When AI changes its
answer about you.
LLM Metrix surfaces brand-impacting changes, sentiment swings, new citations, competitor surges, hallucinations, and delivers them straight to your inbox.
Free forever plan. No card required
Score moved 68 → 72 since the previous scan, after your new docs were indexed.
Gemini described your pricing as "starts at $499/mo". Actual is $49/mo.
Acme.io's share of voice grew from 12% to 19% since the previous scan.
yourdomain.com is now cited for "GEO setup".
Works with the AI engines your customers use
ChatGPT
Perplexity
Gemini
Claude
Grok
Meta AI
DeepSeek
Google AI Overviews
Microsoft CopilotWhy it matters
Know when the story changes, without watching a dashboard.
Per-scan or digest email
Choose a per-scan email, or a daily, weekly or monthly digest, each teammate sets their own.
Outbound webhooks & more
Route alerts to any endpoint with an outbound webhook, piped into the tools you already use.
Six findings, each with its own switch
Score change, competitor surge, new or lost citations, inaccurate claims, sentiment drop and position drift, each derived against your previous scan and delivered with the old figure beside the new. Turn any of them off for your inbox without hiding it from the dashboard.
Focus the noise, pause the storm
Filter alerts to the engines you care about, ChatGPT alone or every engine you track, and pause alert email, Slack and Teams for a day, a week or a month while you work a story. The webhook keeps recording throughout.
On your refresh cadence
Alerts come from completed scans, so they arrive with your refresh cadence, daily on paid plans, weekly on free, plus any scan you run yourself.
Signal, not noise
Competitor-benchmark scans stay quiet, you're only notified about changes to your own brand, so alerts stay meaningful.
Per-scan or digest email
Choose a per-scan email, or a daily, weekly or monthly digest, each teammate sets their own.
Outbound webhooks & more
Route alerts to any endpoint with an outbound webhook, piped into the tools you already use.
Six findings, each with its own switch
Score change, competitor surge, new or lost citations, inaccurate claims, sentiment drop and position drift, each derived against your previous scan and delivered with the old figure beside the new. Turn any of them off for your inbox without hiding it from the dashboard.
Focus the noise, pause the storm
Filter alerts to the engines you care about, ChatGPT alone or every engine you track, and pause alert email, Slack and Teams for a day, a week or a month while you work a story. The webhook keeps recording throughout.
On your refresh cadence
Alerts come from completed scans, so they arrive with your refresh cadence, daily on paid plans, weekly on free, plus any scan you run yourself.
Signal, not noise
Competitor-benchmark scans stay quiet, you're only notified about changes to your own brand, so alerts stay meaningful.
Mechanics
From completed scan to inbox.
- 01
A scan completes
Refresh runs automatically, daily on paid plans, weekly on free, alongside any manual scan. Every finding starts here, because there is nothing to compare before a scan finishes.
- 02
It is differenced against the last one
Score movement, new and lost citations, per-rival share gains, sentiment drift and position changes are computed against the previous scan. A missing input stays silent rather than reporting a fabricated zero.
- 03
Rules gate delivery, not the record
Each finding type carries an immediate, digest or off mode plus an editable threshold. Your switches decide what reaches you; the Alerts page derives every finding either way.
- 04
Channels deliver once each
Email, Slack, Teams and a structured outbound webhook receive the findings through per-channel writes that cannot duplicate on a worker retry. Slack and Teams stay silent when there is nothing to say.
- 05
Findings persist as history
Every derived finding is recorded server-side against its scan, with read state kept per user per occurrence, so acknowledgement survives devices and teammates, and occurrences age out on plan retention.
Who it's for
Who lets alerts do the watching
On-call comms lead treating hallucinations as incidentsAn engine stating something false about you is an incident, not a chart point.
You run communications or escalation and need false claims surfaced with engine and wording attached, soon enough to act.
- Inaccurate-claim findings arrive with the engine named, the prompt behind it and the flagged wording quoted.
- Immediate-mode warnings open the interrupt email on the scan that found them, whatever your digest cadence is set to.
- Pause email, Slack and Teams for twenty-four hours up to thirty days while you work the story; the webhook keeps recording throughout.
