Google Ads Optimization Score: What To Apply, What To Skip

Last updated: October 9, 2026

1. Executive Summary: Five Key Principles

Google Ads Optimization Score is a 0–100% real-time diagnostic that measures how closely an account’s configuration aligns with Google’s current recommendations—not how profitable, efficient, or strategically sound the account is.^[1] Senior marketers managing accounts in 2026 should operate by five non-negotiable principles.

• Score is a prioritization signal, not a performance KPI. A 10-point Optimization Score increase correlates with a median +14% lift in conversions according to Google’s own vendor claim, but the correlation between score and ROAS across 17,380 global accounts is only r = 0.29, while the correlation between score and ad spend is r = 0.81.^[2][3] Chasing score without validating business outcomes is a documented path to higher spend, not higher profit.
• Fix measurement before touching anything else. Automated bidding, Smart Bidding targets, and most high-impact recommendations optimize toward the conversion signal you supply. Incorrect, incomplete, or unqualified conversion data makes every downstream recommendation unreliable regardless of its score impact.
• Apply, test, or dismiss every recommendation deliberately. Apply low-risk measurement and policy repairs immediately. Test bidding, budget, match-type, and expansion recommendations against a control. Dismiss recommendations that conflict with intentional strategy, and record the business reason for each dismissal.
• Keep auto-apply narrowly scoped. As of January 26, 2026, Google removed “Add responsive search ads” from auto-apply entirely.^[4] That change illustrates how the auto-apply catalog shifts without notice. Enabling broad auto-apply transfers recurring strategic decisions to Google’s system and compounds changes before they can be reviewed.
• A lower score can be the correct score. Nineteen percent of accounts in a study of 17,380 global accounts reached 90%+ Optimization Score without accepting Google’s recommendations—primarily by dismissing irrelevant ones.^[2] A deliberate dismissal policy that reflects actual strategy is preferable to a high score achieved by accepting recommendations that increase spend or reduce targeting precision.

2. Benchmarks and Numbers at a Glance

Metric	Typical range or threshold	Applies when	Source
Optimization Score scale	0–100%	All active Google Ads accounts; vendor claim, not a study finding	[1]
Accounts scoring below 70%	32% of sampled accounts	Study of 17,380 global accounts; active ≥ 90 days; monthly spend $500–$1,000,000	[2]
Accounts reaching 90%+ without accepting recommendations	19% of sampled accounts	Same 17,380-account study; achieved by dismissing irrelevant recommendations	[2]
Accounts that accepted Google’s recommendations	5.5% of sampled accounts (fewer than 1,000 of 17,380)	Same study; note: not identified specifically as auto-apply adoption	[2]
Accounts scoring 90–100 after accepting recommendations	333 accounts	Subset of the 17,380-account study	[2]
ROAS difference: accounts scoring 90–100 vs. accounts scoring below 70	186% higher ROAS at 90–100 score band	Same 17,380-account study; global sample; correlation, not causation	[2]
Correlation between Optimization Score and ROAS	r = 0.29 (weak positive)	17,380 accounts; six-month period; monthly spend $500–$1,000,000; global	[5]
Correlation between Optimization Score and ad spend	r = 0.81 (strong positive)	Same 17,380-account study; higher score is strongly associated with higher spend	[5]
ROAS relationship above 70% Optimization Score	Essentially flat between 70% and 90%	Same study; incremental ROAS gain from pushing score above 70% is minimal	[5]
Median conversion lift after a 10-point score increase	+14% conversions	Vendor claim; Google internal data; sample size not disclosed; not a controlled study	[3]
Additional conversions after switching phrase match to broad match (Target CPA campaigns)	Approximately +25%	Vendor claim; Google internal data from 2020; sample not disclosed; requires stable Smart Bidding	[6]
Additional conversion value after switching phrase match to broad match (Target ROAS campaigns)	Approximately +12%	Vendor claim; Google internal data from 2020; sample not disclosed; requires stable Smart Bidding	[6]
CTR correlation with Optimization Score	No correlation detected	17,380-account study; score improvement does not predict CTR improvement	[2]
CPA relationship with Optimization Score	Higher scores (especially 80%+) correlated with improved CPA; no USD figure published	Same study; directional finding only; no controlled CPA change figure available	[2]

3. What Optimization Score Measures

Optimization Score is calculated in real time from four inputs: account and campaign statistics, current settings and status, the estimated impact of available recommendations, and the account’s recent history of applying or dismissing recommendations.^[1] Each unresolved recommendation reduces the score by its estimated impact weighting. Applying or dismissing a recommendation resolves the gap—which is why score movement does not reliably indicate that performance improved.

