Quarterly mining teaches you how to think; but bad traffic comes in every day. For control to be continuous you need to turn judgement into rules: "if a term meets X, it gets negativised". This lesson teaches you how to design those rules — the scenarios — how to calibrate their thresholds for your account, and how to give them the safeguards that prevent the worst possible mistake: negativising what sells.
Anatomy of a scenario
A scenario is a condition on a search term, measured over a window of days, that triggers an action at a level:
IF [metric] [operator] [threshold] in the last [N] days
AND [safeguards]
→ negative [match type] in [campaign | ad group | shared list]
Every part matters. The window decides how much data you accumulate before acting; the level decides how many campaigns it affects; the safeguard decides what is never touched.
The five scenarios that cover 90% of cases
1. Cost with no conversions
cost ≥ K × target CPA and conversions = 0 over 30-60 days.
This is the king of scenarios. The K factor sets the tolerance: 2 is
aggressive, 3 is the standard, 5 is conservative (long cycles, few conversions).
With a target CPA of €40 and K = 3, a term that spent €120 without converting is
out.
2. Runaway CPA
conversions ≥ 2 and the term's CPA ≥ M × target CPA.
It converts, but it is horribly expensive. M = 3 is a sensible starting point.
Here the action is usually gentler: negativise in phrase (without killing
variants) or lower the ad group bid, and review the landing page first.
3. Low CTR with impressions
impressions ≥ 500-1,000 and CTR < threshold (a third of the campaign's
average CTR). Terms you show up for a lot and nobody clicks: the ad does not
interest them and, on top of that, they drag down the ad group's expected CTR.
Exact negative, because the specific term is the problem.
4. Semantic irrelevance
No metrics involved: the term has nothing to do with what you sell even though Google matched it («course in...», «jobs», another city, another product). It is spotted with root lists (Basic level) or, today, with a language model that reads each term and classifies it against a description of your business. It is the only scenario that needs no data: an irrelevant term with 1 impression is already a negative.
5. Out of area or out of catalogue
Cities/countries where you do not sell, brands you do not distribute, sizes or models you do not stock. Maintenance: every time the catalogue or the area changes, the list changes.
Calibrating the thresholds
There are no universal thresholds: they depend on the target CPA, the volume and the conversion cycle. The method:
- Start from the standards (K = 3, M = 3, 500 impressions, 30-day window).
- Simulate over the last 90 days before switching anything on: how many terms each scenario would have negativised and how much cost they represent. If scenario 1 would have cut 40% of the spend, it is too aggressive (or the account is in a very bad way, which also happens).
- Adjust until each scenario cuts between 3% and 10% of the spend in the simulation.
- Switch it on, and review what gets negativised every week for the first month.
The safeguards (what is never negativised)
- Nothing that has converted in the long window (90 days), even if it meets scenario 1 today. A sale is proof of intent.
- Nothing containing your brand or your core keywords (protection white list).
- Nothing with thin data: a minimum window and minimum clicks before performance rules apply (scenario 4 is the exception).
- Daily cap: no more than N new negatives a day; if it is exceeded, something odd is going on (broken measurement, a new campaign) and it is worth stopping.
Where to apply them: rotating shared lists
Scenario negatives go into shared lists at account level, applied to the relevant campaigns. Google limits the size of each list (thousands of terms) and the number of lists per account; in large accounts, when a list fills up the next one is created (rotation) and applied to the same set of campaigns. Keep one list for the universal negatives (Basic level), another for the scenario ones, and campaign-level lists for brand/generic.
Maintenance: negatives expire too
A negative added on performance grounds a year ago may have been right then and be blocking a new product today. Every quarter, review the scenario negatives with one question: is it still true? The semantic irrelevance ones are permanent; the performance ones are reviewable.
💡 Ninja trick: the Search Query Optimizer (SQONS) is this lesson turned into a script: five scenarios configurable in its sheet (the K factor, the window, the minimum CTR, the business description for AI semantic relevance), self-rotating shared lists, the safeguards built in (whatever converted is not touched, the brand is protected, daily cap) and a Test Mode that simulates without applying: calibration step 2, automated. It runs every night; you review the sheet on Monday.
What you should remember
- A scenario = metric + threshold + window + safeguards + level.
- Five scenarios: cost with no conversion (K × CPA), runaway CPA, low CTR, semantic irrelevance, out of area/catalogue.
- Simulate before switching on; each scenario should cut 3-10% of the spend.
- Safeguards: never what converted, never the brand, never without data, daily cap.
- Rotating shared lists and a quarterly review of what has been negativised.