Deep Social — big social research without the noise

How to start a run

You start it. The chat cannot start one for you — a run takes about five to eight minutes, so it is always your call.

1. 01

   Click the telescope button

   The telescope icon sits next to the + button in the chat composer. Click it and pick Deep Social. The button turns green while it is armed.

2. 02

   Type your research question

   Whatever you send next becomes the research goal. "What hooks are working for money creators on X right now?" "Study outlier YouTube titles in the fitness niche." "Compare how these three creators open their videos." The toggle clears itself after that message, so the one after it is a normal chat turn again.

3. 03

   Walk away

   The chat replies in seconds to say the run has started, then keeps researching in the background. You can keep chatting, switch chats, close the tab, or shut the laptop. A card in the chat streams what the agent is doing — each search, each creator it checks, each post it reads, each finding it writes down. The finished report lands in that same chat, on its own, in about five to eight minutes.

One depth, the full sweep

There used to be Quick and Standard options. They are gone. Arming Deep Social always runs the exhaustive version: up to 12 findings from dozens of searches and full-post reads, about five to eight minutes, always in the background.

Small lookups do not need it. Ask the chat normally for one creator, one post, or a single search — it has fast tools for those and will use them. Deep Social is for the sweeps those tools cannot finish in one turn.

What the report looks like

Deep Social is deliberately a fact gatherer, not an opinion machine. The agent is banned from editorializing. Its report contains:

• Findings — plain-spoken claims about what is working and why it appears to work.
• Verbatim quotes — exact hooks, titles, or comment lines, copied word-for-word from the posts. The system rejects any quote that doesn't match the source text, so what you see is what was actually written.
• Real metrics — views, and outlier multiples (how many times better a post did than that creator's baseline).
• Receipts — links to every post behind a finding. Tap one to open the post.

The chat then interprets the report in your context: what it means for you, what to try, what to skip. The thinking layer stays in the conversation where you can push back on it.

When the run finishes, the card also offers "Everything it found" — the full list of every post the agent scanned, not just the ones that made the report. Tap any row to open that post.

Example: YouTube title research

This is a job Deep Social is built for. Ask something like:

"Study the top-performing YouTube titles in the productivity niche from the last 3 months. I want the exact titles, what patterns repeat, and which ones are beating their channel's baseline the hardest."

The agent restricts the sweep to YouTube, pulls outliers, reads the top performers, and returns the exact titles as verbatim quotes with each video's outlier multiple. Follow up in the chat: "rewrite my draft title in the three strongest patterns."

The same pattern works for X hooks, LinkedIn openers, Instagram carousel covers, TikTok, Threads, and Substack — all seven platforms Eden indexes.

Tips for better runs

• Give it a focus. "Focus on the hooks" or "focus on how they frame pricing" steers what the agent quotes. Without a focus you get general findings; with one you get exactly the material you need.
• One niche per run. A run on "fitness AND business AND crypto" spreads thin. For multiple niches, arm the toggle again for each one — a finished report stays in the chat, so the takeaways stack up.
• Interrogate the report. The report is the start, not the end. Ask "which of these fits my voice?", "expand on finding 2", "show me the full list of posts you scanned."
• Creator-centric missions study full histories. "Study everything @creator has posted" makes the agent go through their whole archive, not just recent posts.

Where to go next

• To research your own saved library the same way, read Deep Synthesis.
• To understand the outlier multiples in the report, read the outlier multiplier explained.