Optimization Score in Google Ads 2026: What to Apply

Once a week someone sends a screenshot: “optimization score is 68%, we need to fix that.” A month later the score is 94%, spend is up 50%, and orders are up 12%. On paper the account is optimized. In reality, margin was traded for spend growth. The optimization score in Google Ads measures how much you agree with Google’s suggestions — not how good your account is — and knowing the difference is a baseline skill in 2026.

Below: how the score is actually calculated, why 100% is unreachable without expanding spend, which recommendations are worth applying almost every time, which belong in the bin, and why auto-apply remains the most dangerous checkbox in the interface.

How the optimization score in Google Ads is calculated

The 0–100% score reflects how fully, in Google’s model, your account uses the platform’s available features. It isn’t computed from results — it’s computed from the list of active recommendations. Each carries a weight, and the “more important” one is in Google’s model, the more it drags the score down while it sits unactioned.

The scoring mechanics:

Apply a recommendation — the score rises by its weight.
Dismiss a recommendation — the score also rises, usually less, and the item leaves the list (sometimes returning later).
Ignore it — nothing changes and the item keeps sitting there.

First practical takeaway: an unactioned recommendation hurts the score more than a deliberately dismissed one. If you’ve decided a suggestion doesn’t fit your account, dismiss it explicitly — the score improves and the list stops accumulating clutter.

The second takeaway is less comfortable. The heaviest-weighted recommendations are the ones that expand spend: raise budgets, broaden match types, launch Performance Max, add similar audiences and locations. Which means 100% is arithmetically reachable only by agreeing to nearly everything that increases spend. A score of 100 doesn’t mean “perfect account.” It means “account that agreed with Google.”

What score is actually healthy

A sensible band for a hands-on account is 70–85%. In that range you’ve typically applied everything sensible and dismissed everything that conflicts with your economics. Below 60%, go look — you may be missing something genuinely useful. Above 95% is almost always a sign that someone is applying in bulk, spend-expanding items included.

Optimization score is an agreement metric, not a performance metric. It has no relationship to your margin, and Google doesn’t pretend otherwise.

Six categories, and what to do with each

Google groups suggestions into six buckets: bidding, budgets, keywords, ads and assets, targeting, and campaign structure. A more useful split is by what each does to your money.

Class	What it does	Share of the list	How to handle
Spend-expanding	Raise budget, broaden match, launch a new campaign type, add audiences and geos	Majority	Experiment or dismiss
Genuinely efficient	Add negatives, pause non-converting keywords and ads, fix tracking	Minority	Apply almost always
Neutral	Add sitelinks, callouts, another RSA variant	Moderate	Apply if it matches your asset plan
Technical	Fix disapproved ads, feed issues, conversion tracking errors	Small but critical	Apply immediately

What to apply almost every time

Negative keywords surfaced from search terms. Google flags queries burning budget without conversions. Reviewing and adding them is direct savings — though review rather than bulk-apply: long-consideration queries occasionally end up in that list. The method is in the guide to negative keywords and search terms.
Pausing keywords and ads with volume and zero conversions. Here Google is on your side: those items are demonstrably wasting money.
Technical fixes. Disapproved ads, conversion tracking errors, product feed problems — apply immediately, no deliberation needed.
Relevant assets and extensions. Sitelinks, callouts, structured snippets improve visibility without raising bids — provided they actually describe your business rather than being generated filler.
Promoting consistently converting search terms into their own keywords. Worth doing when it buys you control over bid and copy.

What to dismiss almost every time

Broadening existing keywords to broad match. The most frequent recommendation and the most expensive mistake. Broad match can work — but only alongside a trained conversion model and a dense negative list, and that’s a deliberate strategy rather than a side effect of a button. The conditions under which it’s justified are in the piece on keyword match types.
Budget increases “to capture more conversions.” The suggestion models uplift in Google’s terms and never looks at your incremental economics. Budget decisions belong on the return curve — how to read it is in the guide to Performance Planner and budget planning.
Auto-generated Performance Max “based on your existing campaigns.” PMax is a strategic decision with its own structure, signals, and exclusions — not a button. What to settle first is in the complete Performance Max guide.
Similar audiences and geo expansion with no hypothesis behind them. That’s traffic dilution presented as reach growth.
One-click bid strategy changes. Changing strategy restarts learning and rewrites the campaign’s economics — validate it with an experiment instead. Strategy trade-offs are compared in the breakdown of Google Ads bidding strategies.
Lowering target CPA or ROAS “to grow volume.” The phrasing is contradictory but such items appear; each is an economics question, not a score question.

