## An Overlooked Time Sink

An Overlooked Time Sink

As an indie developer or member of a small team, you've probably been through this: the code is merged, the release is packaged, deployment is live—and then the product manager (or just you yourself) reminds you: "Did you write the changelog? Did you publish the announcement?"

So you open your Git history, scroll through dozens of commits, trying to recall "why did we make this change again?" while attempting to translate fix: 修复了若干问题 into something users can actually understand. After finishing the changelog, you still have to rewrite it in a different tone for the release announcement, then post it to the product dashboard, mailing list, and user groups—each channel with a slightly different voice.

This task has zero technical difficulty, but it has three characteristics: repetitive, time-consuming, and easy to procrastinate on. The reality for many teams is that the changelog lags two or three versions behind, and release announcements simply never get written. Users are left to discover on their own: "Huh, why did the button move?"

Estimating 90–120 minutes per release and 4 releases a month, that's 6–8 hours. For a large team, this is nothing; for an indie developer, it's almost an entire day of productive development time.

What AI Can Actually Do for You

The core idea of handing this over to an AI pipeline is: let AI handle the two most time-consuming steps—"translation" and "multi-channel rewriting"—while you only review and fine-tune.

Here's the division of labor:

What AI can do:

• Extract user-facing change points from raw commit history, PR descriptions, and task tickets
• Distinguish between "what users need to know" and "pure internal refactoring" (users don't care at all which dependency library you swapped)
• Generate different versions for different channels: an announcement for all users, an email for paying customers, a short message for community groups
• Maintain a consistent tone and expression style across every release

What you still need to do:

• Confirm the priority and categorization of changes
• Check for sensitive information leaking out (unreleased features, internal codenames)
• Decide which content deserves to be highlighted or pinned

Note one key point: don't make the AI guess from scattered, raw commits. The quality of the input determines the quality of the output. My own approach is to compile the titles and descriptions of the PRs merged in this release, along with key commit messages, into a single text block before feeding it to the model. This prep step takes about 2 minutes, but it raises the usability of AI output from "needs major rework" to "basically ready to use as-is."

Breaking Down the Actual Workflow

The process is now fixed into four steps:

1. Collect (2 minutes): Export the list of changes included in this release from Git or your project management tool, including PR titles, descriptions, and affected modules.
2. Generate (1–2 minutes): Use the prompt template below to generate three pieces of content in one go—a user-facing changelog, a release announcement, and a short community message.
3. Review (3–4 minutes): Read through everything, remove internal information that shouldn't be exposed, adjust individual wordings, and add screenshot links.
4. Publish (1 minute): Paste to each channel.

The before-and-after time comparison is striking:

In other words, you save about 100 minutes per release, which adds up to 6–7 hours across 4 monthly releases. What matters even more is the quality improvement: when writing changelogs manually under time pressure, they often end up as lazy copy like "fixed some bugs, improved the experience." AI-generated versions describe each change as a complete sentence from the user's perspective and automatically categorize them into "New Features / Improvements / Bug Fixes." User feedback has become noticeably more positive—though I won't share specific numbers here, you can try the comparison yourself.

As for cost: these text tasks consume very few tokens per run. Even with a flagship model, a month's worth of release copy costs far less than one hour of your hourly rate. For specific pricing, refer to the official pricing page.

A Ready-to-Copy Prompt Template

你是一个 SaaS 产品的技术内容写手。我会提供本次发版的原始变更记录
（PR 标题、描述、commit message），请帮我生成三份内容。

【背景信息】
产品名称：{产品名}
产品定位：{一句话描述，例如"面向小团队的轻量级项目管理工具"}
本次版本号：{版本号}
目标用户：{用户画像}

【原始变更记录】
{粘贴 PR 标题与描述、关键 commit}

【要求】
1. 只保留用户能感知到的变更；纯内部重构、依赖升级等一律不写。
2. 每条变更从用户视角描述"这对我有什么用"，不使用内部术语和代号。
3. 分为三类：✨ 新功能 / 🔧 改进 / 🐛 问题修复。

【输出三份内容】
A. 更新日志（Changelog）：条目式，每条一行加粗标题加一句说明。
B. 发布公告：面向全体用户，150-250字，友好、专业、不过度营销，
   开头一句话概括本次更新重点，结尾引导用户反馈。
C. 社区群短消息：80字以内，口语化，可带一个 emoji，突出最值得说的1-2个变更。

如果某条变更意图不明确，不要猜测，单独列出向我提问。

The last line is crucial—having the model proactively ask questions instead of fabricating when information is insufficient blocks most "AI hallucination-style feature descriptions." This is far more cost-effective than later having to retract a nonexistent feature from your announcement.

A Few Additional Tips

• Maintain a tone sample: Put your most satisfying past announcement into the prompt as a style reference; consistency will improve significantly.
• Generate Chinese and English separately: If you have overseas users, have the model translate based on the already-reviewed Chinese version rather than regenerating English from the raw commits—this avoids the two versions diverging.
• Integrate the API for semi-automation: If your release process is already scripted, you can encapsulate this step as part of your release script, calling an LLM API to generate drafts and automatically create documents pending review. By connecting to multiple models through a gateway like ThisToken.AI, you can also automatically select more cost-effective models for low-risk tasks like copywriting, reserving flagship models for scenarios that truly need them.

To get this workflow actually running, you can start by registering an account: https://api.thistoken.ai/register

The one or two hours saved per release adds up over time—enough for an extra user interview, or simply to clock out a bit earlier.

---

Ready to try it yourself? Sign up at https://api.thistoken.ai/register to get your API key and start building.