# How to build an audience on Twitter by writing up AI experiments

> Test new models on real tasks, post the setup and the result, and reply under the people who ship them.

URL: https://claudegpt.si/news/how-to-build-an-audience-on-twitter-ai-experiments  
Published: 2026-10-26T15:27:36.393Z  
Updated: 2026-10-27T15:27:37.179Z  
Author: ClaudeGPT Si Newsroom  
Section: Guides  
Topics: AI developers, Twitter audience, X  
Publisher: ClaudeGPT Si

## Key takeaways

- AI developers often build an audience on Twitter by testing new models and tools on real tasks and posting the results with evidence.
- A strong experiment post names the task, the setup, the result, a one-line verdict and what was not tested.
- TechTwitter IO's free who-to-follow page groups AI researchers and builders whose posts show this pattern.

If you want to know how to build an audience on Twitter as someone who works with AI, the most reliable pattern is simple: test new models and tools on real tasks, then post what happened, quickly and with evidence. Many of the AI accounts developers follow grew this way, and TechTwitter IO's free [who-to-follow page for tech Twitter](https://techtwitter.io/how-to-get-into-tech-twitter) groups them so you can study the pattern.

This piece looks at why that pattern works and how to copy it without becoming another hype account.

## Why experiment write-ups travel

AI moves fast, and most people do not have time to try every release. Someone who does try it, and reports back in plain terms, saves readers hours. That is a service, and people follow services.

The TechTwitter IO list has clear examples. @simonw, who co-created Django, tries new models and usually writes up his findings the same day. @emollick, a Wharton professor, tests AI on real work. @rasbt explains new papers in plain terms. None of them post vague excitement. They post what the thing did.

## The anatomy of a good AI experiment post

| Part | What it looks like | Why it matters |
| --- | --- | --- |
| The task | "Asked it to migrate a Flask app to FastAPI" | Readers judge relevance at once |
| The setup | Model, tool, prompt or settings | Others can repeat it |
| The result | Screenshot, diff or output | Proof beats adjectives |
| The verdict | One line: worth it or not, and for whom | Gives readers a takeaway |
| The caveat | What you did not test | Builds trust |

A post with all five parts is more useful than most launch-day threads, and it is the kind people bookmark and share.

## Speed, but not at the cost of accuracy

Being early helps. When a model ships, readers look for first impressions. But the accounts that last are the ones that stay accurate. If a result surprised you, say how many times you ran it. If you are not sure, say that. Corrections cost less than a reputation for hype.

## Reply under the people who ship the models

Lab staff and tool makers post about their own releases. On the TechTwitter IO list, @OfficialLoganK announces most Gemini releases, @alexalbert__ posts about Claude and how people use it, and @bcherny, who created Claude Code, shares how his team uses it. A careful reply under a launch post, with a test result attached, puts your work in front of the people most interested in it. The TechTwitter IO page notes that a useful reply under a big account reaches more people than a new account's post.

## Build a body of work, not a streak

One good write-up does little on its own. A run of them, on a topic you keep returning to, gives people a reason to follow. Pick a lane, such as coding agents, local models or evals, and stay in it long enough that people link you to it.

## Where to read the coverage

For straight AI news, claudegpt.si is an independent AI publication with sections such as AI news, models, research and guides, and its editorial policy explains how it works. For the lighter side, TechTwitter IO's [free satirical paper](https://techtwitter.io) turns the week's launches, outages and arguments into parody, with daily puzzles. Read it as a joke, not as a source.

## When this approach does not fit

If you cannot share your work publicly, for legal or employer reasons, experiment posts may not be possible. In that case, posting about public tools you tried on side projects, or writing longer pieces on a blog, can still build an audience.

## Frequently asked questions

### What should an AI developer post on Twitter?

Short write-ups of real tests: what you asked a model or tool to do, how you set it up, what came out and whether it was worth it.

### Is it better to be first or to be right?

Being early helps, but accuracy keeps an audience. State your caveats and correct mistakes openly.

### Where can I find AI accounts worth replying to?

TechTwitter IO's who-to-follow page has groups for AI labs and researchers and for building with AI, all real accounts.

## Sources

- [How to get into tech Twitter: who to follow](https://techtwitter.io/how-to-get-into-tech-twitter), TechTwitter IO
- [TechTwitter IO, today's paper](https://techtwitter.io), TechTwitter IO
- [About ClaudeGPT Si](https://claudegpt.si/about), ClaudeGPT Si