The score reflects alignment with Google’s recommendation framework for the account’s selected advertising objectives—such as maximizing conversions or conversion value—not alignment with the advertiser’s actual business economics, margin structure, lead quality requirements, or incremental revenue goals.^[1]

Google’s campaign goal is now displayed alongside the Optimization Score in the interface.^[7] Goals are determined primarily by bidding strategy and can be set to conversions, conversion value, clicks, or impression share. This makes it critical to confirm that the campaign’s bidding goal matches the actual business objective before interpreting or acting on score data.

Score can move without any meaningful change in business performance. New recommendation types introduced by Google, campaign condition changes, and shifts in recommendation availability can all alter the score independent of advertiser action. Treat an unexplained score drop as a prompt to inspect the Recommendations page—not as evidence that the account degraded.

4. Is 100% the Goal?

Google states that a 100% score means the account “can perform at its full potential,” but this is a vendor claim, not a verified performance guarantee.^[1] The practical evidence is less encouraging. In the 17,380-account study, the correlation between Optimization Score and ROAS is only r = 0.29, and ROAS is essentially flat between 70% and 90% score bands.^[5] The correlation between score and ad spend is r = 0.81—meaning that the surest outcome of a higher score is higher spend, not higher profit.^[5]

A 100% score is achievable in two very different ways: by applying all recommendations Google offers, or by dismissing every recommendation that does not apply to the account’s strategy. Nineteen percent of accounts in the same study reached 90%+ without accepting recommendations.^[2] This proves that dismissal is a legitimate and score-effective strategy, not a workaround.

The appropriate target score depends on account maturity, strategy, and risk tolerance. A score above 70% is a reasonable operational baseline—the study data suggest meaningful ROAS improvements are concentrated below 70%, while incremental gains above 70% are marginal.^[5] A score below 70% warrants a systematic review of what recommendations are outstanding and why. A score above 90% achieved by accepting recommendations that broadened targeting, increased budgets, or switched bidding strategies should be validated against CPA, ROAS, conversion quality, and profit—not accepted as proof of success.

5. Recommendation Types Explained

Google organizes recommendations into six broad categories, each of which can affect score by a different weighted amount depending on account conditions.^[8] The categories are: Bidding and budgets, Keywords and targeting, Ads and assets, Conversions and measurement, Campaign setup and maintenance, and Automation and audience (including Customer Match, broad match with brand inclusions, and app-related actions).^[8]

Recommendation availability is account-specific and changes dynamically as Google’s best practices evolve. The Auto-apply settings page should be treated as the current source of truth rather than any static external checklist.^[4] The estimated score impact of each recommendation is displayed in the interface; Google states this is the mechanism by which score uplift is communicated, though per-type benchmark values are not publicly reported as standardized figures.^[2]

For bidding recommendations, Google may suggest a Target ROAS based on recent actual ROAS (excluding conversion-delay-affected data) or a Target CPA based on the previous 30 days adjusted for conversion delays.^[9] These suggested targets are starting points for evaluation, not commercially validated figures. A suggested CPA based on 30 days of data may reflect a period with atypical seasonality, a landing-page test, or a tracking anomaly.

A specific newer recommendation type combines broad match keywords with brand inclusions, restricting broad-match coverage to specified brands or related products and services.^[8] This is a more constrained version of standard broad match and may be appropriate in cases where standard broad match would be dismissed.

6. Recommendations to Apply

Not all recommendations carry equal risk. The following categories are generally safe to apply after a brief human review because they repair operational defects, improve measurement fidelity, or add incremental value without altering strategic targeting or spend levels.

Measurement and tracking repairs are the highest-priority category. Conversion tag fixes, enhanced conversions setup, consent-mode adjustments, offline conversion imports, and attribution corrections should be resolved before any other recommendation is evaluated. Every bidding and budget recommendation downstream depends on the accuracy of this data.

