Ad platforms increasingly include AI assistants that suggest changes: raise budgets, widen audiences, accept recommendations. Some suggestions are useful, others mainly serve to spend more. Knowing how to assess each recommendation avoids accepting everything or rejecting everything.
The essentials in a definition
AI assistants inside ad platforms recommend changes to budget, structure, creatives and targets. Some recommendations help; others aim to increase spend or feature adoption, and do not always suit the business. Assess each by expected effect on your return, test on a small scale and keep human responsibility, in line with the oversight principles of the AI Act. Options such as Performance Max and Smart Bidding are often suggested.
What assistants do
They analyse the account and propose actions: new terms, assets, budgets, bidding strategies, structure. Some apply with one click. They usually come with an “optimisation score”, which indicates how far the account follows the recommendations.
They have a stated aim: improving performance. But part of the platform’s incentives coincides with more spend.
Recommendations that usually pay off
Fixing errors (disapproved ads, badly configured conversions, broken links), adding missing assets, improving data quality, resolving budget limitations when return is good. These tend to bring clear benefit and low risk.
Always check the account’s context before accepting.
Assessing platform recommendations
- Review recommendations every week
- Classify: accept, test, postpone or reject
- Test the riskier ones on a small scale
- Record the reason and the result
Recommendations to analyse carefully
Budget increases without evidence of return, expansion of audiences or terms that may dilute quality, strategy changes that restart learning, addition of campaigns that overlap with existing ones. Ask: what changes? What is the risk? How do I measure the effect?
Do not use the optimisation score as a target: it measures adherence, not results.
A platform recommendation is a suggestion, not an instruction.
A simple process
Review recommendations weekly. Classify them as accept, test on a small scale, postpone or reject, and record the reason. Test the riskier ones with experiments. Compare results. Over time you learn which types of recommendation work in your account.
Avoid applying everything in bulk.
Decision table
| Type of recommendation | Assess | Typical action |
|---|---|---|
| Increase budget | Is there budget limitation with return? | Apply in steps if so |
| Change bidding strategy | Are there reliable data and tracking? | Test with an experiment |
| Add broad keywords | Are there exclusions and control? | Only with negatives and review |
| Accept new generated assets | Do they comply with brand and policies? | Review one by one |
Common mistakes to avoid
- Applying everything automatically. Some recommendations raise spend without improving return.
- Chasing the score. The optimisation score is not a business objective.
- Ignoring all of them. Some are useful and fix real flaws.
How to measure the result
Record each recommendation, the decision and the measured effect, and track the proportion of useful recommendations.
Limits and exceptions
Recommendations are generic and based on average patterns, and may not consider your margin, capacity or market. Applying them without criteria can raise costs and reduce control. Conversely, ignoring everything may leave errors uncorrected. Balance, recording and testing are the best defence.
Illustrative example
Illustrative example: an account receives ten recommendations. The team classifies them by expected impact and risk, applies two low-risk ones (fix disapprovals and add a useful asset), tests a third in an experiment and ignores the rest with written reasons. The account improves in a controlled way, and results are recorded for comparison.
Frequently asked questions
Should I follow the optimisation score?
Not as a goal. It measures adherence to recommendations, not results.
Which recommendations can I accept without fear?
Error fixes and data improvements, with low risk.
How do I test a recommendation?
With an experiment or by applying it to part of the account.
Should I follow the platform’s recommendations?
Assess each by its expected effect on your objectives; some help, others suit the platform more.
What is the optimisation score?
An indicator of adherence to recommendations, not of business performance; use it only as a clue.
Treat platform assistants’ recommendations as hypotheses: assess, test on a small scale, record and decide based on your objectives.
Sources and recommended reading
- Google Ads Help: Performance Max campaigns
- Google Ads Help: Smart Bidding
- Artificial Intelligence Act (AI Act)
Last editorial review: October 2026. Informational content; always confirm current conditions in the official sources before deciding.