Every content creator hits the same wall. You open Buffer, you click Create, and you stare at the empty box. The problem is never a lack of opinions. It is the gap between "I have things to say" and "I know which one is worth a post this week."
A software engineer wrote on buffer.com about building a side project that sits next to Buffer's Create flow. It asks three onboarding questions, fetches current trends from the user's industry, and pushes a packaged draft, hook, body outline, call to action and hashtags, into Buffer via the API. The author's stated goal was simple: kill the blank-screen problem.
What happened:
- The tool is a personal side project, not an official Buffer product.
- Stack: Nx monorepo, NestJS backend, Angular 19 frontend.
- Onboarding covers industry, sub-topics and target audience.
- Each idea returns as a structured JSON package.
- Ideas flow straight into Buffer's Create Space via the API.
What the app actually does
The author describes a short onboarding form: industry, sub-topics inside that industry, and the audience you want to reach. Those answers stick to the profile and shape every later request. The author argues this is what separates the tool from the usual idea generators, which throw generic suggestions like "post a tip" no matter who you are.
On the generation side, the app fires off a single request to the OpenAI Responses API with web search turned on. Five ideas come back at once. Each one is wrapped in a strict JSON schema so nothing is fuzzy or unstructured. The author treats the schema as the heart of the product.
For every idea the user gets a ready-to-draft package: a working title, a hook line, a body outline, a call to action, the main talking points, a suggested format, hashtags, the platforms that fit, the trending topic it ties back to, and a source URL when one exists. Rich-text markers, bold, italic, bullets, are written in Unicode so they survive the trip through Buffer and onto the feed.
Our analysis
The interesting design choice is the schema. Most AI idea tools return prose, which the user then has to interpret. Returning a fixed shape, title here, hook there, CTA at the end, turns the model from a brainstorming partner into a content assembly line. The user still writes the post. The scaffolding is already there.
The other smart call is the onboarding. The author correctly diagnosed that generic output comes from generic input. A tool that knows your industry, your sub-niches and your audience produces tighter suggestions than a tool that only knows your job title. The trade-off is the upfront cost. If the first five ideas disappoint, there is no reason to come back and refine the answers.
There are limits the author does not address. The "trending in the last 72 hours" window is a snapshot, not a longitudinal view. The single web search inherits its model's search bias. And the structured output is tuned for LinkedIn's norms. A second platform would mean a second schema, and a second round of tuning.
The hidden risk is sameness. If a thousand people adopt a tool like this, a thousand LinkedIn feeds start to converge on the same hooks and the same CTAs. The differentiator is still the writer.
A worked example
Say you are a backend engineer who posts on LinkedIn about developer tools, SRE practices and incident retros. You run through onboarding once. The model returns five ideas tied to whatever is moving in the dev-tools world that week. One idea lands well: a post about a recent outage at a major SaaS vendor, framed as "what we can learn from their postmortem." The hook, outline and CTA are already wired up. You click to push it to Buffer and the blank-page deficit drains.
Bottom line
This is the kind of tool the Buffer ecosystem was built for. It does not replace the work of writing. It replaces the slowest part of writing, which is choosing the topic and the angle. The risk, as with any AI shortcut, is sameness. The edge is still the writer.
Key takeaways
- The schema is the real product, not the AI call.
- Onboarding answers are the lever; bad input still produces generic ideas regardless of model quality.
- Unicode formatting is the small detail that keeps the draft readable on LinkedIn.
Frequently asked questions
Is the Buffer Ideas Extension an official Buffer product?
No. The buffer.com post describes it as a side project built by a Buffer user, separate from Buffer's official product line.
Which API does the app use to generate ideas?
The author describes using the OpenAI Responses API with web search enabled and a strict JSON schema on the output.
How are ideas pushed into Buffer?
Via the Buffer API, after the user picks which ideas to use from the generated batch.
Does the app preserve formatting when pushing to LinkedIn?
The author states that rich-text formatting uses Unicode so bold, italic and bullet markers survive the trip through Buffer and onto LinkedIn.
Can a non-technical user replicate this?
The author claims yes in principle, but the stack, Nx monorepo, NestJS and Angular 19, is a developer stack. A non-technical user would need a no-code clone or a community port.