Social

Social listens to public conversations on X, Reddit, and LinkedIn for moments worth replying to — someone complaining about a tool you replace, asking their network what to use, or announcing they just stepped into the role that owns the problem you solve — so you can join the conversation while it's live. There's nothing to configure: Star turns your product, ideal customer profile, and competitors from Settings → AI into listening angles and starts scanning on its own.

What gets found

Two kinds of opening, weighted equally. Problem signalsare the ones you'd expect:

• Pain points — someone publicly describing the problem you solve
• Competitor complaints — unhappy users of a named competitor, or people asking for alternatives. Competitor watches also scan fresh G2 and Capterra reviews, where reviewers are usually anonymized — those finds surface as account-level intelligence rather than contacts
• Tool evaluations — buyers actively comparing options in your category
• Asks for recommendations — "what do you all use for…?" posts, including open-ended ones that never name a category

Moment signals are the ones most tools miss. Nothing is wrong in these posts — something just changed:

• New role — someone just joined, was promoted, or was handed a number. They own the problem now, and have no stack yet
• Company milestone — a raise, a launch, or an expansion, so budget and pressure just arrived
• Team changes — a team being built or rebuilt
• Sharing an approach — describing how they run the process you improve

Every find is a real, public post: Starlight keeps only posts it can quote verbatim with their actual URL, and drops anything it can't source. Scanning uses licensed X data and the public web — no scraping, no logged-in automation, nothing that touches your accounts.

The feed

The feed is ranked by how worth a reply each post is, not by recency. Star scores relevance to your ICP and product separately from keyword match, then prefers posts in the engagement sweet spot — enough likes that the account is real, few enough comments that yours will be seen. Competitor and vendor posts, recruiter listings, and press about someone else stay in Low fit instead of the main feed. Auto-drafted outreach only fires on the highest-scoring finds.

Each opportunity is a card: the post itself (linked), who wrote it, why it's an opening for you, and a suggested commentwritten to best practices — it references a specific point from the post and adds something of your own. Comments weave in a light, human emoji accent mid-sentence (never a trailing flourish); Reddit stays conservative and will skip the emoji when it would read as a brand account. On an announcement post it congratulates specifically instead, naming the actual role or milestone and never implying a problem they didn't mention.

When what you sell genuinely fits what they wrote, the comment says so straightforwardly — and says you work here. A plain "full disclosure, I work at [your company]" followed by a simple reason it fits and your honest experience, including where it isn'tthe right answer, earns replies; the same comment without the disclosure gets you buried. It never uses marketing language, never links, and stays quiet about your product entirely when it isn't relevant to the post — a forced mention costs more than it earns. Everything it claims comes from your product description in Settings → AI, so it can't invent features or numbers.

On LinkedIn opportunities, with the Chrome extension installed, Post comment posts it for you: the card opens the comment for a final look (edit it right there if you like), and confirming with Post via Extension has your own browser open the post and publish the comment from your account. Nothing is ever posted without that explicit confirmation, and the card flips to Engaged once the post is confirmed.

Copy & open post remains alongside it — and is the way to engage on X and Reddit: it puts the comment on your clipboard and opens the post in a new tab, so you paste, tweak if you like, and post it yourself. New opportunities also appear on Tasks and as notifications.

The skill

The Skill tab holds one thing: an open markdown document describing how Star hunts social for you — who to look for, the moments worth acting on, competitors to shadow, the literal phrases your buyers type, what to skip, and how comments should sound. Star writes it for youfrom your product and ICP in Settings → AI, so there's nothing to set up — but it's yours to edit. Change anything and save; the AI immediately re-plans what it searches for behind the scenes. "Rewrite with AI" drafts the whole document fresh from your profile whenever your product or ICP has moved on.

Under the hood, the search engines Star uses run literal keywords rather than prose, so it compiles your instructions into a concrete set of daily searches (stretching to every few days when an angle stays quiet) and manages that list itself — you never have to curate it. The skill also guides the relevance judgment on every post found, and its tone section shapes the suggested comments.

Deeper scans with the Chrome extension

With the Starlight Chrome extension installed, scans go further than web search can. LinkedIn hides most post bodies from search engines, so vendor results are often snippets — but your own logged-in LinkedIn sees everything. When a watch scans, Starlight also asks your extension to quietly open LinkedIn and X search for the same phrases in a background tab, read the results — full post text, engagement counts, real profile links — and send back what it found. The extra finds land in the same feed, judged by the same relevance check.

The extension only reads. It never posts, likes, comments, or connects during a scan. On LinkedIn it does use each post's "Copy link to post" menu item, because that is the only way to get a post's address — but it catches the link in transit, so your own clipboard is never overwritten. Scans run at most once every half hour or so, a few searches at a time, roughly the pace of a person browsing. You can turn it off any time in Social settings with Scan with your Chrome extension; it costs no extra blocks.

Mining a competitor's engagement

People who react to a competitor's LinkedIn posts are already in the market — they follow the space closely enough to engage with it. On a Competitor watch, turn on Mine reactionsand, once a week, your extension opens the competitor's LinkedIn page in a background tab and checks who reacted to their recent posts. Reactors whose headline reads like your buyer are verified against an independent data source, saved to Contacts, and appear in the feed tagged Engaged with competitor. This scan is read-only like everything else the extension does — it never reacts, follows, or comments.

Outreach to these people is deliberately ordinary cold outreach. Their reaction is used only to decide who to contact — drafts never mention the post, the like, or the competitor, so nothing about the message suggests their activity was watched.

From post to pipeline

For strong finds, Starlight identifies the person behind the post — their real name, title, company, and LinkedIn profile — and adds confident matches to Contacts. With Auto-drafton (the default), a strong opportunity also drafts a personalized sequence that references the conversation naturally, like a peer joining it — never "we saw your post." The post is passed to the writer as sourced research, every draft is quality-judged, and everything lands in Messages as pending review — nothing sends without you.

Autopilot is different, and off by default: when you enable it in Social settings, social drafts that pass the quality check are approved and sent without your review. Suggested comments are never affected — a comment only posts when you click Post via Extension (or paste it yourself).

Pausing Social

Heading out for a while? Click Pausein the feed's top bar. While paused, no listening scans, extension scans, or auto-drafts run and no blocks are spent — the button flips to Resume, and listening picks back up on the next cycle after you resume.

Blocks

• Listening: 4 blocks per watch per daily scan (suggested comments and poster identification included)
• Drafting from an opportunity: the standard 4 blocks per lead, same as Draft Messages with AI
• Posting a comment via the extension: 2 blocks, charged only when the post is confirmed
• Mining a competitor's reactions: 2 blocks per competitor per weekly run, refunded if the scan never runs