Publit.Si
Docs›Competitors

Using Publit.Si

Competitor research

See what your competitors publish – Shorts, posts, blog articles – and on which platforms. Your AI researches on the web, writes down the tactics that stand out and suggests posts you can make right away. Competitors are saved per project.

Start the research

  1. Open Research › Competitors. With All projects in the sidebar, pick the project first.

  2. Enter a competitor you know on the start card and click Add. A name is enough. You can skip this – the research suggests some.

  3. Click Research with AI for a quick run, or open the arrow next to it and choose Thorough. It takes a few minutes and you can leave the page. Open in Studio shows what Claude is doing right now.

What it reads

The research runs with Claude Code on your computer – or with Codex, if you choose it under Settings › AI › Research runs with. Codex runs locked down in a configuration of its own (no shell, no files, none of your own tools) and has no cost cap: a run stops after 15 minutes (quick) or 40 (thorough) and uses your ChatGPT plan. Publit.Si hands it your repository's README, AGENTS.md and CLAUDE.md so it knows what the product is; it searches the web and reads your competitors' websites, blogs and public profiles.

YouTube channels are read through YouTube's API with the Google sign-in of a YouTube account you connected: subscribers and the newest videos with views, likes and comments. On other platforms it saves only what a public page shows. A number it could not see stays empty – nothing is estimated. Competitors' YouTube numbers are kept for at most 30 days, as YouTube's developer policies require: what a refresh does not renew within that time is deleted.

Some platforms show little without a login. The summary says what could not be read.

A research run reads the competitors' YouTube channels through YouTube's API: subscribers and the newest videos with their numbers. Other platforms have no such API – there the run saves what the public pages show.

What the research can and cannot do

A research run reads pages that anyone can write. That is why it is locked down: no shell, no file tools at all – it cannot read your repository, so a page cannot talk it into reading .env –, no other MCP servers, and none of the tools that create, change or schedule posts. It can only save competitors, their posts and the report. A later message in the same Studio chat keeps these limits.

Quick runs on Claude Sonnet, looks at up to 3 competitors and stops at $3 of API value. Thorough uses your own model, looks at up to 6 and stops at the amount under Settings › AI ($15 unless you change it, $5–$50). With a Claude subscription you pay its flat price instead. The header shows when the last run was, what it cost and how long it took.

The run is kept in Publit.Si's database: a restart does not lose it, and a second run of the same page cannot start while one works. A run cut off by a restart says so, and you can start it again.

Read what works

The page has three tabs: Ideas, What works and Competitors. Once there is a research it opens on the ideas. The header shows the day of the last research; while a run works, a line says how many competitors it has found and how many posts it has read.

What works starts with the research in one sentence; The whole summary opens the rest. Most viewed posts shows the competitors' saved posts with the most views, as the platform reported them – YouTube videos with their preview image. Publit.Si adds no score of its own. Tactics that stand out and Gaps – what competitors do that you do not, and the other way round – show one line per finding; click a line to read it in full.

An opened finding shows the competitor posts it is based on, with their preview image where there is one. Evidence always points to a post the run saved; a link it did not save is dropped when the report is stored.

Decide who counts

What the research found waits in a strip above the tabs until you decide. Follow keeps a competitor in your list. Ignore folds it away under Ignored on the Competitors tab – the research never suggests it again. Remove deletes it with its channels and saved posts.

The Competitors tab is one table: a row per competitor, a column per platform – the bigger the dot, the more followers – and each one's strongest post. Click a name to see every channel with its saved posts, the most viewed first.

Turn ideas into posts

The Ideas tab shows each idea with its platform, its hook and the competitor posts it builds on; Details opens the caption and why it should work, and the chips at the top filter by platform. Open as draft puts the text into the editor. The More menu has Make it in the Studio, which asks your agent to make the post and hand it over, and Save to backlog, which keeps it as an idea folder for later.

Ideas adapt a tactic to your project. They never copy a competitor's text.

Your AI uses it too

For 60 days after a research run, every AI text in Publit.Si and the Shorts automation get a short list of the tactics and gaps. Claude and Codex can read the whole research with the tool get_competitor_research (see MCP tools).

The header says until when the research steers AI texts, and when it is older than 60 days and no longer does.

Research again automatically every 30 days at the bottom of the page keeps it fresh: the worker starts a quick run (up to $3) in the background – also while the app is closed, as long as the computer is on – and sends a notification when it is done. It is off until you switch it on, separately for Competitors and Reddit; a failed run is tried again the next day.

Past research below it lists every run with its date, quick or thorough, by hand or automatic, and its cost; the arrow opens its Studio chat. Delete removes a run's report and moves its chat to the Trash; Delete all but the newest clears out the rest. Competitors, subreddits and saved posts stay.

Check

Research › Competitors lists at least one competitor on the Competitors tab, and the What works tab starts with one sentence and has tactics below it.

Stuck? Search the troubleshooting list or write to us . Leave out secrets and tokens.