ClipWise
AI reading and writing assistant for Windows and macOS
Localisation was the cheapest feature I ever shipped and the worst growth bet I have made so far. The app runs in 21 interface languages, two of them right to left. The translation files were the easy part. The real cost was Arabic and Persian: mirroring the whole layout properly took longer than the other nineteen languages combined, and I did it anyway because I work in a second language myself and I refuse to ship a tool built for people like me in English only.
Here is the uncomfortable result. I cannot point to sales that the languages produced on their own. The reason is obvious in hindsight. A translated interface is a product feature, but discovery still happens in one language. My listings, my posts and my search presence are all English. Someone searching in Indonesian or Turkish never finds the product, so they never learn that the interface would have greeted them in their own language. Localisation converts users you already reached. It does not reach anyone.
If I did it again I would still ship all 21, but I would pair every language I claim to support with at least one distribution channel in that language, even something as small as one directory listing or one forum where that language actually lives. Support without distribution is a museum exhibit.
For those of you who localised: did any single language ever pay for itself, and if it did, how did those users find you in the first place?
I work all day in English. It is not my first language. Everything about the product comes from that: the rewriting, the translation, the part where you can speak a thought instead of typing it clearly.
Building for yourself is supposed to be an advantage, and mostly it is. You do not need research to know which moments hurt. But there is a cost nobody mentions, which is that you lose the ability to tell the difference between a real problem and your own habits.
A concrete example. I was certain the killer feature was rewriting, because that is the thing I personally do twenty times a day before hitting send. I built the whole pitch around outgoing writing for months. It took a stranger to point out that half my own feature list is about understanding things coming in, summarizing, clarifying, translating something someone sent me, and that for a lot of people that half is the reason they install it at all. I had underweighted it because reading in English stopped hurting me years ago. It still hurts other people.
So the product was right and my model of the product was wrong, and no amount of using it myself would have surfaced that. Only somebody else's confusion did.
If you build for yourself: what have you found that you were blind to precisely because you had already adapted to it?
Two of those languages are right to left, which means Arabic and Persian users get a fully mirrored interface rather than English text shoved into a layout facing the wrong way. That work is not a translation file. It is layout, icon direction, text alignment and a long tail of small things that only break for people you will never watch using the product.
The case against doing it early is obvious. I had almost no revenue. Every hour spent on mirroring was an hour not spent on distribution, which is still my actual constraint. A rational allocator would have shipped English, waited for demand, and localized on evidence.
The case for it is that the product is built for people who work in a language that is not their first, and shipping that product English only would have been a contradiction sitting at the centre of the pitch. Also, retrofitting right to left into a mature interface is genuinely painful, while building it in early is merely annoying.
What I cannot claim is that it paid. I do not have the numbers to prove those languages brought users I would not otherwise have had, and I am suspicious of my own reasoning here, because the work was interesting and interesting work is very easy to justify after the fact.
So a real question rather than a rhetorical one: has anyone localized early and been able to show it mattered? Not a feeling, a number.
I moved my plan quotas out of the app and into a config the server serves at runtime. Sensible change: I can adjust the free daily allowance without shipping a release, without waiting for auto-update to reach everyone, without version skew between Windows and macOS.
Then I went to update a directory listing and noticed what I had actually done. Roughly a dozen published pages, store listings, comparison pages and affiliate blurbs all contain a sentence like "10 free actions a day". Every one of those is now a claim about a value that a single admin console change can falsify, quietly, with no deploy, no alert, and no way for me to know which pages just went stale.
Marketing copy is usually treated as static text. The moment a number in it is served by config rather than compiled into the binary, that copy has an undeclared dependency on a system nobody thinks of as customer-facing.
The fix was a writing rule rather than a code change. The plan is permanent, the quota is a current value. So copy now says "free plan with no time limit, currently 10 actions a day", never "free forever at 10 a day". Anywhere the digit is not commercially necessary it says "a free daily allowance" instead, and the website pulls the number from the same config at render time so the page cannot disagree with the product.
Worth checking if you have runtime flags: which of your published claims are quietly downstream of a value somebody can change on a Tuesday afternoon?
-
1
The strongest insight is that runtime config turns marketing copy into a live product dependency. A small quota change can make dozens of otherwise correct pages instantly misleading, so keeping the product and its published claims tied to the same source is a smart safeguard.
-
1
This is a sharp catch, and it's a broader pattern than plan quotas — the same thing happens with feature flags gating UI copy, or A/B test configs that quietly change what "New!" or "Limited time" means on a page nobody's looked at in months. The "plan is permanent, quota is a current value" distinction is a genuinely useful writing rule, not just a patch. Do you have any automated way to flag when marketing copy references a config value (so it doesn't rely on remembering to phrase things that way going forward), or is it purely a discipline/checklist thing for now?
