570 visitors, 17 clicks, zero downloads: what a too-small OpenAI ad taught me about growing VoxStudio

A solo founder working at a bright desk by a sunny window, with soft city towers outside

The day my OpenAI ad ran, I watched the Cloudflare dashboard. By the end of the day the site had about twice its usual traffic, 570 unique visitors. There were 17 clicks. The download count didn't move. Neither did subscriptions.

I'm David, and I build VoxStudio by myself in Singapore. This is the story of how that day changed the mix of how I grow it: more organic work now, and paid ads again later, with a real budget.

Why a solo founder was buying ads at all

The idea behind VoxStudio is simple. Recordings pile up. Interviews, meetings, lectures, voice notes. Everything useful in them was said out loud once, and finding it again means scrubbing through audio. I wanted a tool where you could ask your recordings a question and get an answer that points back to the exact line, with the speaker and the timestamp, so you can check it yourself.

That idea turned into three apps. Each one exists because of a specific problem.

Three devices side by side, a browser, a laptop and a phone, each showing an audio waveform and transcript lines

The web app is the easy door. You open a browser, upload a file or paste a video link, and get a transcript with speakers and timestamps beside the player, plus a summary and a translation if you want them.

The Mac app is where most of my time went, because a lot of recordings are private. Once you download the local models, transcription, speaker labels and caption timing run on the Mac and the files stay there. You can record a meeting from the Mac without a bot joining the call. And because a transcript is often the start of something else, the Mac app also became a video editor that treats subtitles as cut points, so you can trim on words and add a voiceover.

The iPhone app came last, for recording away from a desk. Phones run out of space and lose signal, so it uploads audio in small pieces while you record and deletes each piece once it's safely uploaded. If the connection drops, the recording keeps going and catches up later.

I can build all of that. What I didn't know was how to get it in front of the people who need it. Traffic had been low for a long time. So I tried to pay for attention.

The ad that worked, technically

People now ask ChatGPT which tool to use, so putting VoxStudio inside AI answers felt like the obvious bet. I set up OpenAI Ads with CPM bidding and bid USD 40 per 1,000 impressions.

When it finished, the numbers looked like this:

WhatResult
BiddingCPM, USD 40 per 1,000 impressions
Impressions5.3k
SpendUSD 41
Unique visitors that day (Cloudflare, whole site)570, roughly double
Clicks17
Downloads or subscriptions0
Bar chart of the OpenAI Ads test: 5.3k impressions, 570 unique visitors, 17 clicks, 0 downloads or subscriptions

In a narrow sense the ad worked. It was shown, people came, the graph went up. My guess is that almost nobody came because they had a recording to deal with that evening. They clicked because something showed up while they were asking a chatbot about something else. The spend was only USD 41, and the whole test sat on a budget of about USD 100. That was too low. The campaign also didn't meet OpenAI's requirements for expanding reach and scaling up, including the certification and scale-up steps they expect before a campaign can grow. With a limited budget, I never got past the small test.

I didn't conclude that ads were useless. I concluded that this test was too small, aimed at the wrong kind of moment, and never got a chance to expand. So I paused paid ads and tried somewhere people go when they're already looking.

The card that wouldn't go through

That meant Google Ads. Search ads at least catch people mid-question. I never found out whether they would work for me, because Google wouldn't accept my card. I tried a second card. Same result. The payment never went through and no campaign ever ran.

I could have opened tickets and spent days on it. I decided not to. One small OpenAI test and one Google attempt that never started were enough to make me ask a different question. If the people I want are typing questions into Google and asking ChatGPT, why am I only trying to rent a spot next to the answer instead of also writing the answer?

Perplexity is still on my list, both its ads and its partnerships. It's in planning. I'll write about it when there's something real to report.

