Writing High-Performing Search Ads: RSA, Pinning, Ad Strength and Testing

by Francis Rozange | Apr 4, 2026 | Google Ads

rediger-annonces-rsa-performantes-google-ads-1

Writing high-performing Search Ads: RSA, pinning, Ad Strength and testing

Responsive Search Ads are the only Search format Google Ads accepts in 2026. Expanded Text Ads were sunset on June 30, 2022, and the migration was final: no new ETAs, no edits to existing ones, no exceptions. Yet four years later, most accounts are still running RSAs as if they were two-headline ads with a few extras bolted on. They pin too much, they fill the asset slots with synonyms, they chase Excellent Ad Strength as a KPI, and they declare a winner after seven days and 200 clicks.

The mechanics are not intuitive. An RSA is not an ad, it is a template that can produce thousands of permutations. Ad Strength does not measure performance, it measures input quality, and Google’s own internal research found the two correlate weakly in many verticals. Pinning a single headline cuts your effective testing surface by 75 percent. Asset performance ratings (Best, Good, Low, Pending) take 30 days to stabilize and lie to you for the first two weeks. None of this is in the onboarding flow.

This guide walks through the full stack: how RSAs actually serve, how Ad Strength is calculated and where it breaks, when pinning is the right answer (rare) and when it is a nervous tic (common), the copy frameworks that survive contact with Smart Bidding, the policy traps that get ads disapproved, and a 30-day testing playbook you can run on accounts spending anywhere from 2,000 to 200,000 USD per month. Sources are inline: Google Ads Help, WordStream, Search Engine Land, Search Engine Journal, Tinuiti, Optmyzr.

Why ad copy still matters in the Smart Bidding era

A pattern has set in among advertisers who came up after 2020: Smart Bidding will fix it. The thinking is that target CPA and target ROAS are powerful enough to compensate for mediocre creative, that the algorithm finds buyers regardless of what the ad says. This is wrong, and it is wrong in a way that costs real money.

Smart Bidding optimizes the auction. It decides whether to bid 2.40 USD or 4.10 USD on a given query, given the user’s signals. It cannot decide whether to click. The click is a function of the ad: the headline that appears, the description that supports it, the relevance to the query, the tone, the offer. According to WordStream, advertisers combining machine learning bidding with diverse RSA assets see up to 15 percent more conversions than those running fixed creative. The two systems multiply, they do not substitute.

Search Engine Land made the same point in a 2024 audit of 200 Google Ads accounts: the gap between top-quartile and bottom-quartile performers was not bidding strategy (most used target CPA or maximize conversions), it was Quality Score. And Quality Score is driven by ad relevance and expected CTR, both of which trace back to the headline and description copy you wrote.

The Smart Bidding signal stack also depends on copy. The algorithm reads headline performance per query intent, per device, per audience segment, per time of day. If your assets are seven variants of “Buy Coffee Online,” the system has nothing to learn from. If your assets cover price, urgency, social proof, brand, and feature in distinct ways, the algorithm builds a richer model of which message converts which user, and it bids accordingly. Diverse copy makes Smart Bidding smarter. Homogeneous copy makes it conservative.

Tinuiti’s 2025 paid search benchmark report showed that accounts that refreshed RSA copy quarterly saw 8 to 12 percent CTR lift year-over-year, while accounts that left RSAs untouched for 12 months saw flat or declining CTR despite higher spend. Copy decay is real. Audiences habituate to messaging, competitors copy your angles, market language shifts. The ad you wrote in January is not the ad your November searcher needs.

Anatomy of a Responsive Search Ad

An RSA accepts up to 15 headlines, each capped at 30 characters, and up to 4 descriptions, each capped at 90 characters. You also provide one final URL, optional path fields (two segments, 15 characters each, that appear in the visible URL), and tracking parameters. Google then assembles ads at serve time, choosing 2 or 3 headlines and 1 or 2 descriptions per impression, with placement and combination optimized per query.

The combinatorics are larger than most advertisers realize. With 15 headlines taken 3 at a time, ordered, you get 2,730 permutations. Multiply by description pairs from a pool of 4 (12 ordered pairs), and you reach 32,760 unique creative outputs. Add the desktop variant that may show 2 headlines instead of 3, and the single-description mobile variant, and the search engine has tens of thousands of distinct ad surfaces it can produce from a single RSA. Google Ads Help describes this as the system’s main advantage: maximum relevance per query, drawn from a single creative asset library.