Policy and serving repairs—disapproved ads, broken destination URLs, billing issues, and suspended assets—are operational defects, not optional optimization ideas. These reduce serving efficiency and are always worth fixing immediately.

Ad assets—sitelinks, callouts, structured snippets, image assets, price extensions, and promotion extensions—add inventory for Google’s auction-time serving system and are generally low-risk, provided the copy is accurate, commercially appropriate, and consistent with the landing page. Verify that each asset reflects current pricing, availability, and claims before applying.

Ad-quality improvements such as adding higher-rated responsive search ad headlines and descriptions can be applied when the suggested copy preserves brand voice, legal claims, and intended positioning. Do not apply AI-generated copy automatically; read every suggested asset before accepting.

Technical cleanup such as resolving conflicting negative keywords, enabling auto-tagging, and fixing duplicate conversion actions are low-risk when the change does not remove intentional segmentation or historical data continuity.

7. Recommendations to Dismiss

Dismissing a recommendation is not a failure—it is a legitimate governance action that resolves the recommendation from the score calculation just as applying it does.^[1] The key discipline is to dismiss deliberately, with a documented business reason, rather than leaving recommendations unresolved or dismissing them reflexively without evaluation.

The following recommendation types most commonly warrant dismissal in well-managed accounts:

• Budget increases that exceed profitability limits, approved monthly spend, or cash-flow capacity. A “limited by budget” label is a delivery diagnosis, not a profit guarantee.
• Bid-target changes based on insufficient conversion history, a skewed measurement window, or a suggested target that exceeds the margin-adjusted maximum allowable CPA or minimum acceptable ROAS.
• Broad match or audience expansion that would dilute a deliberately narrow, high-intent funnel—particularly in regulated industries, luxury categories, B2B accounts with long sales cycles, or accounts where offline lead qualification reveals significant quality variation.
• Keyword additions that introduce competitor terms, low-intent informational queries, or irrelevant product categories.
• Automatically generated ad copy that alters pricing claims, legal language, tone, promotional terms, or brand positioning without human review.
• Campaign-type upgrades—such as recommendations to move a Standard Shopping campaign to Performance Max—where the advertiser cannot isolate, audit, or qualify the resulting traffic mix.
• Removal of keywords, ads, or structures retained intentionally for testing, regional segmentation, brand protection, or historical reporting continuity.

When dismissing, use Google’s account-level or campaign-level dismissal controls and record the reason, owner, and review date in an external log or using campaign labels. This creates an audit trail and prevents the same recommendation from being applied inadvertently by another team member.

8. The Risks of Auto-Apply

Auto-apply is not inherently unsafe, but enabling it broadly transfers recurring strategic decisions to Google’s system and creates compounded changes that can be difficult to diagnose or reverse.^[4] The most significant structural risk is that multiple automatic changes can accumulate before a weekly review cycle catches them, making it impossible to attribute a performance shift to a specific change.

The risks most frequently realized in practice include:

• Spend escalation from reach-expanding recommendations. Auto-apply does not directly increase campaign budgets, but recommendations that broaden match types, add keywords, or expand audiences can increase impression volume and total spend within the existing daily budget.^[4]
• Bad optimization signals scaled by automation. If form fills, call conversions, or imported conversions are not properly qualified, auto-applied bidding changes will optimize and scale low-quality leads.
• Brand and compliance exposure. Automatically created or modified ads may require legal, compliance, or brand-team review before serving—review that auto-apply bypasses.
• Loss of strategic segmentation. Auto-applied broad match, audience expansion, or campaign-structure changes can dissolve carefully designed segmentation that took months to build.
• False confidence from a rising score. A higher score after auto-apply activity reflects adoption of Google’s recommendations, not verified improvement in profit, lead quality, or incremental revenue.^[2]

The clearest recent example of auto-apply risk is structural: from January 26, 2026, Google removed the “Add responsive search ads” recommendation from Auto-apply entirely.^[4] Advertisers who had relied on this feature to maintain ad coverage discovered the change only when new ad creation stopped. This illustrates that the auto-apply catalog changes without advance notice, and governance policies built on a static assumption about available categories will drift.