Writing the answer while paid ads wait

Two funnels: one with lots of dots poured in and almost nothing coming out, one fed slowly by pages and books with steady drops falling onto a small plant

So for a stretch after that, most of my marketing time went into organic traffic. I kept improving the blog and the tutorials on voxstudio.me and worked on SEO one page at a time. It's slower than flipping on a campaign. But someone who searches "transcribe audio offline on Mac" already has the problem. If my page answers it properly, trying the app is just their next step.

Writing these pages taught me more about marketing than that first ad did.

A person at a laptop writing a long article, with printed pages, a notebook and a small microphone on the desk

The first thing I learned was that "trust me" doesn't work. When we wrote the tested guide to offline transcription on Mac, a claim like "works offline" wasn't enough. The guide names the Mac and the OS version we tested on. It explains how to confirm where processing and storage actually happen. It links the raw results so readers can see the mistakes as well as the good parts. It also says plainly which steps still need internet. A page that admits its limits is easier to believe, and that's the only kind of page I want a stranger to land on.

The second was to write the page people are already searching for, even if it's uncomfortable. Anyone choosing a Mac transcription app ends up comparing it with MacWhisper. So we wrote MacWhisper vs VoxStudio for Mac. It opens by saying we make VoxStudio, and it says where MacWhisper is the better choice: if all you want is a dedicated transcription and transcript-editing tool. Pointing a reader at a competitor felt odd at first. It also made the rest of the page more credible.

The third was that one article answers one question, but people have whole jobs to do. A journalist doesn't want a transcript for its own sake. They want quotes they can use. That became the interview to quote sheet tutorial. It's three lessons: transcribe the interview, check who said what, then export the verified quotes to a spreadsheet with speaker, timestamp and context. Of everything on the site, it's the page closest to what the product is for.

Each app needed its own pages too, because people search differently depending on the device in their hand. The guide to recording and transcribing an interview on iPhone covers getting consent, what happens when the signal drops mid-interview, and checking speaker labels against the audio. It's also honest that phone transcription needs a connection.

The last one brought me back to where this started. I'd paid to appear next to AI answers and it didn't convert. Meanwhile VoxStudio had become something people could use inside those assistants. Searching cloud recordings with ChatGPT shows how to connect VoxStudio Cloud MCP, ask about your own recordings, and open the cited passage to check the answer. For someone who already lives in ChatGPT, that's a better reason to try VoxStudio than any ad I could buy there.

Where things stand

After improving the blogs and promoting on other platforms, organic traffic is now roughly on par with the traffic that OpenAI Ads trial brought on the day of the 5.3k impressions. Most of that credit goes to the SEO work on the site itself. I'll keep going.

The other channels have been much quieter. Visits from X are very low. That part I need to look at honestly: how often I post, and whether the content is missing the practical pieces people can learn from, like cutting clips in the Mac app's video editor, or MCP examples. I plan to stick to posting one example a day for the X community, and hope the cumulative effect shows up slowly. Product Hunt got little engagement. I have posts on Substack, and almost nobody is reading them. On Reddit I haven't bought ads yet. I plan to try. Posting promotional copy there is hard, and most of what I submit gets rejected by the mods or the review process.

Paid ads aren't off the table. I plan to try other types next, click and purchase campaigns among them, with a larger budget, to break out of the long stretch of low traffic. The OpenAI test was too small to tell me much about what a properly funded campaign can do. Perplexity is still in planning. Organic stays the base I keep building either way.

I don't have a growth curve to show you, and I won't draw one. What I have is a product that runs in a browser, on a Mac and on an iPhone, a growing shelf of pages that try to answer real questions honestly, organic traffic that can now match what that first tiny ad trial brought in a day, and a plan to try paid again with enough money and the right bid types to learn something real.

If you have recordings you keep meaning to go back through, VoxStudio is where I'd start. The web app works in any browser, the Mac app has a 14-day trial, and the iPhone app is free.

Comments

Popular posts from this blog

Speaker Labels vs. Speaker Names: Getting "Who Said What" Right in Mac Transcripts

The quote her editor asked about at 9:40 p.m.