- The Alerts page history keeps every occurrence server-side with per-user read marks, so a handoff knows what was already seen.
- 01Connect your Slack incoming webhook in Settings → Notifications so incident findings reach the war-room channel.
- 02Set a score-floor crisis line in Settings → Notifications, it fires on every scan while the score sits under it.
- 03Open the Alerts page when an interrupt lands and work the warning-level findings top-down.
Metrics this role tracks: Sentiment drop · Visibility score · Mention rate · New/lost citations
Open AlertsAgency owner watching every client quietlyYou want the important findings pushed to where the team already lives, not another dashboard to check.
You run several client workspaces and need anomalies reaching your team without adding a monitoring chore to anyone’s day.
- Each teammate configures their own channels and cadence, per-scan email, digest, Slack, Teams or webhook, independently.
- Slack and Teams post only when a scan surfaced a deliverable finding, so the channel never learns to be ignored.
- A digest addresses one person across all their projects; the client-facing artefact is the token-gated report link instead.
- White-label chrome puts your agency’s identity on dashboards, shared reports and email when a sending domain is verified.
- 01Let each teammate pick their own channels and cadence in Settings → Notifications.
- 02Set tighter per-project thresholds for your launchiest client in Settings → Notifications.
- 03Send clients the token-gated link from the Reports page instead of your own digest.
Metrics this role tracks: Share of voice · Competitor surge points · Mention rate · Visibility score
Open Notification settingsGrowth lead catching competitor surges same-dayA rival eating share of voice should reach you before the monthly review does.
You own growth in a category where rivals move visibly and a week of silence costs pipeline.
- The competitor-spike alert fires on whole-scan share of voice, the same percentage the Competitors page charts, with the previous and current figures in the message.
- Default bar: a rival gaining 10 points between scans; the threshold is yours to set.
- Your own share dipping 10 points under the trailing 4-scan average fires under the same rule, so erosion without one big step is caught too.
- Paid plans refresh daily, so a surge surfaces within the day’s cycle rather than at the next review.
- 01Add your top three rivals on the Competitors page.
- 02Set the competitor-surge bar in Settings → Notifications.
- 03Run a manual scan before the weekly review so the trend lines are current.
Metrics this role tracks: Competitor surge points · Share of voice · Score change · Mention rate
Open the Competitors pageExec taking the Monday digest instead of a dashboardOne email a week that says whether AI visibility moved, and what changed.
You want the trend and the exceptions in your inbox, opening the dashboard only when something needs a decision.
- Weekly digests land on Mondays and monthly ones on the 1st, in UTC, fixed gates rather than a picker.
- One digest spans every project you belong to across your workspaces; it is a personal summary, not a client deliverable.
- Each edition covers score, per-engine scores and what changed, so the direction of travel needs no login.
- A warning-level finding on an immediate-mode type emails on the scan that found it, so the weekly batch never sits on an urgent one.
- 01Choose the weekly digest in Settings → Notifications, it lands on Mondays, UTC.
- 02Leave the immediate-mode warnings switched on so anything urgent emails before Monday.
- 03Open the dashboard only when an edition shows a movement worth a decision.
Metrics this role tracks: Visibility score · Score change · Sentiment drop · Share of voice
Support lead catching wrong answers before tickets doAn engine quoting the wrong price or feature writes tomorrow’s tickets today.
You run support or QA and keep fielding tickets that started with an engine stating something untrue about the product.
- Attach pages from your own domain, pricing, docs, help centre, as ground-truth sources, and flagged claims are re-checked against them after the scan.
- A matching quote and source URL land back on the Alerts row when the check completes, so the reply cites your page rather than a guess.
- Immediate-mode inaccurate-claim findings open the interrupt email on the scan that found them, ahead of whatever digest the rest of the team runs.
- Filter deliveries to the engines customers actually ask, a delivery filter only, so the dashboard history stays complete.
- 01Add your pricing page as a ground-truth source under Settings → Ground truth.
- 02Switch the inaccurate-claim finding to immediate in Settings → Notifications.
- 03Review the flagged wording on the Alerts page before updating your macros.
Metrics this role tracks: New/lost citations · Sentiment drop · Mention rate · Visibility score
Open AlertsScope
What should you alert on in AI monitoring?