Decoding the wording: what each recommendation actually does

Recommendations are carefully phrased and nearly always quote an expected uplift. The catch is that the uplift is modeled in Google’s terms and almost never comes with the price of that uplift. Here are the common phrasings and what they mean in practice.

Wording in the interface	What actually happens	Verdict
“Use broad match to get more conversions”	Keywords expand to broad; volume rises and so does irrelevant traffic. Without a dense negative list, CPA climbs	Dismiss or test
“Raise your budget to stop losing impressions”	Spend rises immediately, conversions follow a diminishing curve. Incremental CPA is never shown	Decide on the return curve
“Create a Performance Max campaign”	Auto-generated from existing campaigns, frequently cannibalizing brand traffic	Dismiss; launch deliberately instead
“Add negative keywords”	A list of non-converting queries ready to add	Review, then apply
“Pause underperforming keywords”	Pauses keywords with spend and zero conversions	Apply
“Fix your disapproved ads”	Surfaces the specific policy reason	Apply immediately
“Add similar audience segments”	Look-alike expansion with no hypothesis attached	Dismiss
“Optimize ad rotation”	Usually a switch from even rotation to optimize	Apply unless an ad test is running
“Set up conversion tracking”	Flags a real measurement failure	Apply — this is critical
“Add a seasonality adjustment”	A one-off bidding correction for a short spike	Apply only for a genuine 1–7 day event
“Expand your location targeting”	Adds regions without regard to logistics, language, or margin	Dismiss
“Add more assets to this ad group”	More combinations for the algorithm, wider format coverage	Apply if the assets are meaningful

A reading heuristic: if the wording contains “more” applied to reach, match, audiences, or budget, it’s a proposal to expand spend and it demands your arithmetic. If it says “fix,” “pause,” or “exclude,” it’s usually savings and can be applied quickly.

A note on seasonality adjustments

Seasonality adjustments are a rare case of a recommendation that is genuinely useful and easy to break through misuse. They exist for short, predictable spikes in conversion rate: a weekend sale, a launch, a one-to-seven-day promotion. Applied to long stretches like all of December, they feed Smart Bidding a distorted signal that takes weeks to unwind. Long-run seasonality is already handled by the bidding models on their own; you intervene only where a sharp, brief shift in conversion rate is coming that the algorithm cannot know about.

Auto-apply: the most dangerous checkbox in the account

Auto-apply lets Google implement selected categories of recommendations without your involvement. It sounds like a time-saver and functions as a transfer of budget control to a party with different incentives.

What actually happens with those boxes checked:

broad keywords get added into settled ad groups, and the structure you spent months building dissolves in a week;
budgets rise at the same time as traffic broadens — a compounding effect on spend, none of it recorded in your change log;
match types shift and audiences appear, so when CPA degrades two weeks later you go hunting through creative, because you don’t remember making any changes;
bid strategies re-enter learning at the worst possible moment.

The practical rule is blunt: turn auto-apply off entirely in every working account. The only arguable exception is purely technical categories such as fixing disapproved ads — and even that is better handled through alerts and your own scripts and automated rules, where the logic is written by you and visible in code.

Where to switch it off

Recommendations → Auto-apply tab.
Uncheck everything in both groups — “Improve performance” and “Account maintenance.”
Check at manager account (MCC) level: some settings are enabled from above and propagate to all child accounts, including newly linked ones.
Re-check quarterly. New accounts and interface updates occasionally reintroduce defaults in the enabled state.

Also audit who holds edit access. In teams, auto-apply is usually enabled not by the person running the account but by whoever took a call from a Google rep offering to “improve the optimization score.”

Handling calls from your Google rep

Google support calls regularly, offering to optimize the account for free. This isn’t a scam or a conspiracy: the rep has their own metrics, and recommendation adoption and spend growth are among them. They can deliver real value — beta access, policy clarification, help with technical issues. They are not accountable for your margin.

A workable protocol:

Never grant edit access “for the duration of the optimization.”
Ask for suggestions in writing — what specifically, and with what expected effect.
Queue contested proposals as experiments rather than applying them off the back of a call.
Use the relationship where it’s genuinely valuable: policy questions, restrictions, betas, and troubleshooting.

A weekly routine for handling recommendations

Open Recommendations and sort by type — not by weight. Weight reflects Google’s priorities, not yours.
Technical first. Disapproved ads, tracking errors, feed problems: apply immediately.
Then the money-savers. Negatives, pausing dead keywords and ads: eyeball them, then apply.
Then the neutral ones. Assets and extensions: apply if they match your plan and aren’t generated nonsense.
Spend-expanding items go into the hypothesis backlog. Not applied, not ignored — logged as a hypothesis and, if it’s worth it, validated by split test. The mechanics are in the guide to experiments in Google Ads and Experiment Center.
Dismiss the rest explicitly, with a reason where the interface asks for one — that trains the feed and reduces repeats.
Log what you applied and when. A month later, that log is the only way to connect a metric shift to a specific action.

