Some of my App Store descriptions had not been touched in nearly ten years. Open one and you could still smell "version 2016" in the tone. Since Gemini Advanced shipped Canvas mode, I have been chipping away at this backlog, and I just finished updating the fortieth one.
I have been shipping iOS and Android apps as a solo developer since 2014, and the combined catalog has crossed fifty million downloads. Even so, sitting down to seriously rewrite store copy is something I rarely make time for. Here is what changed once I started doing it inside Canvas, what stayed the same, and what I think about Canvas as a tool after three weeks of real use.
Why I picked up the work now
The trigger was simple. The latest review guidelines tightened the requirement that store copy must match the actual functionality of the app. When I audited my forty apps, more than half had either described features I had since removed, or were missing features I had added in the last two years.
Both ChatGPT and Claude can rewrite copy. Until recently I would clean things up one app at a time inside Claude. Canvas caught my eye because it offers a side-by-side document and chat surface, which felt like the right shape for "edit the details while keeping the whole structure in view."
Back when I used to sit in coffee shops near Kichijoji Station and write longhand, my best drafts always came when I could see the whole page. Canvas reproduced a fair bit of that feeling inside a browser tab.
How I broke the work into phases
Forty descriptions in one sitting is a recipe for fatigue and a quality drop in the second half. I split the three weeks into three phases instead.
Week one was "ground truth." I pulled each app's current description, the latest App Store Connect metadata, and six months of reviews into a single spreadsheet, then fed each block into Canvas with the prompt "list what is missing from this description compared to the actual app." Ten minutes per app, forty apps, but at the end I finally had a complete picture across the catalog.
Week two was "draft generation." I put the current description on the left in Canvas and asked the chat on the right to "rewrite this in natural 2026 Japanese while preserving the original positioning." Canvas applies drafts directly to the document, so it is easy to highlight only the paragraphs you dislike and regenerate them in place.
Week three was "human edits and A/B candidates." Drafts never quite sound like me, so I rewrote endings, examples, and rhythm by hand. For each app I kept two variants side by side in Canvas so I could later run them through Custom Product Pages in App Store Connect.
What I let the AI handle, what I always touched
Structural reorganization was safe to delegate. The old descriptions were inconsistent — some opened with feature lists, others with a long poetic preface, others with a developer's greeting. Canvas reliably reorders things into "reader benefit → features → how to use → app philosophy," and the same prompt worked across nearly all forty apps.
It is also good at preserving the rhythm of long sentences while restating the same point more concisely. This is a property of the whole Gemini 3.x family, but Canvas makes it more useful because the model can see the surrounding paragraphs while it works.
Three things I always touched by hand.
First, trademarks and product names. Canvas helpfully tries to write "iPhone and Android" together even when the app is iOS only. I always did a final visual pass.
Second, language taken from reviews. When a reader had written "I open it before bed and feel calm," I wanted that exact phrasing near the top of the description. Pulling reader language into the copy is closer to a blogger's job than a writer's job, and the AI tends to sand it down into something more generic.
Third, the closing line. My catalog leans toward wallpaper and well-being apps, so I want the last sentence to invite a small exhale. Ever since 2019, when I saw a ring of light above Kichijoji Station and started taking visual expression seriously, I have cared about the negative space between images and words. This is one part I cannot delegate.
Small ASO effects
Across the three weeks I also tidied the subtitle and keyword fields. Afterward, Search Ads impression share rose ten to twenty-five percent on seven apps, stayed within noise on twenty-eight, and slightly dropped on five.
The five that dropped had the same problem. Canvas had made the description so smooth that the slightly awkward search terms had been removed. When you ask Canvas to "make it natural Japanese," it shaves off the keywords you actually need for search. Halfway through I started asking for "natural at seven, searchable at three," which produced better results.
The numbers are not dramatic, but multiplied across forty apps the work was worth doing. If you go into Canvas expecting your downloads to double, you will be disappointed. The honest summary is "small, real, durable lift."
Where Canvas fits and where it does not
Canvas felt right for these situations. Editing a long document while keeping the whole thing in view. Summarizing a multi-section reference like Apple's guidelines and applying it to my own app. Comparing two candidate versions side by side. In a chat-only UI these all turn into a lot of scrolling and lose context quickly.
It felt wrong for editing a single sentence. Opening Canvas and waiting for the document surface to load is heavier than a sub-one-minute task deserves. The regular chat box, or Gemini Code Assist inside the editor, is faster for that.
It also struggles with documents that mix prose and code. Twice during the three weeks I watched the model edit inside a markdown code fence and break the snippet. For technical writing with lots of code, I would still draft in a normal editor.
What I want to build next
Two follow-ups came out of this project.
First, a "review the store copy every six months" workflow. The week-one work of pulling current copy, reviews, and metadata into one document is the right thing to automate with Apps Script. If I can get Canvas to receive that bundle automatically, the maintenance pass becomes feasible at a half-year cadence.
Second, the English versions of the same descriptions. This round I only touched the Japanese copy. Canvas understands the stylistic differences between languages well enough that "preserve the nuance of the Japanese version but make it natural for English-speaking readers" should land most of the way.
Both of my grandfathers were temple carpenters, and they would re-thatch the roofs of old buildings on a quiet, decades-long rhythm. App descriptions deserve the same kind of patient, regular care if you want the app to keep finding new readers over the years.
If you also have a backlog of stale store descriptions sitting in App Store Connect, I hope some of this is useful. Thanks for reading.