempt and figured I had it figured out. Next time it dropped to 25. Then to 22, while the other sections were going up. Just writing alone, free-falling.
Reading and Listening are input sections. The logic’s pretty simple: if you understood the passage, you got it right. Writing and Speaking don’t work that way.
After the 22 in writing I did something kind of demoralizing — I went back through all my old writing practice trying to figure out where things had gone wrong. The more I looked, the more my mind went blank. I could see these were essays that hadn’t gotten full marks, but I couldn’t say what was actually wrong with any of them. I tried to remember what writing them had felt like in the moment, and that just made things worse. I remembered writing smoothly, no sentence that made me hesitate. But what had cost me points? I had no idea.
I sometimes joke with friends — if I could’ve seen what was wrong, I wouldn’t have written it that way to begin with.
I was studying entirely on my own back then. ChatGPT and Gemini didn’t exist yet — this was before 2023. The best I could do was scan sample essays online, eyeball the differences, and revise by feel. But “by feel” was the actual problem. I had no idea what I was doing well, what I wasn’t, or what I should specifically be practicing against next time.
New method, sixth attempt, finally past 100
The 8-point jump from 98 to 106 looked different in each section. But underneath, it was the same thing.
For Listening, I quit the “more reps + 1.5x speed to season my ears” approach completely and switched to intensive listening. Same passage, played sentence by sentence, until I could hear every contraction and reduction. My daily volume halved. And the score finally started moving.
For Speaking, I started building my own personal language corpus. Sounds nerdy, but it actually worked. I’d take passages from listening practice, write down everything I could remember, then time myself against the old format’s 15-20 second prep window and rewrite the spoken version until it became muscle memory. By exam time I wasn’t composing live. I was retrieving a structure I’d already rehearsed dozens of times. The new TOEFL Speaking has no prep time anymore, but I think the corpus + rewrite loop still matters; the response window just gets shorter.
Writing was the hardest. I eventually paid for a single-section writing course at 托你的福 — about 10,000 RMB. The mechanics were nothing fancy:
- Send my essays to the instructor as a Word doc
- They’d return notes and proposed rewrites
- I’d revise once and drill the same prompt type again
Repeat. But here’s the thing — this loop was the one I couldn’t run on my own. I literally couldn’t see what was wrong with my own writing, so I couldn’t self-correct. Every batch of feedback was another inch of my essay scaffolding being chiseled out. Eventually writing settled at 27-28.
Looking back, the core of those 8 points wasn’t more practice. It was finally having someone who could tell me what was wrong, where, and what to do next time.
What self-study actually feels like
The hardest part of self-studying for the TOEFL isn’t running out of practice questions. There’s always more practice. The hard part is that output sections don’t have native feedback — you finish, and you have no idea how you did.
The tools I actually used were a short list:
- 小站托福 — free basic practice; real exam questions and detailed feedback both sit behind their paid courses
- 考满分 — free basic practice + free auto-scoring for writing and speaking; same deal — real exams and meaningful feedback are paid
- A single-section paid instructor at 托你的福 — about 10,000 RMB per section
That free auto-scoring back in 2022 was basically a paragraph telling you you’d written well and could polish further from this angle — vague to the point of being useless for shaping the next attempt. I get the business model: free tools as top-of-funnel, then upsell into a 50,000 RMB full-prep package or a “guaranteed-pass” track.
I’m not going to trash paid courses. I’m a literal beneficiary. But here’s the thing: to get a single piece of feedback that actually improved my writing, I had to:
- Pay 10,000 RMB
- Work around the instructor’s calendar
- Wait three days for the next round
That bar locks most self-study takers out of the loop entirely. Honestly, only people with the money, the connections, and the luck of finding a good instructor get to run it at all.
Why Toeflair
Starting in late 2025, the base capabilities of frontier models — GPT, Gemini, Claude — kept stepping up, to the point where even speech recognition models can now correct pronunciation at the phoneme level.
That means the loop I once paid 10,000 RMB and weeks of calendar time to access can — for the first time — be fully automated. Not just essay scoring; pronunciation correction down to the phoneme level too.
My reasons for building Toeflair aren’t complicated. It isn’t that I think AI is going to give better feedback than a great human instructor; on certain fine-grained things, a good human is still sharper. It’s that feedback shouldn’t have to cost what it cost me, or take what it took me. A 90-point TOEFL taker trying to figure out where they’re stuck shouldn’t have to drop 10,000 RMB just to start the loop.
The other reason — honestly the bigger one — is that I just walked this exact path. Every place I got stuck, every “I can’t even tell what’s wrong with my own essay” moment, every time I looked at my score and went “wait, what?” — I lived through. I’m not inferring from a market report that self-study takers have a feedback gap. I was the gap.
So this product is for people who are where I was: already taken the test once, stuck somewhere between 90 and 105, studying mostly on their own, not quite sure what to fix next. It’s not for:
- People aiming for 110+ — I never hit that mark, and I’m not the right person to teach it
- People planning to drop tens of thousands on a guaranteed-pass course — I can’t make that promise, and I don’t think it’s the right shape for the underlying problem
If you’re where I was, this is for you.