One OpenAI dots Plugin for Social Listening Across Five Networks

Reddit, X, Bluesky, Hacker News and Facebook all land in the same RedReplier queue, so an OpenAI dots plugin for social listening needs one connection rather than five. OpenAI launched dots on 29 September 2026 as always-on agents with their own cloud computer, and a social listening routine is the kind of job that benefits from an agent that never closes the tab. The hard part is stopping the loudest network from burying the other four.

The short version: add RedReplier as a custom MCP plugin, have your dot call count_mentions once per source before it reads anything, then list_mentions sorted by RELEVANCE. Every call is a read, so a Pro account's read/fetch limit on custom plugins costs you nothing here.

Why does one plugin cover five networks?

One RedReplier plugin covers five networks because RedReplier merges and scores the mentions before your dot sees them. The OpenAPI spec lists seven source values: REDDIT_POST, REDDIT_COMMENT, TWITTER, BLUESKY, HACKERNEWS, FACEBOOK and FACEBOOK_GROUP. The source is a field on each mention, so "what did people say about us today" and "what did people say on Bluesky today" are the same call with a different filter.

Five separate plugins mean five sign-ins, five sets of API terms and five things that break on different days. The setup for the one plugin lives in the OpenAI dots plugin for Reddit monitoring guide: developer mode, the https://mcp.redreplier.com/mcp URL, an OAuth sign-in with your RedReplier account, and a read call that proves your dot uses the plugin. OpenAI has not documented that dots use developer-mode plugins as of 29 September 2026, so run that proof first.

Can a dot reach these networks without a plugin?

A dot reaches any public site through its own Chrome browser, but a browser read gives you pages, not a scored feed. OpenAI's launch post says plugins connect dots to "over 4,000 apps" and publishes no breakdown, so check the Plugins tab yourself before assuming any of these five networks has its own listing.

The browser route has a specific cost for listening across five sources. Each site has its own layout and its own login wall, and nothing ranks a Hacker News comment against an X post. RedReplier attaches a relevanceScore to each mention, and the spec defines a higher score as a better candidate for a reply. That number is what lets one digest compare five networks at all.

What does the first social listening read look like?

The first read counts before it lists, because volume across five networks is uneven and the counts tell you where to cap. count_mentions returns one total for whatever filters you pass, so one call per source shows the skew before it reaches your digest.

Use the RedReplier plugin. Read only: do not call any tool that
changes, approves, rejects or deletes anything.
1. Call count_mentions with statuses NEW, once for each source:
   REDDIT_POST, REDDIT_COMMENT, TWITTER, BLUESKY, HACKERNEWS,
   FACEBOOK, FACEBOOK_GROUP. Show me the seven counts first.
2. Call list_mentions with statuses NEW and sort RELEVANCE.
   Show at most five mentions per source.
3. Put mentions with no relevanceScore in their own group at the end.
4. Mark any mention with no link as "no public link".

Steps three and four handle two fields the spec marks nullable. A mention fetched before the scorer reaches it has no relevanceScore, and a numeric sort either drops it or parks it at an end of the list. And url is null "when the platform gives no public link, as with some Facebook and Bluesky posts", which a dot will otherwise paper over with a guess.

How do I stop one network drowning the others?

You stop one network drowning the others by sorting on relevance and capping per source, not per digest. RedReplier's sort parameter takes RELEVANCE or RECENT, and the spec declares no default, so name it every time.

Recency is where multi-source listening fails. Reddit and X produce far more matches per hour than Hacker News or Bluesky, so a feed sorted by time returns the last hour of the two loud networks. Relevance puts a Hacker News thread scored 90 above an X post scored 30 whenever either landed. Reddit also arrives under two source values, REDDIT_POST and REDDIT_COMMENT, so a cap written for "Reddit" has to cover both or half the Reddit rows slip past it.

Can my dot spot recurring requests the way OpenAI's example does?

OpenAI's launch post describes a dot that "watches customer feedback for recurring requests", and RedReplier's tags field gives your dot the raw material for exactly that. explain_mention fills in relevanceReason, tags and aiReplySuggestion when they are missing, so a weekly task that explains the top mentions and groups them by tag turns five networks of chatter into a short list of themes.

Ask for "every Friday, the top 20 mentions of the week by relevance, explained, grouped by tag, with the count per tag and one example link each". That report is the social listening output a product team reads. Test explain_mention on Pro before you schedule it: the call runs AI scoring and stores the result, and OpenAI has not said whether ChatGPT treats a tool like that as read/fetch. If Pro refuses it, mentions RedReplier already explained still carry their tags.

Where does the social listening digest go?

A dot answers in ChatGPT on desktop, web and mobile, and OpenAI's launch post adds Slack and Microsoft Teams as places to message it. OpenAI's admin guide says Teams access is an invite-only alpha, and OpenAI's getting-started article in the help centre says texting is a limited beta for Pro users in the US, both as of 29 September 2026.