This is also the source of most confusion. When an advertiser says “my ad is converting at 4 percent,” they are usually describing aggregate performance across thousands of underlying combinations, not the performance of a fixed creative. The RSA report shows totals, not the winning combination. Until 2024, you could not see which headlines drove which clicks at all. The new asset-level reporting changed that, but only partially: you see headline-level click and conversion attribution, not full combination attribution. Google still treats the assembly logic as proprietary.

The asset slots themselves are not interchangeable. Google distinguishes between unpinned headlines (which can rotate through any of the three displayed positions), and pinned headlines (locked to position 1, 2, or 3). The same applies to descriptions, with positions 1 and 2. Pinning has cascading effects on Ad Strength and on combination diversity that we cover below.

One asset rule trips up new advertisers: you cannot have duplicate headlines, even with case differences. “Free Shipping” and “FREE SHIPPING” are flagged as duplicates. You also cannot use the same word more than once across your headline pool in some cases, depending on Google’s duplicate-content detection. The system will accept the asset but assign it low expected CTR, which lowers Ad Strength.

Headline pinning rules and when to break them

Pinning forces a headline or description into a specific position. You select the asset, click the pin icon, and choose position 1, 2, or 3 (for headlines) or 1 or 2 (for descriptions). From that moment, the pinned asset shows in that exact slot every time the ad serves, or the ad does not serve at all if no other asset can be combined.

The 75 percent figure is not folklore. Search Engine Land’s analysis of pinning impact found that pinning a single headline to position 1 reduces the number of testable combinations by approximately three quarters, because the system can no longer rotate the remaining 14 headlines through that slot. Pinning two headlines compounds the effect. Pinning all three positions reduces an RSA to something that behaves like an ETA: a fixed creative with no learning surface.

Optmyzr ran a study on 1,200 accounts in 2024 and found that accounts with zero pins outperformed accounts with one or more pins by 14 percent on conversion rate, on average. The exception was accounts with strict legal or compliance requirements, where pinning was unavoidable. In those cases, the workaround is to pin multiple headlines to the same position rather than just one.

The “loose pinning” technique

Loose pinning means assigning two or three headlines to the same pinned position. If you must have a brand name in position 1, do not pin a single brand headline. Write three brand variants (“LaFactory Coffee,” “LaFactory Roasters,” “LaFactory: Since 2010”), and pin all three to position 1. Google can now rotate among them, learning which brand framing performs best while still satisfying your compliance constraint.

This is the technique Google’s own optimization specialists recommend in account reviews. It preserves Ad Strength, it preserves combination diversity, and it gives the algorithm a microbattle to optimize. The cost is the discipline of writing three good brand headlines instead of one and accepting them all as legitimate.

When pinning is the correct call

Pin when you have a non-negotiable constraint:

Legal or regulatory copy that must appear in every ad (financial disclosures, pharmaceutical disclaimers, regulated industries).
Brand-mandated copy from clients or legal review (always-on tagline, trademark notice).
Time-sensitive offers where missing the message defeats the campaign (“Sale Ends Sunday,” “Black Friday Only”).
A single asset with overwhelming historical evidence of dominance (50 percent higher conversion rate than the next best, on more than 1,000 conversions). This is rare in practice.

Do not pin to satisfy a manager who wants their favorite tagline visible. Do not pin because you are not sure which headline is best and want to “control” the output. Do not pin all positions to recreate an ETA. Each of these is the most common failure mode in pinning, and each costs measurable conversions.

Ad Strength meter: what it measures, what it doesn’t

Ad Strength is the meter Google shows next to each RSA in the Ads interface: Incomplete, Poor, Average, Good, or Excellent. Google Ads Help describes the calculation as a function of three inputs: number of headlines and descriptions, asset relevance to keywords in the ad group, and asset diversity (lack of repetition). Pinning lowers the score because it reduces combinations. Missing assets lower it because the algorithm has fewer combinations to learn from.

Google’s published claim: improving Ad Strength from Poor to Excellent yields, on average, 12 percent more conversions. This number is repeated everywhere, often without context. The context is critical.

The Google research that nuanced the claim

In 2023, Google published an internal study (referenced in Search Engine Land’s coverage) that examined the correlation between Ad Strength and ROAS across hundreds of thousands of campaigns. The headline finding: Ad Strength correlated positively with ROAS in aggregate, but the correlation was weak and varied wildly by vertical. In some verticals (B2B SaaS, high-consideration purchases), accounts with Good Ad Strength outperformed accounts with Excellent on ROAS terms. In others (e-commerce with broad audiences), the relationship held strongly.