Recommended auto-apply policy for 2026: Keep auto-apply off by default for budgets, bids, keywords, match types, targeting, audience expansion, and measurement settings. Consider enabling narrowly scoped categories—such as fixing disapproved ads or broken destinations—only after confirming their scope in a test account or reviewing recent auto-apply history for the specific category. Review Auto-apply History and Change history at least weekly for any account with auto-apply enabled.

9. Using Score in Audits and Across Accounts

At the individual account level, Optimization Score functions as a recommendation workflow metric: it shows which recommendations are outstanding and how much score each represents. At the portfolio or manager-account level, it functions as a triage tool for identifying accounts with significant unresolved opportunities—but only when scores are aggregated correctly.

Google’s API documentation specifies that a multi-account aggregate score should use each customer’s optimization score weight, not a simple average.^[10] The correct formula is:

Portfolio score = Σ(customer score × score weight) ÷ Σ(score weight)

Averaging raw percentages equally across accounts of different sizes, spend levels, and campaign counts produces a misleading portfolio figure. A $500/month account with a 95% score and a $500,000/month account with a 62% score should not be averaged to 78.5%—the larger account’s recommendations carry far more dollar impact.

Audit framework for individual accounts:

• Export account and campaign scores, all active recommendations with estimated score uplift, auto-apply subscriptions, Change history for the prior 90 days, and current conversion tracking status.
• Classify each recommendation by type: measurement, policy, creative, bidding, budget, targeting, or expansion.
• Rank by expected business impact—not score impact—combining estimated uplift with spend concentration, conversion volume, CPA/ROAS relative to target, and recommendation reversibility.
• Fix measurement defects before evaluating any other recommendation.
• Document every dismissal with a reason, owner, and review date.
• After applying material changes, monitor conversion quality, search-term mix, CPA/ROAS, and impression share for a minimum of 14–28 days before drawing conclusions.

10. Broad Match and Budget Recommendations

Broad match and budget recommendations are the two categories most likely to increase spend without a proportional increase in qualified conversions. They are also the two most frequently surfaced recommendation types in accounts with significant unresolved opportunities. Treating them with the same workflow as measurement or asset recommendations is one of the most common and costly errors in Google Ads account management.

Broad Match

Google’s vendor claim is that switching from phrase match to broad match in Target CPA campaigns generates approximately 25% more conversions, and in Target ROAS campaigns approximately 12% more conversion value.^[6] Both figures are from 2020 internal data with undisclosed sample sizes and should be treated as directional, not as applicable benchmarks for any specific account.

Broad match is appropriate when conversion tracking is accurate, Smart Bidding is stable with meaningful conversion history, search-term quality is monitored at least weekly, negatives and exclusions are well maintained, and the business can tolerate exploratory traffic while performance is judged on valuable conversions.^[1] It is not appropriate when lead quality varies substantially, offline qualification data is absent, queries must be tightly controlled, or the recommendation would replace proven exact-match coverage without a controlled experiment.

A 2026 recommendation variant combines broad match with brand inclusions, restricting matching to specified brands or related products and services.^[8] This constrained version may be testable in cases where standard broad match would be dismissed.

Budget Recommendations

A “campaign limited by budget” flag means additional eligible impressions exist above the current daily budget ceiling. It does not prove that additional impressions will convert at the current CPA or ROAS.^[1] Increase budget only when the campaign is meeting its CPA or ROAS target, incremental conversions are likely to be valuable, and Search impression share and lost-to-budget data support expansion at the margin.

Auto-apply does not increase campaign budgets directly, but the broader recommendation ecosystem can increase traffic and spend through other changes.^[4] Budget decisions require explicit human authorization at every increment.

11. Common Mistakes to Avoid

The following errors recur across accounts at all spend levels. Each one is observable in the Recommendations page, Change history, or downstream performance data and can be corrected with a clear process.

