01
Waste
Spend without conversions and without buying intent for your services. Exact negatives when the rules and health checks allow.
Optimizer
Keywords get verdicts. Search terms sort into waste, winners and mismatch. No language model decides spend. Growth and judgment calls wait for you. Learning and broken tracking stop the machine.
The problem
Most “AI optimisers” either chat about the account or push changes and explain afterwards. The middle ground is empty: deterministic rules that know when not to touch the money.
Humans miss mismatch. A search term looks like your keyword. The caller wants something adjacent. You pay. The report still says the keyword is fine.
Or the opposite: someone pauses a keyword that was about to convert, because the lag had not landed yet. Or they act inside a Smart Bidding learning window and reset the strategy for nothing.
Search terms
Waste is often obvious. Winners are pleasant. Mismatch is where accounts bleed quietly, and where a human still has to decide.
01
Spend without conversions and without buying intent for your services. Exact negatives when the rules and health checks allow.
02
Converting queries that are not yet keywords. Proposed as new Exact keywords. You approve before they enter the account.
03
Shares a real phrase with a keyword, means a different need. Always a proposal. Human judgment required.
What it does
01
Keywords already in the account get a verdict. Search terms that triggered ads get classified. Different engines. Same rulebook. No language model chooses where money goes.
02
Each keyword is judged against the ad group’s own cost-per-lead baseline. Pause when a fair trial shows zero conversions with confidence. Trim when a keyword converts but runs too expensive. Hold when tracking is unhealthy or conversions may still be landing. Feed winners that convert at or below the baseline when they are starved of reach.
03
Queries that spend without converting, and that show no genuine buying intent for what you sell, can become exact negatives. Clear junk does not need a meeting. Tracking must be healthy first. Learning and post-change holds still apply.
04
Queries that convert well and are not yet keywords become proposals to add. Growth waits for you. After approval, the new keyword enters the same verdict engine on later runs.
05
This is the subtle one. The query shares a real multi-word phrase with a keyword you already run, but the need is different. It looks related. It is not on-theme. Auto-tightening or auto-negating would change targeting without consent. Those stay as proposals so you choose: tighten the match type, or negative the term.
Example shape: your keyword is “emergency plumber”. The search term shares the phrase but wants a different job. Auto-negating could cut real coverage. Auto-tightening could miss the leak. The product puts the choice in your queue with the data attached.
06
If conversion tracking looks broken or unknown, judgment stops. While Smart Bidding is learning, or after a campaign or bid strategy change, autonomous changes hold. Bid increases for starved winners stay proposals. High-impact pauses escalate for approval.
The rules
Rule 01
Classification is account arithmetic and fixed rules. A model does not decide pauses, trims, negatives or bids.
Rule 02
Keywords are judged against your own cost-per-lead baseline. Nothing is paused on a small sample or while conversions may still be landing.
Rule 03
Nothing autonomous runs while Smart Bidding is learning, or in the period after a campaign or bid strategy change.
Rule 04
Related-looking queries with the wrong need are proposed, not auto-rewritten. You choose tighten or negative.
Rule 05
When conversion health is degraded or unknown, the optimizer stops moving money and waits.
What you see
Get started
Connect read-only. See waste, winners and mismatch before you approve a change.