On the Recommendations page in Google Ads there is a percentage with a circle that fills up. It is the optimisation score, and it has a striking psychological effect: 62% feels like a failing grade, 94% feels like a well-run account. Neither reading is right. The score does not tell you how well the account performs; it tells you how closely the account matches what Google’s model believes could improve it.
What the score actually measures
According to Google’s documentation, optimisation score is an estimate of how well your account is set to perform, on a scale from 0 to 100%, where 100% means the account can perform at its full potential. It is calculated in real time from the settings, status and statistics of your account and campaigns, the estimated impact of available recommendations, and recent recommendation history.
The key words are estimate and potential. The score does not look at cost per acquisition, ROAS, margin or lead quality in your CRM. It looks at the gap between your current setup and a setup the model considers better. Each recommendation shows how many points it is worth: that is its estimated weight in closing the gap, not a forecast of results.
The point to hold on to. An account can score 95% and lose money on every conversion, or score 60% and be the most profitable one in the portfolio. The score indicates alignment with Google’s proposals, not economic health.
Why dismissing counts as much as applying
Google says so explicitly: the score rises whether you apply recommendations or dismiss them. That is consistent with what the number is: dismissing a recommendation means you have assessed it, and the system records that the proposal is no longer open. In practice, anyone can take an account to 100% in five minutes by rejecting everything.
Two consequences follow. First, the score should never be used as a KPI for whoever manages the account, because it rewards throughput, not the quality of decisions. Second, there is no reason to accept a bad recommendation "for the score". Dismissing it has the same effect on the number and none on the account. A dismissed recommendation can also come back later if the condition that triggered it persists.
Why agencies care about it
One detail worth knowing when you choose a supplier: to keep Google Partner status, an agency needs an optimisation score of at least 70% at manager-account level. It is a legitimate requirement, and it can be met honestly by dismissing the bad recommendations. But it explains why some agencies lean towards accepting more than they should. If whoever runs your account accepts everything, ask them on what basis.
Recommendations usually worth accepting
There is no universal list, but there are families of recommendations with very different risk-to-benefit profiles. The ones that tend to deserve attention are maintenance and hygiene recommendations, which fix objective problems without moving budget or widening targeting:
- Measurement problems. Inactive conversion tags, conversions not recorded for days, missing consent signals. These are nearly always real and should be fixed first, because everything else rests on that data.
- Keywords conflicting with negatives. If a negative blocks a keyword you want live, the recommendation is flagging a genuine configuration error.
- Redundant or non-serving keywords. Cleaning up makes the account easier to read. Do check they are not keywords kept deliberately for coverage.
- Missing or weak assets. Sitelinks, callouts, responsive search ads with few headlines: adding them is generally positive. Accept the diagnosis, but write the copy yourself rather than taking the suggested text as is.
- Policy and approval issues. Limited or disapproved ads need fixing regardless of the score.
Even here, "accepting" means reading the detail. A redundant-keywords recommendation can propose removing the one with the better history; a generated asset can contain a promise your business does not keep.
Recommendations to dismiss, or nearly
The second family is the one that widens spend or scope. The recommendation has a logic — more coverage, more volume — but that logic ignores your margin and often the quality of the traffic that will arrive.
| Recommendation type | When to dismiss it | When it may make sense |
|---|---|---|
| Raise the budget | Low impression share lost to budget, CPA above target, unreliable conversions | A profitable, budget-limited campaign with room to grow |
| Switch to broad match | Few conversions, shaky tracking, nobody watching search terms | Smart Bidding with solid volume and disciplined negatives |
| Change bid strategy | A stable campaign where the switch resets learning for no reason | A changed business goal, or a clearly wrong current strategy |
| Add suggested keywords | Generic or off-target terms for your offer | Relevant variants you genuinely missed |
| Expand targeting | Local offer, niche B2B, an audience narrowed on purpose | High-volume campaigns with a scale objective |
Budget and broad match, without data
The two cases we see most often are budget increases and moving to broad match. Both can be worth plenty of score points, and both are dangerous when solid conversion data is missing.
A budget increase only makes sense if the campaign is genuinely budget-limited and every extra conversion is profitable. If the constraint is the bid target, more budget buys worse auctions. That is the reasoning we set out in the piece on how much budget Google Ads actually needs.
Broad match works when Smart Bidding has enough conversions to learn which queries are worth paying for, and when someone reviews search terms regularly. Without both conditions, it extends coverage onto queries the system cannot yet judge. Before accepting it you need a method for negatives, like the one in the guide to governing search terms.
Auto-apply: the risks and how to switch it off
Auto-applied recommendations let you choose recommendation types that Google will apply on its own, without your sign-off. It is convenient for anyone running many accounts with little time, and that is exactly why it is risky: changes happen while nobody is watching, and if nobody looks for them, nobody notices.
The risk is not that Google acts in bad faith. It is that a general logic gets applied to an account with specific constraints: a keyword paused for commercial reasons, a campaign kept narrow on purpose, a budget agreed with the client. The system does not know those reasons.
How to check it and turn it off
- Open the Recommendations page for the account and go into the auto-apply settings.
- Check which types are selected. Do not assume everything is off: it may have been switched on in the past by another user or during onboarding.
- Deselect anything touching budgets, bids, keywords and targeting. If you want to leave something on, restrict it to low-impact types you understand well.
- Review the history tab in the same settings: it shows recently applied recommendations and when each type was switched on.
- Read Google’s emails. Notices about upcoming or applied recommendations are to be read, not archived.
Our rule. In the accounts we manage, auto-apply is off by default. If a change is good, applying it by hand takes thirty seconds; if it is bad, it can take weeks to notice.
Using change history to audit
Change history records every change to the account with date, level, previous and new value, and the user who made it. Changes made by auto-applied recommendations do not appear under a person’s email address but under a dedicated system user, identifiable by its reference to auto-apply. That is the thread to pull.
A proper audit follows three steps:
- Filter by user and isolate changes not made by anyone on the team. If you find some, you are looking at what the system changed.
- Filter by change type — budgets, bids, keywords, ads — and line the dates up against spend and conversion trends. A shift in trend that coincides with an automated change is the first suspect to check.
- Record and decide. For each meaningful change: keep it, revert it or amend it. Change history shows the previous value, so restoring it is easy; the hard part is knowing it needed doing.
It is the same principle that applies to every automation in the account, from Smart Bidding to the options covered in Smart Bidding Exploration: delegating execution is reasonable, delegating oversight is not.
A twenty-minute weekly routine
You do not need to live on the Recommendations page. You need a fixed slot, always in the same order:
- Change history for the last seven days (5 minutes). Who changed what, paying attention to non-human changes.
- Auto-apply status (1 minute). Confirm the configuration is the one you decided on.
- New recommendations, by family (10 minutes). Measurement and policy first, then maintenance, then budget and targeting.
- An explicit decision on each (4 minutes). Apply, dismiss, or defer with a note. What matters is not leaving recommendations open for weeks: by then nobody knows whether they were ever assessed.
A worked example: an account at 68% has three open recommendations. An inactive conversion tag, a 40% budget increase and a move to broad match. The first gets fixed immediately. The second is judged on impression share lost to budget and real CPA: if CPA is above target, it is dismissed. The third is deferred until measurement has been reliable for a few weeks. The score will rise either way — but for the right reasons.
If you want to know what has been applied to your account over the past months, and by whom, the free audit starts precisely from change history. You can see how we work on accounts and the answers to the questions we get most often in the site’s FAQ.