The reason is that Ad Strength rewards diversity and quantity, but conversion rate rewards focus and clarity. A 15-headline RSA that covers price, urgency, social proof, feature, brand, CTA, value prop, and three more angles scores Excellent. But if your audience converts primarily on social proof and price, the other 12 headlines are noise that dilutes serving frequency for the two that work.

Optmyzr’s 2024 analysis confirmed the pattern: 8 percent of accounts they reviewed had Good Ad Strength outperforming Excellent on conversion rate. The driver was always the same: tightly focused asset libraries beat broad ones in narrow-intent verticals.

What Ad Strength actually tells you

Use Ad Strength as a diagnostic, not a target. Specifically:

If it flags “Add more headlines,” check whether you have under 10. Below 10, you genuinely lack input diversity. Above 10, the warning is often noise.
If it flags “Make your headlines more unique,” read your assets. If three of them say the same thing in different words, prune. If they cover distinct angles, ignore the warning.
If it flags “Include more keywords,” check whether your headlines reference your top 3 keywords at all. If yes, the warning is calibration; if no, fix it.
If it flags “Reduce pinning,” and you do not have a legal reason to pin, unpin. The warning is correct.

The trap is treating Ad Strength as a KPI in itself. Account managers who report “we got 80 percent of RSAs to Excellent this month” are reporting an input metric, not an output metric. The output metric is conversions per dollar.

Asset performance ratings interpretation

Google rates each headline and description with a performance label: Pending, Low, Good, or Best. These ratings appear in the Ads interface under each RSA, in the asset details panel. They are based on relative performance within the same RSA: a Best rating means the asset outperforms others in the same ad on contribution to conversions.

The 30-day rule

Asset ratings are unreliable for the first two weeks. Google Ads Help notes that ratings require sufficient impression volume to stabilize, and that volume varies by ad group. A high-traffic ad group might produce reliable ratings in 7 to 10 days. A low-traffic ad group might take 30 to 45.

The practical rule: do not act on asset ratings before 30 days, unless the volume is overwhelming. Acting too early creates noise: you pause a “Low” headline that was actually Best material, and you replace it with something worse.

What each rating means in practice

Pending: not enough data. Ignore. Wait.
Low: this asset underperforms relative to others in the ad. Candidate for replacement, after 30 days.
Good: middle of the pack. Keep.
Best: top performer. Keep, study, and use as a template for new variants.

Note the relativity. A Best rating in a weak ad means “least bad,” not “objectively excellent.” If your overall RSA is converting at 1.2 percent and the category average is 4 percent, the Best headline is still part of an underperforming ad. Use Best ratings to identify what to keep when iterating, not to congratulate yourself.

Iteration cadence

The recommended cycle is to review asset ratings every 30 to 45 days, replace the bottom 20 to 30 percent of headlines with new variants inspired by the Best performers, and let the new asset library run for another 30 days. Search Engine Journal’s audit data suggests this cadence produces 3 to 5 percent compounding monthly improvement in CTR over six months, before plateauing.

Copy frameworks that convert

The headline frameworks below are not theoretical. They are pulled from Tinuiti, WordStream, and Search Engine Journal teardowns of high-performing accounts, and they consistently outperform “We sell good products” copy by margins of 30 to 80 percent on CTR.

Problem-Agitation-Solution (PAS)

Old framework, still works. Headline 1 names the problem, headline 2 amplifies the cost of the problem, headline 3 presents the solution. Example for a payroll software: “Payroll Eating Your Friday?” / “Stop Manual Entry Errors” / “Automate in Under 10 Minutes.” This works for high-friction B2B and B2C purchases where the buyer is already aware they have a problem and is searching for relief.

Before-After-Bridge (BAB)

Headline 1 describes the current state, headline 2 describes the desired state, headline 3 is the bridge (your product). “Stuck With Bland Coffee?” / “Wake Up to Specialty Roast” / “Single-Origin Subscription.” Works for aspirational products and lifestyle categories.

Feature-Advantage-Benefit (FAB)

The classic sales copy structure. Feature (“Cold-Brew Filtered Twice”), advantage (“Lower Acidity Than Standard”), benefit (“Easier on Your Stomach”). Strong for technical products where buyers do feature comparison shopping. The trap is stopping at feature: “Cold-Brew Filtered Twice” alone is a fact, not a reason to buy.

