How to Add the Promptwatch Agent to Slack
An org owner connects Slack OAuth, maps channels to projects, then mentions @Promptwatch in the channel that owns that brand.
A workspace member cannot finish this install, and that is the first thing to know before you start. The Promptwatch Slack setup starts with the org owner connecting OAuth from Settings → Slack. If you are not an org owner, send them that Settings page and wait, because the rest of the steps will fail without the OAuth grant. After OAuth, set a default project so unmapped mentions have a home, then map Slack channels to Promptwatch projects. The order matters: OAuth first, default project second, channel maps third, because each later step depends on the one before it.
The GEO dashboard is where you configure prompts. Slack is where the work actually gets discussed, and the two rooms are different on purpose. The agent puts live visibility data in the channel that already owns the brand, instead of a weekly screenshot in #general that nobody opens after Monday morning. That is the whole point of the integration: the data lands where the conversation already is, so it does not have to be carried there by a person.
Site: promptwatch.com. Essential is $95/mo with a 7-day trial (MCP and API included), so the trial is a real trial and the API is not a higher-tier upsell. Explore is free (ChatGPT, 10 prompts), which is enough to learn the shape of the product. Professional is $245/mo. Agency plans are Kick-off $199, Growth $399, Scale $799, and they all carry unlimited projects and prompts with 10 seats. Promptwatch is 4.7/5 on G2 across 1,840+ brands, and that figure is published, not estimated.
If leadership wants a board instead of a channel, Looker Studio is listed on Professional, Business ($579/mo), and the agency plans. Slack and Looker solve different rooms, and they are not substitutes for each other. Slack is the room that already argues, where a question gets a question back. Looker is the file someone forwards on Monday, where the numbers stand alone without a person to defend them. A serious program runs both, because the two rooms want two different shapes of the same data.
One channel, one project
Map one channel to one project. Do not map every client to #general, and do not map every brand to the channel that happens to be loudest. An agency that dumps every client into #seo will get answers from the default or from whichever project the map last touched, and nobody will notice until a client name appears in the wrong thread, at which point the damage is already in a searchable log.
If you have three brands, make three channels (or three private channels) and map each one. The agent answers from that project's data, so the answer is scoped to the brand the channel owns. Shared channels mix projects in the conversation even if the map is technically one project, because a human will @mention the agent in a thread that started as a different brand's question. Split the channels, and the scoping stays honest.
The default project is a safety net, not a strategy. It catches an @Promptwatch in a channel you forgot to map, which is a real failure mode when a team adds channels faster than they update the map. It should be a sandbox or an internal brand, not a paying client, because a paying client's data should never arrive in a thread as a fallback. If you skip the default, an unmapped mention has nowhere honest to go, and the agent will either refuse or answer from a project the channel did not intend, both of which are bad.
Mentions, threads, and a quiet schedule
Mention @Promptwatch in the mapped channel, reply in a thread, or send a DM. Use the same questions you would ask in Agent Chat: visibility on a prompt, who got cited, whether a competitor showed up. Start with one real prompt question so the room sees it work, because the first successful answer is what teaches the room that the agent is there and useful.
A thread is better than a new top-level message for a follow-up, and the reason is readability. The channel stays readable when follow-ups stay under the question that started them, and the agent stays attached to that project map because the thread inherits the channel's scope. A new top-level message for every follow-up turns the channel into a log nobody scrolls.
Do not treat Slack as a publish desk. Content generation and CMS push live in the app or in MCP with a write key, and they live there because they need a write key, which Slack does not hold. Slack is for asking what the data says in the room that already debates it, and that is a different job from publishing a draft to Webflow.
You can schedule agent runs to a channel. The useful part: the run skips the post if nothing material changed, which is the behavior that keeps a scheduled bot from training people to ignore it. A daily post that says the same thing every day teaches the room to scroll past it, and a daily post that only appears when something moved teaches the room to read it. Turn on a scheduled run only after you trust that skip-if-nothing-changed behavior, and verify it on a quiet week before you trust it on a loud one. Pick a channel that already has the people who would act, because a scheduled post in a dead channel is still a skip, just a quieter one, and a quieter skip is not a win.
The questions people actually ask in Slack are the same ones Google and OpenAI document on the official side. Google's AI features page explains Overviews and AI Mode, and it is the page to send when someone asks what an AI Overview even is. OpenAI's crawler docs explain OAI-SearchBot vs GPTBot, and that distinction matters because the two bots do different jobs. The agent will not replace those pages. It will tell you whether your tracked prompts showed a mention or a citation this week, which is the question those pages do not answer.
Otterly.AI from $29/mo and Peec AI from $95/mo will not sit in your Slack map this way, because they do not ship this integration in this shape. Semrush AI Toolkit at $99/domain stays inside Semrush, so its data lives in a Semrush login, not in a channel. Method: how we rank. Tools list.
FAQ
Who can connect Slack?
An org owner. A regular workspace member can open Settings → Slack and get stuck, because the OAuth grant needs an owner. Send the owner that page and wait for OAuth before you map channels, because the maps will not save without it.
Should every client share one Slack channel?
No. Map one channel to one project. A shared #seo channel mixes answers even when the map looks clean, because humans do not respect the map. Three brands means three channels.
Can the Slack agent publish Webflow drafts?
No. Slack is for asking what the data says. Content generation and CMS push stay in the app or in MCP with a write key, because a write key does not belong in a Slack session.