-
1Honest answer: discipline plus a list, with only one part automated. The automated part is the website. The hero pulls the number from the same config the app reads, so that page physically cannot disagree with the product. If I could do that everywhere I would, and the problem would mostly disappear. The part that cannot be automated is everything I do not control. Store listings, payment provider pages, directory profiles, affiliate blurbs, other people's comparison articles. No script of mine can reach them. So instead of a linter I keep a plain registry: every published surface that names a digit, with a note on where it lives and who can edit it. When the value changes, that list is the blast radius, and updating it is a chore rather than an archaeology project. Your feature flag and A/B examples are worse than my case, because a quota at least changes deliberately. A flag that silently flips "New!" off on a page nobody has opened in months has no moment anyone would think to check. If I were you I would treat any copy string whose truth depends on a flag as needing an owner and an expiry date, the way you would treat a certificate. I am not aware of a tool that does this well. If someone built one I would pay for it, which is probably a hint.
-
On my comparison page, every competitor gets a sentence about where they genuinely beat me. Grammarly corrects live as you type, with a real grammar engine; I only work on demand, on text you select or copy. DeepL translates whole documents with glossaries and formality control; I translate selected text. Dedicated dictation tools stream into the field you are typing in; I record a note and transcribe it. All of that sits directly underneath my own pitch.
Two reasons I did it. The first is that I sell to people who are already suspicious of AI marketing, and the fastest way to be believed on the claims that are true is to be visibly honest about the ones that are not. The second is duller: in the EU a published comparison has to be material, relevant, verifiable and representative. Vague superiority claims are not just tacky, they are a liability. So every price I quote is dated and sourced, and I re-check before publishing.
What I do not know is whether it converts. It might read as confidence. It might also hand a hesitant visitor three good reasons to close the tab and go buy the incumbent instead. Those are very different outcomes and I cannot tell them apart from the traffic I have.
If you have run an honest comparison page against a conventional one, I would love the number. Otherwise I am about to find out the expensive way.
For most of this product's life my tagline said "AI Actions for Your Clipboard." It was accurate. It was also the most expensive sentence I have ever written.
If you call yourself a clipboard tool, the buyer's reference points are Ditto, Maccy, and the clipboard history built into Windows. All free, all good, all mature. Every pricing conversation then starts from zero and I have to argue my way up. The ceiling is not set by my features. It is set by the category I put myself in.
The same product, described as a reading and writing assistant, sits next to a grammar subscription and a dictation subscription. Two of those together run about $24 a month, before anyone adds an OCR tool. I start at $4.99. Same binary, same shortcuts, same code, and a completely different question in the buyer's head. Not "why would I pay for this?" but "why am I paying four times this?"
The uncomfortable part is that I had all the information a year ago. The clipboard is how the product works; it was never what the product is for. I let the mechanism name the category because the mechanism was the clever part, and clever is a bad reason to choose a name.
Has anyone else priced themselves into a corner with a tagline? I would like to know I am not the only one.
-
1
The idea of turning one piece of content into multiple short-form clips is useful, especially for people who know they should be posting consistently but don't have the time to edit everything manually. The biggest challenge is probably making sure the clips still feel intentional and not just like random cuts from a longer video. Curious to see how Clipwise handles that part as it develops.
-
1
Thanks for reading, and let me correct one thing, because the fact that I have to is useful information for me.
ClipWise has nothing to do with video. It is an AI reading and writing assistant for Windows and macOS. You select or copy text, press a shortcut, and it rewrites, translates, summarizes, clarifies or drafts a reply, right where you were working. Or you press the voice shortcut, speak, and it turns the recording into written text you can then rewrite or translate. The "clip" is clipboard, not video clip.
You are not the first person to read it the other way. In a post that spends four paragraphs describing a writing tool, the name still pulled you toward video editing, which tells me it works against me in any context where someone has not already seen the product. That is the real cost of a name I picked because it was clever, and I would rather learn it now than after another year of listings.
So thank you for the accidental user test. It was more useful than the compliment would have been.
-
-
1
The category point is unusually sharp. The product didn’t change, but changing the reference point changed how the same $4.99 price gets interpreted.
-
1
That is the sentence I wish I had written a year earlier.
The part that took me longest to accept is that the price ceiling is not set by what the product does. It is set by whatever the buyer is mentally holding it next to. I spent months trying to justify $4.99 against free clipboard managers by adding features, which was the wrong lever entirely. Nothing about the binary changed when I stopped. The comparison set changed.
The uncomfortable follow-on is that it is reversible. If my own copy drifts back toward clipboard language, the anchor drifts back with it, and the drift is easy because the clipboard is genuinely how the thing works. So I now treat the category sentence as something to defend in every listing rather than something I decided once.
-
1
That’s a strong observation. The fact that the anchor can drift back so easily makes the positioning lesson even more interesting — would you be open to sharing the best email to reach you on?
-
-
Understand what you read. Say what you mean.
I read and write in English all day, and it is not my first language.
