Policy template
Backlink alert thresholds, written as an incident policy
An alert that fires every week gets muted by Friday. Here is a severity ladder for link health, with the arithmetic behind each line so you can move the numbers.
Policy
Page on referring domains, not on DR
DR is a log of the referring-domain count, so it hides movement. A site with 1,000 referring domains scores 41; losing 200 of them drops it only to 40. Alert on the raw count and use DR as a dashboard colour.
- SEV1 and SEV2 use the referring-domain count.
- SEV3 uses the anchor mix.
- SEV4 is a note in the weekly review.
The ladder
| Level | Trigger | Evaluate | Notify | Owner acts within |
|---|---|---|---|---|
| SEV1 | Referring domains down 20% or more versus the 4-week median | Weekly | Page, then chat | 1 working day |
| SEV2 | Down 8% to 19%, or three weekly declines in a row | Weekly | Chat channel | 3 working days |
| SEV3 | Any one anchor above 15% of all anchors, or +5 points in a month | Monthly | Ticket | 2 weeks |
| SEV4 | Competitor adds 3x your new domains in the same month | Monthly | Digest | Next review |
| INFO | A new spike of domains with a known cause | Weekly | Log line | None |
Worked example (hypothetical numbers)
Picture a store with 1,000 referring domains on 1 October. On 29 October the snapshot shows 780. That is a drop of 22%. SEV1 fires. The DR formula moves from 41 to 40, a change nobody would notice on a chart, even though the store lost 220 domains. Page the owner.
Hence the raw count. One DR point is the visible part of an iceberg, and the lost-link recipe shows how to list exactly which 220 domains left. Often it is one network or one vendor, not a general collapse.
Now the comparison. If the same week brought a Google spam update, such as the March 2026 one, check whether competitors fell too before you blame yourself. Run the same call on three rivals using the competitor diff job. If they are flat, the cause is yours.
Say the owner finds that one resource site, with 190 outbound links, was taken offline. That explains most of the loss. The ticket closes with a note, and the baseline resets to 780 for the next four weeks.
Choosing the baseline
Use a median. The median of the last four weekly snapshots, not last week alone. A single odd week then cannot trip an alarm. Ahrefs, Semrush and Moz dashboards show only the latest number, which is one reason a scripted history is worth keeping.
Small sites need caution. At 50 referring domains, losing 10 is a 20% drop and may be one directory cleaning itself out. Set a floor: no SEV1 below 25 lost domains. The rule is yours to change, but write it down so the next on-call does not argue about it.
Seasonal sites need a second baseline. A tax-advice site gains and sheds links around filing deadlines, so compare October with last October when you can, and with the previous month when you cannot.
Before you accept a threshold
- Take at least four weekly snapshots first, so the median is real.
- Remember the index refreshes about monthly, so two identical weeks are normal.
- Write the owner's name next to every level, or the alert has no pager.
- Re-tune after any migration, relaunch or purge of old pages.
- Keep one quiet level (INFO) so people see that spikes were logged, not ignored.
Tuning after a quarter
After three months, count how many alerts fired and how many needed action. If more than half were noise, widen the band by a few points. If none ever fired, narrow it. A threshold that never trips has no value.
Review the ladder after each Google update window too. The September 2025 and March 2026 spam updates, for instance, are worth marking on your chart so a future on-call reads context beside the numbers.
