Routy

Link monitors

What this is for

A broken affiliate link is broken income — and you usually find out when your dashboard goes quiet, not when the link actually broke. Link monitors check your links on a schedule and alert you the moment something changes: the destination URL flips, the page returns a 404, the offer is pulled, or the landing copy changes.

You can watch the same link from several countries (an offer can be live in the UK but pulled in Germany), and you can capture a screenshot and the page content each time something changes so you can see exactly what the page looked like.

Adding a monitor

You monitor your existing links — the quickest way is from Routy Links:

  1. Open Routy Links.
  2. Find the link you want to watch and click the Monitor action (the radar icon) in its Monitor column.
  3. Give the monitor a name.
  4. Optionally pick one or more locations to check from.
  5. Click Create.

You can also open the Link monitors page (the button at the top of Routy Links) and click Add monitor, then choose one of your links from the list.

Routy monitors the page content daily and captures a screenshot of each change by default. If you need a different schedule, a different thing to watch (final URL, status code, …), or want to turn capture off, your account manager can adjust those for you.

Locations — one monitor per country

A monitor runs from a single location. When you pick several countries while creating a monitor, Routy creates one monitor per country and groups them together under the same link — so you'll see, for example, "Promo link" with US, UK, and DE monitors listed beneath it.

You can target whole countries, or — for the United States — individual states.

Screenshots & page content

Every time a change is detected, Routy saves a screenshot of the page and a copy of its content, which you'll find in the monitor's history. Capture is on by default for your monitors. (Account managers can toggle it; when it's turned on it only applies to changes detected afterwards — past checks aren't filled in.)

Reading the history

Click History on any monitor to see its checks, newest first. Each row shows:

  • When the check ran and which country it ran from.
  • A badge for what happened — a change (URL changed, content changed, became unavailable, …), an error, or no change.
  • The before and after values when something changed.
  • A screenshot thumbnail and a View DOM link, when capture was on for that check. Click the thumbnail to open the full screenshot.

Use the filters at the top to narrow to a change type, or to show only checks with changes or only checks with errors.

Statuses

  • Active — checking on schedule.
  • Paused — checking is stopped until you resume it.
  • Error — the last check failed; open the history to see why.
  • Pending sync — the monitor is being set up and will start shortly.

Common tasks

  • Pause / resume: use the pause or play button on the monitor row.
  • Check now: the refresh button runs a check immediately, outside the schedule.
  • Edit name or location: click the edit (pencil) button. (Schedule, what to monitor, and capture are managed by your account manager.)
  • Stop monitoring: the delete (trash) button removes the monitor.

Troubleshooting

  • I turned on capture but the history has no screenshots. Capture only applies to changes detected after you enabled it, and a thumbnail only appears on checks where a change was actually detected and the capture succeeded.
  • A screenshot or DOM link doesn't open. These links expire after about a week. Reopen the history to load fresh links.
  • My monitor shows "Error". Open the history — the failed check lists the error (for example, the page timed out or returned a server error).

Test AI gate (admin)

When a monitored page changes, the AI gate decides whether the change is material (an offer or destination really changed, the page is down) or incidental (a cookie banner, a login modal, a promo carousel rotating). Only material changes raise a notification. "Test AI gate" lets you see what the gate would decide — without sending anything — so you can sanity-check behavior and tune it over time.

You'll find it in the row actions menu on the Link Monitors list (and per affiliate). Nothing is persisted: no notification, no issue, no history row. The verdict you see is exactly what the live gate would produce on the same inputs.

Three modes:

  • Dry-run latest — from the row's actions menu, "Test AI gate…" → Dry-run. Classifies the monitor's two most recent checks. Best for "this just alerted (or didn't) — why?"
  • Compare any two checks — open the monitor's History, tick any two rows in the timeline, and hit Compare these two. The verdict appears above the timeline. The pair is ordered by time on the server, so click order doesn't matter. Best for "this should have alerted three checks ago — let me prove it." (If a selected check predates snapshot-text persistence, the verdict's "Heads up" note tells you the text isn't stored for it.)
  • Classify text — paste a before/after and, optionally, a one-line "what to watch for." Best for iterating on the classifier with no monitor involved at all. Available to platform admins only; opened from the "Test classifier" toolbar button on the Link Monitors list.

Each verdict shows:

  • Would notify — whether the live gate would have raised an alert. Green means "no alert," yellow means "material but suppressed," red means "alert."
  • Category — what kind of change the AI thinks it was (offer changed, destination changed, page down, cookie banner, login modal, promo rotation, cosmetic, other).
  • Severity — how important the AI thinks the change is.
  • AI summary — a one-sentence explanation in plain language.
  • Before / after — the truncated text the AI actually saw, so you can tell whether a "no change" was the gate being smart or the snapshot itself being empty.

Each test run is a real AI call and is billed under platform LLM spend, tagged separately from production gate calls so it doesn't pollute production metrics.

When to escalate to engineering

If a monitor stays in Pending sync for a long time, repeatedly errors on a link that loads fine in your browser, or never records any checks, contact support so engineering can investigate.

If the AI gate's verdict in test mode looks wrong on a real-world change you can reproduce (cookie banner ringing the alarm, or a real offer change being called "incidental"), share the monitor and the two snapshot timestamps with engineering — that's the kind of feedback the classifier prompt is tuned on.