• Treating score as a performance KPI in client reports. The correlation between score and ROAS is r = 0.29; the correlation between score and spend is r = 0.81.^[5] Reporting score alongside CPA, ROAS, and conversion volume without this context implies a relationship that the data does not support.
• Applying bidding recommendations before validating conversion data. A tCPA or tROAS recommendation is only as reliable as the conversion signal Google used to generate it. Apply measurement fixes first; apply bidding changes after.
• Accepting broad match without a negative keyword foundation. Broad match without robust negatives, location settings, and audience layering produces exploratory traffic that raises spend and dilutes conversion quality, particularly in B2B, regulated, or high-CPC verticals.
• Enabling all auto-apply categories at account setup. The auto-apply catalog changes without advance notice—as demonstrated by the removal of RSA auto-creation on January 26, 2026.^[4] A broad auto-apply subscription will eventually include categories that conflict with account strategy.
• Dismissing recommendations without logging the reason. An undocumented dismissal looks identical to an oversight in an audit. Log every dismissal with a reason, owner, and review date.
• Using simple score averages across a portfolio. Equal-weight averaging ignores spend concentration. Use the API’s score-weight field for accurate portfolio aggregation.^[10]
• Evaluating bidding changes during the learning period alone. Smart Bidding learning periods typically last 1–2 weeks depending on conversion volume. CPA and ROAS during this period are not representative of steady-state performance.
• Accepting Target CPA or ROAS suggestions without checking the measurement window. Google’s suggested targets are based on the prior 30 days, which may include promotional periods, seasonal spikes, landing-page tests, or tracking anomalies that make the period unrepresentative.^[9]
• Ignoring the campaign goal displayed next to the score. A score of 90% against a “Clicks” goal means the account is well configured to maximize clicks—not conversions or revenue. Verify that the displayed goal matches the business objective before pursuing score improvement.^[7]

12. What Changed Recently (Last 30 Days)

As of October 2026, the most material documented change to Optimization Score and recommendations is the removal of “Add responsive search ads” from the Auto-apply settings page, effective January 26, 2026.^[4] This change means that Google no longer automatically generates or applies new RSAs through auto-apply. Advertisers who relied on this feature for ad coverage, asset variety, and ad strength improvement must now manage RSA creation manually or through third-party tools.

The practical impact is twofold: first, accounts that had auto-apply enabled for RSA creation may now have static ad inventory that is not keeping pace with asset testing and coverage requirements. Second, the change illustrates the broader governance risk of relying on a static auto-apply configuration—the available recommendation types can change at any time, and a subscription list built in Q3 2025 may no longer reflect what is actually being applied in Q4 2026.

Beyond the RSA change, Google’s current guidance confirms several operational characteristics that practitioners should treat as standing policy rather than new developments:

• Auto-apply operates at the account level, not the campaign level, and applies only when a recommendation is considered relevant by Google’s system.^[4]
• Auto-apply does not increase campaign budgets.^[4]
• The available recommendation types continue to expand dynamically; the Auto-apply settings page is the authoritative source of what is currently available and enabled.^[4][8]
• Google may suggest Target ROAS or Target CPA values after a new bid strategy is created or a recommendation is applied; these are starting points based on recent account data and conversion-delay adjustments, not commercially validated targets.^[9]
• The campaign goal is now displayed alongside the Optimization Score in the interface, making it more straightforward to detect goal mismatches before acting on recommendations.^[7]

Recommendation categories currently documented by Google include: bidding and budgets, keywords and targeting (including broad match with brand inclusions), ads and assets, conversions and measurement, campaign setup and maintenance, and automation-related actions including Customer Match list refresh and Firebase app conversion actions.^[8] Availability of any specific category remains account-specific.

---

References

1. [1] https://support.google.com/google-ads/answer/9061546?hl=en-GB support.google.com
2. [2] https://www.optmyzr.com/blog/does-optimization-score-affect-google-ads-performance/ www.optmyzr.com
3. [3] https://ads.google.com/intl/ky_us/home/tools/recommendations/ ads.google.com
4. [4] https://support.google.com/google-ads/answer/10279006?hl=en support.google.com
5. [5] https://observix.ai/blog/google-optimization-score-guide observix.ai
6. [6] https://support.google.com/google-ads/answer/10195720?hl=en support.google.com
7. [7] https://support.google.com/google-ads/announcements/9048695?hl=en-GB support.google.com
8. [8] https://support.google.com/google-ads/answer/3416396?hl=en-GB support.google.com
9. [9] https://support.google.com/google-ads/faq/10286469?hl=en support.google.com
10. [10] https://developers.google.com/google-ads/api/docs/recommendations developers.google.com