What Should You Optimize in Google Ads? A Guide to Choosing the Next Change

An account can present dozens of suggestions at once: add assets, change bids, expand reach, adjust keywords or increase a budget. Accepting them all creates activity. It also makes it harder to understand which change helped and which introduced a problem.

The useful question is specific: what evidence makes this the next change worth making?

Begin with one outcome and a usable baseline

Define the result you are trying to improve. For a lead campaign, that could be the cost of a suitable enquiry or a customer acquired. For a shop, it could include contribution after advertising. For a reach campaign, use the objective it was funded to achieve and a separate rationale for that spending.

Record the current result, date range, spend, relevant volume and known reporting delay. Check that the conversion configuration did not change within the comparison. Our conversions and profitability guide explains the commercial measures that a conversion total leaves out.

Avoid optimizing a report you do not yet trust. If purchase values are wrong, improving reported return on ad spend may simply reward the measurement error.

Sort findings into repairs, controls and experiments

Repairs address something verified to be wrong: a broken destination, an incorrect price or an event that fires before submission succeeds. Controls contain exposure outside the plan, such as confirmed irrelevant searches or an unintended location. Experiments test an explanation that is still uncertain, such as whether a different offer would attract more suitable enquiries.

These categories determine what to do next. You do not need an advertising experiment to establish that a checkout returns an error. You do need evidence before declaring that a new headline caused a sales increase.

Use four practical questions: How strong is the evidence? How much relevant spend or business outcome is affected? How reversible is the change? What else could it interfere with? A large estimated uplift with weak evidence can deserve less immediate attention than a modest, verified fault.

Read recommendations as proposals

Google's recommendations documentation describes suggestions informed by account history and an optimization score estimating how the account is set to perform. Evaluate the proposal against your own objective and constraints.

For a budget recommendation, ask whether the campaign is producing an affordable outcome and whether the additional exposure fits the business. For an expansion recommendation, ask what new traffic or inventory becomes eligible. For a bidding change, check the conversion signal and the target's economic basis.

Likewise, Google's Quality Score guidance treats Quality Score as a diagnostic tool, not the account's performance objective. Use a weak component to investigate relevant work. Do not substitute a higher diagnostic score for evidence of better customer outcomes.

Match the change to the campaign's structure

These are examples of questions to investigate, not instructions to make those changes in every account. Some reporting is incomplete or aggregated. Google's search terms guidance, for example, explains why not every query is shown.

Consider interactions too. Excluding a term at a broad scope can affect several campaigns. Changing an account-default goal can affect campaigns that use it. Inspect the affected scope before applying a change intended for one problem.

Design a test that can answer its question

Write a hypothesis that connects a change to an outcome. For example: “Stating the minimum project size on this page may reduce unsuitable enquiries without reducing suitable enquiries enough to make acquisition more expensive.” This is a hypothetical test idea, not a promised effect.

Choose the primary measure in advance, plus a guardrail. In that example, suitable-enquiry cost could be the primary measure, with suitable-enquiry volume and total spend as guardrails. Document the lead qualification rule so it is applied consistently.

Where an appropriate Google Ads experiment is available, use it to structure the comparison. Google's experiment guidance recommends isolating a variable and choosing the success metrics before the test. Check support for the campaign type and proposed change rather than assuming every experiment works everywhere.

Before-and-after reports can still be useful, but seasonality, promotions and other changes can affect them. Describe that limitation. A rise after an edit does not by itself establish that the edit caused it.

Keep a change log and let results mature

Record the exact change, affected campaigns, date, owner and intended result. Include other events such as a promotion, site outage or price change that could affect interpretation.

Review enough completed outcomes to make a useful decision, allowing for the business's sales cycle and Google's conversion delay. Set the observation and spending boundaries before the test starts. There is no useful universal promise that every campaign can be evaluated after the same number of days.

At review, choose among keeping, reversing, continuing within the agreed limit or declaring the evidence inconclusive. A test that cannot distinguish the options has not proved success or failure.

Use AdPilot to act on a finding you understand

The AdPilot Optimize walkthrough explains findings from your account, including wasted search terms, weak ad copy and asset-group work. Review the supporting data, scope and proposed action before using a recommendation.

That workflow can make the next task clearer. It does not supply the business's margins or prove that a lead became a customer. Bring those records into the decision when the objective requires them.

For clearly irrelevant Search traffic, follow the negative-keyword guide. Use the audit checklist to schedule ongoing work. If the issue is that the campaign has not served at all, start with delivery troubleshooting before attempting performance optimization.

Choose the change you can explain, scope and evaluate. If you cannot state the evidence and intended outcome, the next task is usually to improve the diagnosis.

Common questions

Should I accept every recommendation to reach a perfect score?

No. Evaluate each recommendation against the business objective, evidence and spending constraints. A score is not a profitability calculation.

How many changes should I make at once?

Repair verified faults as needed and record them. For an experiment, isolate the change you want to learn about; simultaneous changes make attribution harder.

What if a test produces no clear winner?

Record it as inconclusive. Check whether further observation is affordable and likely to answer the question. Do not manufacture certainty from a small difference.

Download the campaign worksheet