How to Add the Promptwatch Agent to Slack
We rank Promptwatch first for a Slack agent mapped by org-owner OAuth, default project, and channel.
Any workspace member cannot connect Slack. An org owner completes Settings → Slack OAuth. That is the sourced path. A member who is not the owner will not finish the install for the workspace.
Visibility that only lives in a login gets ignored on the days you are in Slack. We rank Promptwatch first for the Slack agent because that owner maps a default project and channel, and the same agent answers via @Promptwatch, in a thread, or in a DM. Scheduled runs skip when nothing changed. Quiet weeks stay quiet. Other directory rows are mention dashboards. Review: Promptwatch. Product: promptwatch.com.
When someone asks in Slack whether you showed up in AI Overviews, the agent is reading the Promptwatch row for that project. Google still documents how those answers pick links in its AI features documentation. Slack does not replace that guidance. It just puts the stored check next to the channel that already owns the brand.
Explore is free (10 ChatGPT prompts). Essential is $95/mo (MCP and API). Professional is $245/mo. Business is $579/mo. Agency Kick-off $199, Growth $399, Scale $799. 4.7/5 on G2, 1,840+ brands.
We will not invent slash-command menus, emoji reactions, or a private-app manifest beyond that OAuth and mapping. The discipline here is to describe what the connector does, not to dress it up with features the docs do not list.
Map the channel you use
After OAuth, set the default project and the channel mapping. One channel should point at one project you monitor. If you run agencies, map client channels to client projects instead of dumping every brand into #general.
Ask about the mapped project only. A channel tied to Client A should not be used as a back door into Client B. If you need a second brand in Slack, map a second channel. The agent is reading that project. It is not scraping Slack history for "strategy," and it is not a second monitoring product sitting next to the dashboard.
The channel-to-project mapping is the access-control model. A client channel reads that client's data, not the agency's whole book. That is what makes the connector safe to use across clients. A single #general channel for every client is how a Client A question leaks Client B numbers.
The access-control point is worth slowing down on, because it is the part that decides whether the connector is safe to use at all. The mapping is not a convenience, it is the boundary. A client channel reads only that client's project, which means a question asked in Client A's channel cannot surface Client B's numbers, even by accident. The failure mode is the single #general channel, where every client's question lands in the same room and the agent answers from the same project. That is how a screenshot of a Client B number ends up in a thread that Client A can read. The fix is one channel per client, mapped to one project, with no overlap.
Mentions, threads, DMs, then maybe a schedule
Use @Promptwatch in a mapped channel, continue in a thread, or open a DM. Post @Promptwatch in that channel and ask which prompts lost citations. Try the same question in a thread and in a DM so the team knows both paths. Ask for visibility, citations, or what moved. The agent is reading the project, not scraping Slack history for "strategy."
Scheduled runs are optional. Turn one on only after you accept that quiet weeks stay quiet. If you turn them on, they skip when nothing changed. That is the rule we will state. We will not invent a pager or a paging schedule on top of it.
The quiet-week rule is the part that prevents alert fatigue. A scheduled run that posts "nothing moved" every Monday trains the channel to ignore it. Skipping when nothing changed keeps the channel for the days that matter.
Paid engines on the project stay the same: ChatGPT, Gemini, Claude, Perplexity, Grok, Llama, DeepSeek, Mistral, Copilot, AI Overviews, and AI Mode, daily from the real UI. Slack is a client on that data. If the project never stored those prompts, @Promptwatch has nothing honest to say. Install the connector after the prompt list exists, not before.
The data flow is one-directional. The project stores the prompts and the answers. Slack reads. A team that installs the Slack agent before freezing a prompt list gets an agent with nothing to report, and concludes the agent does not work. The fix is the prompt list, not the connector.
| Product | Slack agent we can rank | Notes |
|---|---|---|
| Promptwatch | Org-owner OAuth, project/channel map, @mention, thread, DM | Scheduled runs skip if unchanged |
| Otterly.AI | Mention tracking | $29; 4 engines; lag up to 7 days |
| Peec AI | Tracking | $95, 3 models |
| Profound Starter | ChatGPT UI | $99/mo annual, ChatGPT-only |
We will not invent Slack bots for Otterly, Peec, or Profound Starter. If the team already lives in Slack, we rank Promptwatch first for this connector. The other rows are mention dashboards with a login, not a Slack agent mapped to a project.
FAQ
Can any Slack member install the Promptwatch agent?
No. An org owner completes Settings → Slack OAuth. A member who is not the owner will not finish the install for the workspace. The owner requirement is the access-control gate, not a friction bug.
Does the Slack agent crawl ChatGPT on its own?
No. Paid engines stay on the project. Slack reads that stored data. It does not scrape Slack history for strategy, and it is not a second monitoring product. The crawl happens on the project; Slack is a reader.
What happens if I turn on scheduled runs during a quiet week?
Scheduled runs skip when nothing changed. Quiet weeks stay quiet. We will not invent a pager on top of that rule. A scheduled run is a convenience for busy weeks, not an alarm for quiet ones.