How do I get alerts when my brand is mentioned in ChatGPT answers? Our client setup
How do you get alerts when your brand is mentioned in ChatGPT answers? How we set up scheduled monitoring, Slack briefings and digests for clients on Promptwatch.
Clients often ask for "a Google Alert for ChatGPT". We understand the instinct, and we have to reset it gently. ChatGPT does not publish what it tells individual users, so there is nothing to subscribe to. Every alert you can have is built the same way: a tool runs a fixed set of prompts on a schedule, records the answers, and tells you when they change.
That means the quality of the alert depends on the quality of the prompt set and the schedule. We run client programs on Promptwatch, and this is how we set up monitoring and notifications for them.
Decide what you want to be told
"Tell me when we are mentioned" is too vague, and it produces noise. We ask clients to pick from a short list:
- We appear or disappear for a core prompt.
- A named competitor overtakes us in share of voice.
- We gain or lose a citation from an important page.
- Sentiment in a description changes.
- A crawler error appears on a page that matters.
- We are about to hit a usage limit.
Most clients pick two or three. Everything else goes into a weekly summary.
What Promptwatch gives us
Promptwatch runs the prompt panel daily per model, country and language, so the underlying data is fresh every day. For notifications we combine several features.
Scheduled Chats. Agent Chat is a conversational analyst over the live data. It can run a workflow on a schedule and deliver the result in the app or in Slack. We use the 7-day GEO briefing skill as a Monday-morning post in the client's Slack channel. Slack briefings are not available on white-label workspaces, so we plan around that when a client needs the white-label setup.
The Action Board and its weekly digest. A Discovery agent examines the data weekly and suggests up to five evidence-backed actions per run, and a digest email goes out to the team. For many clients this is the only notification they need.
Scheduled PDF reports. Any dashboard can be exported as a PDF on a schedule, which suits stakeholders who will never log in.
Usage alerts. Promptwatch warns at 85% of a quota, which stops tracking from halting unexpectedly in the middle of a month.
Weekly site audit emails. The weekly Site Health audit includes 21 checks, a health score and what changed since the last run. Agent Analytics shows crawler errors with failure explanations, which covers the crawler condition.
API and MCP for custom rules. There is a REST API v2 and a hosted MCP server with 110+ read and write tools. For a client who wants an alert on a specific condition, such as Share of Voice falling under an agreed threshold, a small script can pull the data and post to a channel.
What we are upfront about
We tell clients that Promptwatch's published feature list does not include a push notification that fires the moment a single answer names the brand. It offers scheduled monitoring, digests, scheduled chats and the means to build custom triggers. For most programs that is enough, because the number we care about is a trend across many runs, not a single answer.
If a client insists on real-time notification, we say so. Some social listening tools offer real-time email and Slack alerts and have added AI visibility modules. We have not found them to replace a dedicated prompt panel for explaining why something changed, but they may serve a team that mainly wants a ping.
Why we avoid alert-on-every-change
AI answers vary from run to run. An alert on every changed answer would fire constantly and train everyone to ignore it. So we alert on trends. For example, we might flag a drop in Share of Voice that holds across several days, or a crawler error that persists for more than a day. A one-day swing is noise until a second run confirms it.
Our default setup
- A stable prompt set of 30 to 50 questions, grouped by topic and persona.
- A weekly Slack briefing through Scheduled Chats.
- The Action Board digest to the working team.
- A monthly PDF to leadership.
- One custom rule for the single condition the client cares about most, built on the API or MCP.
- A review of all rules after the first month, to remove anything that never led to action.
What alerts cannot do
They cannot tell you what a real user saw in a private conversation. They cannot guarantee that a mention will continue. They tell you about your sample of prompts, and that is enough for most programs. The real value of an alert is a faster reaction to a trend, such as a crawler error or a competitor surge, not the thrill of seeing your name.
Where to start
Build the prompt set first and add notifications second. Promptwatch's free Explore plan covers 10 prompts on ChatGPT, which is enough to see what the data and the reports look like. Scheduled Chats and the other features come with the paid plans.