- Critical and warning alerts reach Slack within roughly 10 minutes of being found, and every alert stays in the dashboard feed.
- Info alerts, such as an edited title, are never pushed; they wait in the dashboard feed.
- A firewall block is confirmed with a second request before it's reported.
- A page that times out or returns a 5xx error gets one retry about 20 minutes later before you're alerted.
- Four or more pages with the same problem in one run become one summary alert.
What do the severity levels mean?
Every alert carries one of four labels. The label decides whether you hear about it right away and how much it can affect your AI Access Score.
| Label | Meaning | Delivery |
|---|---|---|
| Critical | AI search crawlers or user fetchers can't reach or read something important | Slack right away, plus the dashboard feed |
| Warning | A change that weakens access or that you should confirm was intended | Slack right away, plus the dashboard feed |
| Info | A change worth knowing about with no direct effect on AI search access | Dashboard feed |
| Fixed | Something that broke earlier works again | In the same message when it arrives alongside other alerts; otherwise in the dashboard feed |
When are alerts sent?
Alerts go out within about 10 minutes of the check that found them, as one message per site that lists everything found together. A site that was alerted in the last 30 minutes waits before the next message, so a problem spread over several checks becomes one or two messages instead of a stream.
Slack messages go to the incoming webhook you paste in Settings > Alerts. Each message names the site, lists the changes with their severity and links to the dashboard. The same screen holds up to 10 email addresses, but alert emails and the Monday weekly report aren't sent yet: email delivery is listed as Coming in the changelog. Until it ships, use Slack and the dashboard feed. The setup guide covers the settings.
Which alerts cover the whole site?
Two alerts track whether your homepage answers at all. They come from the daily site check.
| Alert | Severity | What triggered it | What to do |
|---|---|---|---|
| Homepage stopped responding | Critical | The homepage loaded on the last check and now gives no answer, or your CDN reports the origin is down (Cloudflare 52x or 530 errors) | Check hosting status, DNS and SSL. AI crawlers that try now get nothing. |
| Homepage is responding again | Fixed | The homepage answers again after a failed check | Nothing. Confirm key pages recover on their next check. |
Which alerts cover robots.txt and llms.txt?
These alerts fire when your robots.txt or llms.txt changes between two site checks. Monitor compares which of the 18 agents could fetch your homepage before and after the change, so the alert names the agents that gained or lost access.
| Alert | Severity | What triggered it | What to do |
|---|---|---|---|
| robots.txt now blocks OAI-SearchBot (or other search agents) | Critical | A robots.txt change disallows one or more AI search crawlers | Find the new rule and remove it unless the block is deliberate. Pages they can't fetch drop out of their answers. |
| robots.txt now disallows ChatGPT-User (or other fetchers) | Warning | A change disallows user-triggered fetchers | Decide whether you meant it. OpenAI and Perplexity both say their user fetchers may not follow robots.txt, so a real block needs a CDN rule. |
| robots.txt now blocks training crawlers | Info | A change blocks GPTBot, ClaudeBot, CCBot or other training agents | Usually fine. Google-Extended is the one to double-check: Google's crawler docs say it also covers grounding in Gemini apps. |
| robots.txt now allows (agents) | Fixed | Agents that were blocked are allowed again | Nothing. |
| robots.txt was edited | Info | The file changed but no agent gained or lost access to the homepage | Nothing. Path rules are rechecked against each key page on its next check. |
| robots.txt disappeared | Warning | The file now returns a 4xx status | Restore it. Crawlers treat a missing file as allow-all, so any rules you had stop applying. |
| robots.txt returns a server error (HTTP 5xx) | Critical | The file fails with a server error | Fix the server or CDN response. The robots.txt standard tells crawlers to treat this as "block everything". |
| llms.txt stopped loading | Warning | An llms.txt that worked now returns an error or an HTML page | Check whether a deploy or redirect removed it. |
| llms.txt was updated | Info | The file's contents changed | Nothing, unless nobody meant to edit it. |
The user fetcher behavior comes from the vendors' own docs: OpenAI's crawler page warns that ChatGPT-User, being started by a user, might not follow robots.txt, and Perplexity's says Perplexity-User generally ignores them. For which agent does what, see the AI crawler list.
Which alerts cover firewalls and CDNs?
