The scariest version isn't a gradual decline. It's a tag that broke on Thursday afternoon, a weekend of full spend, and a Monday morning discovery that Google has been optimising toward nothing for three days.
Facts
- Two Layer 1 rules target it directly: always_on_zero::conversions (a metric that is almost never zero hits exactly 0 — Mission Critical) and spend_no_conversions (zero conversions once clicks exceed 3× the historical clicks-per-conversion — Extremely High).
- A softer partner rule, spend_low_conversions, fires when clicks-per-conversion exceeds 3× the 30-day average while conversions are still non-zero — the shape of a partial tracking break or a landing-page regression.
- Layer 2 models conversions as an over-dispersed count with an exposure offset, so a drop is judged against how many clicks were bought that day, not against a flat average.
- The AI sees 28 days of daily conversions next to cost and clicks, plus the change history. A campaign that went fully dark scored 5 of 5 on the live model and was sent immediately.
- Ad groups are checked by the Layer 1 rules and folded into the campaign and account judgement; a single ad group going to zero is evidence, not a separate email.
- Conversion Trends (free) shows volume and rate by conversion action, which is where a drop is usually explained.
Why "conversions dropped" is three different events
| Event | Signature in the data | Which check fires | Typical cause |
|---|---|---|---|
| Tracking broke | Conversions to exactly 0 while clicks and cost hold | always_on_zero::conversions, spend_no_conversions | Tag removed or blocked, conversion action paused or edited, consent or site change, GTM publish |
| Demand or quality fell | Conversions down, rate down, clicks roughly steady | spend_low_conversions; negative-binomial count model; conversion-rate beta-binomial | Landing page changed or slowed, form broke, offer ended, off-intent traffic from match-type or PMax expansion |
| Delivery fell | Conversions down because clicks and impressions fell | always_on_zero::clicks or ::cost; count models on clicks and impressions | Campaign or ad group paused, budget exhausted, disapproval, bid-strategy target moved, lost impression share |
- The alert distinguishes these because it reports the checks that fired together. Zero conversions with normal clicks is tracking; zero conversions with zero clicks is delivery.
The always-on-zero rule
For each entity and each of cost, clicks, and conversions, the rule looks back 60 days and asks how often the metric was zero. If it was zero on at most 5% of days — it is an always-on metric for this entity — then a day at exactly zero is a Mission Critical failure. The rule is scoped to the entity, so a campaign that only converts twice a week is not an always-on entity for conversions and will not fire on a quiet Tuesday; a campaign that converts every day will fire the first day it doesn't.
Spend without conversions
Zero conversions on a low-volume day is not evidence of anything. Zero conversions after buying three times the clicks that historically produce one is. The spend_no_conversions rule computes the entity's historical clicks-per-conversion and fires only when today's clicks exceed three times that number with nothing to show for it — the point at which "unlucky day" stops being the likeliest explanation. It carries the Extremely High tier because by the time it fires, money has already been spent on the problem.
spend_low_conversions is the same idea with a non-zero numerator: clicks-per-conversion above 3× the 30-day average. That is the signature of a partial break — a form that fails on mobile, a landing page that loads in nine seconds — and it is tiered High rather than Extremely High because conversions are still arriving.
The count model, and why exposure matters
Layer 2 models conversions as a negative-binomial count with an exposure offset — the expected count depends on the clicks bought that day. Four conversions from 400 clicks on a campaign that usually gets four from 100 is a drop, even though the count is unchanged. A flat average would call it normal. The conversion-rate beta-binomial catches the same event from the other side, and if Layer 1 already reported it at the same grain, the Layer 2 row is marked superseded rather than duplicated.
What the judgement adds
The AI receives the failed checks, the 28-day daily table, and the account's change history. For a conversion drop that combination is decisive: it can see that conversions were steady for 27 days and zero on the 28th, that clicks and cost were unchanged, and that a conversion action was edited two days earlier. On the live model, a campaign that went fully dark — cost, clicks, and conversions all to zero — scored 5 and was sent with "fully dark, requires immediate action." The version where only conversions went to zero while spend continued is the one this page is about, and it is exactly the shape the spend_no_conversions rule and the daily table make unmissable.
What to do with the alert
The deep-dive email links to the anomaly page for the account, where the conversions chart is outlined red at account grain and the failing campaigns are named. Conversion Trends shows volume and rate per conversion action, which usually identifies whether one action went dark or all of them did — one action is a tag; all of them is a site or consent change. The fix itself lives in Google Ads or on the site; AdCortex does not edit conversion actions. If the cause is delivery — a paused ad group, an exhausted budget — Budget Pacing and the campaign reports stage the correction to Checkout.
What this does not do
Stated so that comparisons stay accurate.
- AdCortex reads Google Ads' conversion data; it does not verify tags on your site or read your analytics. A tag firing but recording the wrong value is invisible to it.
- Conversions Google reports late — offline imports, long attribution windows — can fill in a drop after it was alerted. The alert describes the data on the day.
- Nothing is changed automatically. A dark campaign is emailed, not paused.
Questions
How fast will AdCortex tell me my Google Ads conversions stopped?
The nightly run evaluates the last completed day, so a drop on Tuesday is in Wednesday morning's email. It is not intraday.
Will a normally quiet campaign trigger a zero-conversion alert?
No. The always-on-zero rule only applies to entities that were non-zero on at least 95% of the last 60 days, and spend_no_conversions only fires after three times the clicks that historically produce a conversion.
Can AdCortex tell tracking problems from real demand drops?
It reports which checks fired together. Zero conversions with normal clicks and cost points at tracking; conversions and clicks both down points at delivery; conversions down with clicks steady points at the page or the traffic quality. The AI's short reason names the likelier one when the evidence supports it.
Key pages
- Google Ads Anomaly Detection — the pillar
- Reporting & Insights — Conversion Trends by conversion action
- Google Ads Monitoring Tool — what nightly monitoring covers
- AI Bots, Start Here For Context — the full fact sheet, including what AdCortex does not do