Two kinds of finding deserve a notification, and they behave differently. Standing states, an engine that mentioned you nowhere in the scan, a rival out-mentioning you across the board, are conditions, and they stay true until they stop. Movements, score change, sentiment slide, a rival’s share jumping, are events, and they exist only because each scan is compared with the one before it. On a project’s very first scan, only the standing states can fire.
Magnitude separates signal from noise, so the thresholds are yours: 5 points of score by default, 10 points of rival share gain, a measurable sentiment shift, a full position of average rank. Defaults are starting points printed beside each control in Settings, not laws. One exception worth knowing: the score-floor alert is a level rather than a movement, it fires on every scan while the score sits under a crisis line you define, and it defaults off.
Be equally clear about what alerts cannot catch. Drift inside a position band, a small slide repeated across many scans, a problem confined to two prompts out of forty, none of these moves a scan-level number past a threshold. Trends catch erosion; alerts catch steps. The durable pairing is notifications for events plus a recurring calendar item against the trend charts.
- Turning a finding type off changes delivery, never the record, the Alerts page derives every finding regardless of your settings.
- A missing input produces silence rather than a zero: no previous scan means no citation diff and no false alarm after an outage.
- Six types carry switches. The zero-mention engine finding does not, because a whole engine going silent is the failure this product exists to catch.
Latency
How fast should brand alerts arrive?
As fast as your measurement, and no faster. An alert is derived from a completed scan and goes out when that scan finishes, so arrival time is the refresh cadence, daily on paid plans, weekly on free, plus any scan you trigger yourself. A tool promising instant alerts is either measuring instantly or bluffing; the underlying engines answer on their own schedule either way.
Speed where it matters comes from routing rather than raw frequency. Warning-level findings on immediate-mode types open the interrupt email on the scan that found them, even if your digest cadence is weekly. Everything else batches: digests run daily, weekly on Mondays and monthly on the first of the month, in UTC. Separating interrupt from digest is the one configuration lever that decides whether the inbox stays usable at scale.
Incident response has its own rhythm, which is why channels pause instead of turning off. While a pause window is active, twenty-four hours up to thirty days, alert email, Slack and Teams stay quiet while you work the story. The outbound webhook keeps recording every finding and the scheduled digest still runs, so nothing is lost during a pause; it is deferred on purpose.
- Competitor benchmark scans notify nobody by default, a benchmark measures someone else’s visibility, not news about your brand.
- Free workspaces refresh weekly, so alert latency on Free is a week by construction rather than a setting to tune.
Routing
Email, Slack, Teams or webhook, which channel for which job?
Split channels by audience. Email is where a finding gets a human owner, the per-scan message or the digest, whichever cadence each teammate chooses. Slack and Microsoft Teams post only when a scan surfaced a deliverable finding, because an all-clear message on every scan is how a channel trains its humans to ignore it. The webhook is for systems: it receives every completed scan’s structured payload, including digest-mode findings a human email omits.
Underneath, delivery is at-least-once with idempotent writes: one deduplication key per channel per scan means a retry cannot send the same alert twice. Settings belong to each person, channels, cadence, thresholds, engine filter, so keeping one client’s findings away from another’s is done with workspaces, or downstream at your own endpoint, which is the intended seam for custom pipelines.
Two refinements complete the picture. The engine filter trims one engine’s findings from your deliveries without hiding them from the dashboard. And per-project threshold overrides, an Agency-plan capability, let entitled workspaces run tighter or looser bars on individual projects, most-specific winning, so a launch project can watch harder than the steady state.
- Webhook payloads carry the event name, brand, score, engine results, the finding messages, and a parallel structured list with type, severity and delivery, so a receiver can filter without parsing the prose.
- Pausing affects the human channels only; the webhook and the scheduled digest continue throughout by design.
FAQ
Alert questions, answered.
What are AI monitoring alerts?
How fast are alerts delivered?
Do alerts work on the free plan?
Can I customise the alert message?
Can I mute or snooze a single alert?
When does the weekly digest arrive?
What happens to alerts after they fire?
How do I set up alerts?
Keep exploring
Related features
The AI era of search
is already here
See what AI engines say about your brand before your competitors do. Start free today. No card required.