Firewall alerts compare today's probe requests with yesterday's. They only fire when the previous check was a clean baseline, meaning our normal browser request got your homepage both times.
| Alert | Severity | What triggered it | What to do |
|---|---|---|---|
| Your firewall started blocking (agents) | Critical | AI search agents or user fetchers that loaded the homepage on the last check now get an error, a challenge page or no response, twice in a row | Review bot rules in your CDN or WAF. On Cloudflare, start with the settings in our Cloudflare guide. |
| Firewall now lets (agents) through | Fixed | A probe that was blocked loads the page again | Nothing. |
Probes sent as GPTBot and ClaudeBot never raise a firewall alert, since blocking training crawlers at the edge is a common, deliberate choice. They still show in the dashboard.
Which alerts cover key pages?
Each key page is compared with its own last successful check. The first check of a new page only sets its baseline.
| Alert | Severity | What triggered it | What to do |
|---|---|---|---|
| /page now returns 404 (or another status) | Critical | A page that loaded now returns an error or doesn't respond. Timeouts, 5xx and 429 responses are retried once about 20 minutes later first. | Restore the page or redirect it to its replacement. |
| /page is back | Fixed | A failing page loads again | Nothing. |
| /page was set to noindex | Critical | A robots meta tag or X-Robots-Tag header now says noindex | Look for staging settings, SEO plugin toggles or header rules shipped in the last deploy. |
| /page is indexable again | Fixed | The noindex is gone | Nothing. |
| /page now needs JavaScript to show content | Critical | The raw HTML dropped under 120 words and looks like an app shell | Server-render or pre-render the page. Most AI crawlers don't run scripts. |
| /page lost 60% of its text | Warning | A page with 200 or more words in its HTML lost more than half of them | Check whether copy moved into tabs, accordions or components that render client-side. |
| /page now redirects | Warning | The page lands on a different URL than before, or stopped redirecting | Confirm the redirect is intended and points to the right page. |
| Canonical changed on /page | Warning | The canonical URL is different from the last check | Make sure it still points at the page you want cited. |
| Structured data removed from /page | Warning | A JSON-LD type present before is missing, for example FAQPage | Check the template or plugin that outputs the markup. |
| Title changed on /page | Info | The page title is different | Nothing, unless the change wasn't planned. |
| /page is blocked for (agents) | Critical | A robots.txt path rule stops AI search agents or user fetchers from fetching this page | Narrow the Disallow rule so it doesn't match pages you want cited. |
Which alerts use bot visit data?
One alert needs connected logs: (agent) has not visited in 3 days, a Warning. It fires when an AI search agent or user fetcher averaged at least three visits a day over the previous two weeks and then made none for three days.
A sudden silence often means a block that the probes can't see, such as a rule that only matches the vendor's IP ranges, an outage on the paths that agent visits, or a change on the engine's side. Each agent can trigger this alert at most once a week. Training crawlers are excluded because their schedules vary too much to alert on.
How does Monitor avoid false alarms?
Several rules keep one-off glitches out of your inbox, because an alert you learn to ignore is worse than none.
- A probe that flips to blocked gets a second request about 1.5 seconds later. If that one loads, no alert is raised.
- IP-verified visits logged for an agent during the previous three days outrank a blocked probe, since your CDN may let the real crawler through by IP.
- If our request for robots.txt or llms.txt gets no answer, or gets an HTML or challenge page, the last good copy is kept instead of reporting the file as changed or deleted.
- When bot protection blocks our browser request, the last clean probe results are kept so tomorrow's comparison still means something.
- A soft 404 that answers llms.txt requests with an HTML page counts as "not found", never as "stopped loading".
Frequently asked questions
Why didn't I get an alert on the first day?
The first check on a new site is the baseline, so it can't show a change yet. Problems that already exist appear as findings in your report and in the Who can get in panel. Alerts start from the second check, when there's a previous result to compare with.
Can I turn off a type of alert?
Not per alert type. You choose where alerts go (one Slack webhook per site, with email to follow once email delivery is live), and severity decides timing: critical and warning alerts go out straight away, while info alerts are only listed in the dashboard feed. Every alert stays in that feed.
Why did I get one alert for many pages?
When one run finds the same problem on four or more key pages of a site, Monitor sends one summary instead of one alert per page. That usually means a site-wide cause, such as an outage, a template change or a robots.txt edit. The summary lists up to 10 paths and one full example.
Does Monitor alert me when a training crawler gets blocked?
Yes, as an Info alert when robots.txt starts blocking training crawlers such as GPTBot or CCBot. It's low severity because training crawlers don't decide AI search citations. Google-Extended is the exception worth noticing, since Google says it also controls whether Gemini apps can use your pages for grounding.
How fast will I hear about a problem?
Site checks run once a day and each key page is rechecked about every 24 hours, so most changes are caught within a day. Critical alerts and warnings follow within about 10 minutes of the check. Click Check now after a deploy if you want an answer within minutes.