How to Set Up Monitoring and Alerts for Your AI Visibility

An AI visibility report tells you where you stood on the afternoon it ran. That is genuinely useful once, and almost worthless as a habit — because the thing being measured changes underneath you.

Models get updated. Competitors publish. A page that earned three citations gets outranked by something newer. None of it announces itself, and none of it shows up in your analytics.

Monitoring is what turns a snapshot into a signal.

Why One Report Is Not Enough

The gap between a report and monitoring is not frequency. It is what each can tell you.

Step 1 — Turn Monitoring On

The Monitoring screen is described as automated tracking of your AI visibility, and it needs two things before it can do anything: a site, and prompts to watch.

If it tells you to add a website to start monitoring, that is the whole message — nothing is configured yet. And with no tracked prompts, there is nothing for a scheduled run to check, so add those from Prompt Research first.

Once running, the header carries four figures worth knowing by heart:

Step 2 — Choose a Frequency

Three cadences are available — weekly, daily and hourly — with the ceiling set by your plan. A frequency above your tier is not selectable, and asking for one returns a plain message rather than failing silently.

Each scheduled run costs one refresh credit. That single fact should drive the decision more than ambition does.

Reading the Status Badge

Monitoring shows one of a few states, and the paused ones say why:

The Schedule Grid

The Monitoring Schedule section shows one cell per day, in four states: Ran, Scheduled, Missed, or quiet.

It takes about two seconds to read and answers the question nothing else on the screen does: is this thing actually running? A row of Missed cells is the earliest warning that coverage stopped — earlier than you would notice from silence in your inbox.

Recent Changes

The Recent Changes feed is what monitoring produces. Every entry is a difference against the previous run, in one of four kinds:

You now appear where you did not. Worth reading for the cause — if it followed a page you published, you have just learned what works on that platform.

The urgent one. Something you had is gone, and the cause is recent enough to find. Start from the page that used to earn it.

Still cited, but earlier or more prominently in the answer. Usually follows a content improvement, and confirms it landed.

The quiet one. You are still there, so nothing looks broken — but you are drifting toward the edge of the answer, and a repeated drop usually precedes a loss.

Each entry names the platform and the prompt, which is what makes it actionable. A change without those is a statistic; with them it is a page to open.

Step 3 — Configure Alerts

The Alerts & Notifications screen decides which of those changes are worth interrupting you for. Six switches:

Turning all six on is the fastest way to stop reading any of them. A sensible starting configuration is narrower:

Citation lost and crawler access issues. Both mean something you had is gone, and both are time-sensitive — a crawler block left for a month costs you every citation in that period.

New citation detected. Positive signal, and genuinely useful while you are shipping content, because it tells you which changes worked.

Score changes at plus or minus five points. Early on, this fires constantly and none of it is meaningful — five points on a young account is variance in how assistants happened to phrase an answer.

Competitor gains and losses and content freshness warnings. Both are useful, neither is urgent, and both are better read as a weekly digest than as interruptions.

Where Alerts Go

Three destinations: Email, Slack and Webhook. They suit different things.

What to Do When One Fires

The alert names a platform and a prompt. Read the current answer for that prompt first — who is being cited instead of you tells you most of what you need.

Almost every loss traces to one page. Check whether its facts have aged, whether a competitor published something more complete, or whether something changed technically.

An access issue is not a content problem and should not wait for the content cycle. It usually follows a deploy, a CDN change or a robots.txt edit, and it is fixed in minutes once found.

One position drop can be phrasing. Two in a row is a trend. Reacting to every individual movement produces churn, not improvement.

Common Mistakes

It empties your credits and pauses monitoring — usually right before something worth catching.

Six alert types firing weekly trains everyone to ignore the one that mattered.

No alerts can mean nothing changed, or that monitoring has been paused for three weeks. The status badge and the schedule grid tell you which.

Every tracked prompt costs part of a run. If a movement on it would not change anything you do, it does not belong on the list.

Frequently Asked Questions