Slowly shaping Rephrazo by sitting inside people’s writing workflows
Over the last few weeks I’ve been doing something very unglamorous but weirdly satisfying:
Opening people’s drafts, watching them write, and asking,
“Okay, where exactly does this start to hurt?”
The product is Rephrazo – a small inline AI rewriter.
Highlight → hotkey → compact popup → one-click insert.
No ChatGPT tab, no copy–paste dance, just “fix this sentence and let me move on”.
That was the idea. The reality only стала вырисовываться после разговоров.
What actually changed after talking to users
A few things we thought were smart… weren’t.
-
We added multiple modes (formal / casual / punchy), sliders, knobs.
Users: “I don’t want to configure. I just want one reliable rewrite.”
→ We killed most of the options and doubled down on a single strong default. -
We assumed people wanted “stronger marketing copy”.
Users (copywriters, founders):“If it sounds like LinkedIn AI, I’d rather keep my messy original.”
→ We re-framed the core promise from “make it better” to
“make it clearer, don’t kill my voice.” -
We thought the magic was “AI”.
Users kept coming back to something more boring:“The best part is that I don’t have to break flow.”
→ We started obsessing over tiny UX things: where the popup appears,
how fast it shows up, how many clicks until you’re back in flow.
None из этого я бы не увидел, глядя только в код и аналитику.
How we’re collecting feedback right now
So far it’s been very manual and very human:
-
DM people on X who write a lot (copywriters, indie founders, newsletter folks).
-
Give them early access and ask them to use Rephrazo in their real workflow.
-
Then ask one stupidly simple question:
“Where did it annoy you or slow you down?”
The best insights пришли не из “feature requests”, а из маленьких фраз:
- “I regret pressing it if it takes more than 1–2 seconds.”
- “I use it mostly on first lines and CTAs, not on everything.”
- “I wish it would under-edit rather than over-edit.”
Every time I hear something like that, я иду не в “идеальный роадмап”,
а в самое узкое место UX и пытаюсь убрать ещё один грамм трения.
Tiny example of an iteration
One user shared screen while editing a landing page.
They highlighted a sentence, pressed the hotkey, saw the popup…
…and then stopped.
“I’m scared it’s going to overwrite everything if I misclick.”
That wasn’t a bug. It was just bad feeling.
We changed:
- added a very clear “Insert” vs “Dismiss” state,
- made sure original text is always easy to restore,
- toned down aggressiveness of rewrites.
Same model, совсем другой эмоциональный опыт.
Why I’m posting this on Indie Hackers
I’m still figuring out two things:
-
When do you know you’re listening “enough”?
At what point do you stop tweaking the core interaction and say:
“Окей, this is the gesture, now we just scale it”? -
How do you separate real signal from nice opinions?
Some people want more knobs, others want fewer.
Some ask for features that would slow down the core “stay in flow” promise.
If you’ve shipped a small focused tool before:
- How did you structure your early feedback loop?
- Was there a moment when you realised: “Now it’s time to stop reshaping the core and just polish + grow”?
If you’re curious to try the current version and tell me where it still gets in your way, you can play with it here:
A one-sentence “this part feels off” is honestly more valuable to me right now
than any generic “looks cool”.