Posting the digest into a shared team channel is a different action from answering you. OpenAI's safety post says that "to send a message or share a file, they are taught to seek authorization", so a Friday report posted to #marketing triggers a request each week. Custom Rules have a setting for this case, "take action when you say so or if pre-approved". Pre-approve that one recurring post and leave everything else on "ask before taking action".

Is ChatGPT Pro enough for social listening with a dot?

ChatGPT Pro is enough for social listening, because every call in this routine reads and none changes state. OpenAI's August 2026 help article says "Pro users can connect MCPs with read/fetch permissions in developer mode", while "Full MCP is only available to Business and Enterprise/Edu." The article predates dots, so confirm it with your first read.

The write tools a listening routine never needs are the ones worth keeping out of reach anyway. update_mention_status rejects mentions that then leave your queue, delete_keyword deletes a keyword and every mention it produced, and update_alert_settings rewrites your email cadence. The pillar guide lists all ten state-changing tools. On Business Premium, where full MCP applies, set Custom Rules to "ask before taking action" for them.

What breaks across five networks?

Four things break once more than one network runs through the same dot, and none of them shows up with Reddit alone.

The first is the Facebook group id. The spec's subreddit field holds the subreddit on Reddit and the group id on Facebook, so a dot that labels every row "r/" mislabels Facebook groups. Tell it to print the source next to the field.

The second is a missing link on Facebook and Bluesky rows, covered by step four above. The third is source skew returning after a quiet week, because a new keyword that matches heavily on X undoes last month's caps. Re-run the counts when you add keywords.

The fourth is memory. Your dot creates its own memories from connected apps, and OpenAI's help centre says you cannot view or edit them and only deleting the dot clears them. Five networks of other people's posts is a lot to hand a store you cannot inspect, so ask for summaries and links rather than full post text.

How does the dots route compare with Muse and Grok?

The dots route is the only one of the three where the agent reads RedReplier over MCP with an OAuth sign-in. The Meta Muse connector for social listening has Muse write a custom connector from the REST spec. The Grok bot template for social listening ships a skill that calls the same API with a redreplier_ token. RedReplier's social listening tool page covers what happens before any agent reads the queue.

Frequently Asked Questions

Does one dots plugin cover all five networks?

RedReplier's list_mentions returns Reddit, X, Bluesky, Hacker News and Facebook in one response, with the network as a source field on each mention. Filtering to one network is a parameter change. The X-only guide covers the one network where volume needs extra filtering.

Can my dot reply to a mention on any of these networks?

RedReplier has no tool or endpoint that posts to any network, so the plugin cannot reply. Each mention carries an aiReplySuggestion, which the spec calls a draft for a human to review. OpenAI's dots docs say "Asking your dot to draft replies doesn't give it permission to send them."

How often does my dot check for new mentions?

A dot runs whenever its recurring task fires, and RedReplier monitors in real time on every plan. Event triggers need a connected service that supports event monitoring, and RedReplier's agent docs list tools and no events as of 29 September 2026. This guide sets up one weekly task: every Friday it explains the week's top mentions and groups them by tag.

Does social listening with a dot count against ChatGPT usage limits?

Conversations with your dot do not count toward ChatGPT usage limits, according to OpenAI's launch post. Tasks the dot starts in Codex or ChatGPT Work do count. OpenAI publishes no rate limits or task caps for dots as of 29 September 2026.

Can I use dots for social listening outside the US?

Business Premium includes dots in every supported ChatGPT region. Pro includes dots everywhere except the European Economic Area, the UK and Switzerland. Enterprise workspaces get a beta that an admin has to turn on.

What does RedReplier cost for multi-network listening?

RedReplier monitors every supported network on all three plans, and API access is included on each. Starter is $19 a month for 15 keywords, Pro is $39 for 30 and Business is $79 for 50. The first dot comes with ChatGPT Pro or Business Premium at no extra cost.

Should my dot sort mentions by relevance or by recency?

Sort by RELEVANCE. RedReplier's sort parameter takes RELEVANCE or RECENT, and the spec declares no default, so name it in every prompt. A recency sort returns the last hour of Reddit and X and crowds out the quieter networks.

Why does my digest label a Facebook group as a subreddit?

The spec's subreddit field holds the subreddit on Reddit and the group id on Facebook. A dot that labels every row "r/" will mislabel Facebook groups. Tell it to print the source next to that field.

What should my dot do with a mention that has no link?

Mark it "no public link" and move on. The url field is null when the platform gives no public link, as with some Facebook and Bluesky posts. Without that instruction a dot may paper over the gap with a guess.

Should my dot keep full post text in its memory?

Ask for summaries and links instead. Your dot creates its own memories from connected apps, and OpenAI's help centre says you cannot view or edit them. Only deleting the dot clears them.

Create your RedReplier account and give your dot one read-only feed across five networks, or read the agent docs first.

Sources: OpenAI's dots launch post, dots safety post, getting started help article, safety FAQ, dots user docs, admin guide and the August 2026 MCP help article, plus the RedReplier OpenAPI spec, agent docs and pricing. Checked 29 September 2026.