Quarterly, this routine deserves a proper audit alongside it. Recommendations only surface what Google can see; they will never flag a broken conversion model or a structural problem. The full list of checks is in the Google Ads account audit checklist.

How the feed behaves across campaign types

The same list means different things in a search campaign and in Performance Max, and it deserves different handling.

Search. The most useful mix: negatives, pausing dead keywords, assets, technical fixes. Here recommendations sit closest to real account work, and the share of genuinely useful items is highest.
Performance Max. The list skews toward “add assets,” “broaden audience signals,” “raise budget.” Only the asset quality and coverage items are reliably useful. Anything about signal expansion or budget is an economics question again, not a score question.
Shopping. Heavy on feed diagnostics, and almost all of it is legitimate: disapproved products, missing attributes, price and availability mismatches. This is the case where the tab genuinely deserves a regular visit.
Demand Gen and video. Mostly creative and reach. Useful as a reminder of formats you haven’t uploaded, useless as guidance on budget growth.
Brand campaigns. A specific trap: the feed regularly proposes broadening match in brand campaigns, which instantly dilutes them with cheap irrelevant traffic. In brand campaigns, dismiss anything that expands reach by default.

Special case: optimization score in agency reporting

A specific failure mode: the score lands in a client report as a KPI. What follows is predictable — the team is asked to “get it to 90,” and starts applying things it shouldn’t. If the number must be reported, report it with two companions: how many recommendations were dismissed deliberately, and why. A single line — “dismissed 14 spend-expanding recommendations; expected effect on CPA was negative” — closes the conversation faster than any explanation of how the metric is weighted.

There’s an organizational angle too. In accounts where several parties can change settings — your team, the platform rep, a contractor — control over edits needs one owner. For teams that need a predictable, fully controlled account with workable spend limits, that’s solved at the infrastructure layer: agency Google Ads accounts and the wider PPC Rebels services cover exactly that.

FAQ: optimization score and recommendations

Does optimization score affect impressions or CPC?

No. The score doesn’t enter the auction and has no influence on Ad Rank or cost per click. It’s an interface metric reflecting the share of recommendations you’ve actioned.

What score should I aim for?

For a hands-on account, 70–85% is healthy. It usually means the useful items are applied and the spend-expanding ones were dismissed knowingly. A persistent 95–100% says more about bulk-applying than about quality.

Does dismissing recommendations harm the account?

No. Dismissal is a normal action and even raises the score. There’s no algorithmic penalty for declining recommendations.

Should I enable auto-apply at least partially?

In working accounts, no. Even the “safe” categories broaden over time, and the changes never appear in your change log — which turns any performance investigation into guesswork. Catch technical issues with alerts and your own rules instead.

Why does Google recommend broad match if it often hurts?

Because Google’s model optimizes reach and conversion volume in its own frame of reference, not your margin. Broad match genuinely delivers volume — the open question is its price and quality, and that’s your call, not the recommendation’s.

How do I tell whether a recommendation is actually useful?

Three questions: does it save money or spend more; is the improvement mechanism clear; can it be reverted in five minutes. If it spends more, the mechanism is fuzzy, and rollback is hard — that’s an experiment, not an action.

Do dismissed recommendations come back?

Some do, after a few weeks, especially when campaign conditions change. That’s expected: the list recalculates on fresh data. Dismiss again; the routine still holds.

Can I work with recommendations at MCC level?

Yes, the score and the list are available at manager level, which is convenient for a portfolio overview. Applying in bulk from there is risky though: campaign context differs across accounts while the action is identical.

Do recommendations account for my margin?

Not unless you pass conversion value into the account. While Google only sees “conversion = 1,” it can’t distinguish a 5%-margin sale from a 60%-margin one. Passing conversion value is the single change that makes the recommendation feed meaningfully smarter.

What if leadership demands 100%?

Show the composition of the score: what share of the weight sits on spend-expanding recommendations, and the expected CPA impact of applying them. One table with two columns — “score gain” and “spend increase” — usually moves the conversation from the score back to money.

Is the Recommendations tab worth opening at all?

Yes, as a signal source rather than a task list. Technical problems and negative keyword candidates surface there faster than through manual report review — everything else you filter yourself.

Optimization Score in Google Ads 2026: Which Recommendations to Apply and Which to Dismiss