Social proof headlines

Numbers and named entities. “Trusted by 50,000 Coffee Lovers,” “Featured in Bon Appetit 2024,” “4.9 Stars Across 8,200 Reviews.” Social proof headlines have the highest CTR of any framework in WordStream’s 2024 cross-account benchmark, but they require real, defensible numbers. Inflated or rounded social proof gets disapproved under Google’s misrepresentation policy and damages Quality Score.

Specificity and numbers

Specific numbers outperform generic claims by 25 to 40 percent on CTR. “Save 23 Percent” beats “Save Money.” “Ships in 36 Hours” beats “Fast Shipping.” “12-Month Warranty” beats “Long Warranty.” The cognitive mechanism is credibility: a specific number signals real measurement, while a generic claim signals marketing hand-waving.

CTA distribution

Across 15 headlines, distribute three CTAs at different intent levels. Soft CTA for top-funnel (“Explore the Range”). Mid CTA for consideration (“Compare Plans”). Hard CTA for bottom-funnel (“Start Free Trial,” “Buy Now Save 15 Percent”). Smart Bidding will route serving by predicted user intent, but only if you give it the asset variety to work with.

The 5-bucket asset library

A working RSA covers five buckets across its 15 headlines: keyword-rich (3 headlines, primary keyword referenced naturally), value proposition (2 to 3 headlines, the “why us”), CTA (2 to 3 headlines at varied intent levels), feature and benefit (3 to 4 headlines on specifics), brand and trust (2 to 3 headlines on credibility). The 4 descriptions split similarly: one logistics, one quality, one experience, one offer.

Common policy traps

Disapprovals waste budget and time. The common traps are:

Excessive capitalization

“BUY NOW” gets disapproved. “Buy Now” does not. Google treats more than 50 percent capitalized characters in a headline as gimmicky and rejects under the editorial policy. Acronyms are exempt: “USA” or “SUV” pass. Title case (“Buy The Best Coffee Online”) passes. Sentence case is the safest default.

Excessive punctuation

Multiple exclamation marks, multiple question marks, ellipses, and unusual character sequences trigger rejection. “Buy Now!!!” fails. “Buy Now!” passes. The rule is one exclamation mark per ad, used sparingly. Many accounts default to no exclamation marks at all because the urgency they convey is rarely worth the disapproval risk.

Trademark issues

You cannot use a competitor’s trademark in your headline if the trademark holder has filed a complaint with Google. The system flags it on submission. You can bid on competitor brand keywords (that is allowed), but you cannot reproduce the competitor’s name in your visible ad copy. The fine line: “Better Than [Competitor]” gets disapproved; “Compare CRMs” does not. Google’s trademark policy details the process.

Misrepresentation

“Lose 30 Pounds in 30 Days” gets disapproved under unrealistic claims. “Free!” without a clear path to the free thing gets disapproved. “Number 1 Rated” without a citation gets disapproved. The rule is that any claim must be substantiated on the landing page, and any superlative must be defensible.

Banned categories and restrictions

Healthcare, financial services, alcohol, gambling, and political advertising have category-specific restrictions that go beyond standard policy. If you operate in these verticals, the disapproval rate is higher and the appeals process longer. Build extra time into campaign launches.

Special character abuse

Symbols (™, ®, ©, $, €, %) are allowed but limited. Emoji, asterisks used decoratively, and repeated symbols get rejected. The rule is: characters that convey real information pass, decorative characters fail.

RSA testing pattern: 4 ads per ad group, ad rotation, statistical significance

Google’s recommended testing pattern, echoed by every major account audit firm, is three to four RSAs per ad group with ad rotation set to “Optimize.” More than four creates testing noise and dilutes serving. Fewer than three under-uses the testing surface.

Why “Optimize” not “Rotate evenly”

The “Rotate indefinitely” option (now called “Do not optimize, rotate ads indefinitely”) forces equal serving regardless of performance. This sounds fair but defeats the purpose of RSAs: the algorithm cannot route serving based on predicted performance, so you lose the relevance gains that justify RSAs in the first place. Google’s documentation actively discourages this setting outside specific A/B test scenarios where you control rotation manually.