That creates two kinds of friction, and they pull in opposite directions. Incoming text takes longer to understand, especially when it is dense, unfamiliar, or full of jargon. Outgoing messages take longer because I check the grammar, the phrasing, and the tone three times before I feel safe hitting send. And sometimes I know exactly what I want to say, but speaking the thought is far easier than typing it clearly.
So I did what most people in that situation do: I opened more tabs. A grammar checker for phrasing. A translation tab for anything foreign. A chatbot for the longer stuff. An OCR tool for text stuck inside a screenshot. A speech-to-text app for when typing was slower. A clipboard history to recover what I copied five minutes ago.
Each one solved a real part of the problem. Together they were still constant switching, prompting, copying, and pasting, plus several subscriptions, several shortcuts, and half of them locked to one browser, one ecosystem, or one platform.
At some point the pattern got obvious. All that switching was in service of just two jobs: understanding what came in, and saying clearly what went out.
So I built ClipWise, an AI reading and writing assistant for Windows and macOS that works across almost any app, with two ways to begin.
Start with text that already exists. Select or copy it, press a shortcut, and rewrite awkward phrasing, translate across 21 languages, summarize something long, clarify dense writing, structure rough notes, extract action items, or draft a reply from copied context. The result comes back where you were already working.
Or start with a spoken thought. Press the voice shortcut, speak, and ClipWise turns the recording into written text you can copy, rewrite, or translate right there. Not real-time dictation into the cursor, but a short note that becomes a finished message.
When text is trapped in a screenshot, a slide, or an unselectable PDF, it can pull the text out so you can carry on from there.
The honest part, which I think matters more than a bigger claim: ClipWise does not try to be the deepest specialist in any single category. Grammarly's live inline underlines are a real advantage I do not match. DeepL translates whole documents with glossaries; I translate selected text. Dedicated dictation tools stream straight into the active field. ClipWise is for the person who wants to retire the whole side-by-side stack and get the everyday version of those jobs done in one place, on both platforms, where the work is already happening.
What made the business case click for me was the price anchor. Framed as a clipboard tool, I would be competing with free. Framed as a reading, writing and voice tool, the comparison is a grammar subscription plus a dictation subscription, together about $24 a month, before any OCR utility. ClipWise starts at $4.99, and the free plan has no time limit: currently 10 AI actions a day, no API key required. You can also bring your own OpenAI or Anthropic key, or run a compatible local Ollama model if you would rather nothing left your machine.
The goal for 2026 is 10,000 copies.
The thing I keep going back and forth on is how much honesty to put in the marketing. My comparison page names what competitors do better than me, in the same breath as the pitch. It feels right, and early readers say it builds trust, but I have no idea yet whether it converts. Has anyone here tested a genuinely self-critical comparison page against a conventional one?
-
1
“AI for clipboard is a great productivity angle.
Would be interesting to see which use cases users actually stick with long-term vs just try once — that retention insight can be gold.”-
1
Honestly, I don't have long-term retention data worth quoting yet. The app is young, and a small sample dressed up as a trend would mislead both of us.
My expectation, which is exactly what I want to test: the two halves of the product should behave very differently. Rewrite and translate have an external trigger (someone sends you something, or you owe someone a reply), so those should be daily. OCR and summarize are situational; you use them in bursts when the work calls for it, then not for a week. If that holds, retention lives in the everyday actions, and the situational ones are mostly what get people to install in the first place.
So the measurement I'm setting up is: for users past day 30, which action was their first, and which one are they still running in week four. If those are usually different actions, then the thing that sells the app and the thing that keeps it are not the same feature, which changes both the marketing and the onboarding.
Thanks for the nudge, it's the right question to be asking this early.
-
-
1
The line about “retiring the whole side-by-side stack” was what sold the idea for me. It explains the product much better than listing all 13 actions.
I’m curious have you found users buy into the convenience first, or do they usually come in looking for one specific feature and discover the rest later?
-
1
That's useful to hear, thank you. I've gone back and forth on whether to lead with the story or the spec sheet, and "thirteen actions" clearly loses.
On your question: I don't have clean data to answer it properly yet, so treat this as an impression rather than a finding. People seem to arrive for one specific thing, usually whatever they were already searching for: text stuck in a screenshot, or fixing a message before they send it. The breadth only registers later, once they hit a second job and find it is already there. Nobody goes looking for "an app that does five things"; that is something you discover, not something you type into a search box.
If that is right, it has an awkward implication. The consolidation story is what makes someone stay and tell a friend, but a single specific job is what makes them click in the first place. So the honest move is probably to lead each channel with one concrete job and let the rest be a pleasant surprise, rather than opening with the full list, which is exactly the mistake I made in the first version of this post.
Did you come at it from one of those jobs yourself, or from the stack problem?
-
1 Comment
The lack of attributable sales seems to show that localization alone isn’t a growth channel, but not necessarily that the languages are weak markets. Did users arriving through any non-English path behave differently once they reached the product, or is that evidence still missing?