“Optimize” lets the algorithm bias serving toward the higher-performing RSA. Both ads still serve, but at unequal rates. Over 30 days, this is more efficient: the winner gets more impressions, the loser gets fewer, total conversions rise.

Sample size for declaring a winner

An RSA test is not a coin flip with 30 results. The combinatorics inflate the sample size requirement. Optmyzr’s statistical significance calculator recommends 500 to 1,000 conversions per RSA before declaring a winner with 95 percent confidence, depending on the magnitude of the difference. For a 10 percent CTR lift, you need around 500 conversions per ad. For a 5 percent lift, closer to 2,000.

The implication: small accounts (under 50 conversions per month per ad group) cannot run reliable RSA A/B tests. They should run a single best-effort RSA, iterate the assets every 45 days based on Best/Low ratings, and accept that they are flying without statistical power.

Test variable isolation

If you change everything between test ads, you cannot attribute the difference to any single change. Discipline: change one dimension at a time. Test ad has different headlines, same descriptions, same paths. Or same headlines, different descriptions. Or same copy, different final URL. Sequential isolation is slower but yields actionable causality.

Parallel multivariate testing (changing multiple variables at once) is faster but produces noisier results. Use it only when speed matters more than causality, which is rare.

Test duration

Minimum 14 days, ideally 30. Shorter tests miss day-of-week effects (B2B accounts have wildly different Tuesday and Saturday behavior). Longer tests risk seasonal contamination (a test running through Black Friday will not generalize to ordinary weeks). The 30-day window is the standard compromise.

Interaction with Smart Bidding

Smart Bidding (target CPA, target ROAS, maximize conversions, maximize conversion value) does not bid on the ad. It bids on the auction. But its inputs include the ad: predicted CTR, predicted conversion rate, asset performance signals, query-to-headline relevance scores. The richer your RSA asset library, the more signal Smart Bidding has to work with.

Concretely, if you run an RSA with 8 weak headlines and pinned aggressively, Smart Bidding has limited inputs. It will bid conservatively, miss high-intent auctions, and produce mediocre results. If you run an RSA with 15 diverse, unpinned headlines covering five message buckets, Smart Bidding builds a richer model: it sees that headline 4 (urgency) drives clicks for users searching during business hours, headline 9 (price) drives clicks for users on mobile after 8pm, headline 12 (brand) drives clicks for repeat searchers. It bids accordingly.

Tinuiti’s 2024 paid search teardown found that accounts pairing diverse RSA copy with target CPA bidding outperformed accounts pairing thin RSA copy with the same bidding strategy by 18 to 24 percent on conversion volume at the same target CPA. The bidding strategy was identical. The creative diversity was the difference.

Learning period interaction

Smart Bidding has its own learning period (7 to 14 days after a major change). RSAs have a learning period (7 to 21 days for asset ratings to stabilize). Changing both simultaneously confounds the two learning curves. Best practice: when you launch a new RSA, do not change bidding strategy in the same week. When you change bidding strategy, do not refresh RSA assets in the same week. Stagger the changes by 14 days minimum.

The “do not pin to control message” trap

Some advertisers pin specific headlines to control which message Smart Bidding sees most often. This is backwards. Smart Bidding’s value is dynamic message routing per query intent. Pinning forces a static message and removes the routing benefit. The result is worse performance from a strategy that sounded controlled.

30-day RSA testing playbook

The playbook below assumes you are starting with one or two existing RSAs in an ad group, on an account spending 5,000 USD or more per month. Smaller accounts can adapt by extending the cadence to 60 days.

Day 1 to 7: baseline and audit

Pull the existing RSA performance data: 30 to 90 days of impressions, clicks, CTR, conversions, conversion rate, cost per conversion. Note the current Ad Strength rating, the asset performance ratings (Best, Good, Low, Pending), and any pinning. Document the baseline. Identify the ad group’s top 5 keywords by impression volume and conversion volume.

Audit the existing assets against the 5-bucket framework: keyword-rich, value prop, CTA, feature/benefit, brand/trust. Mark which buckets are underrepresented or overrepresented. Mark headlines that are duplicate variants of each other.

Day 8 to 14: build the new RSA

Write 15 new headlines and 4 new descriptions, addressing the gaps identified in the audit. Apply the copy frameworks (PAS, BAB, FAB, social proof, specificity, distributed CTAs). Avoid policy traps. Review for duplicate-content issues. Verify all assets reference the ad group’s keyword theme.

Pin only what you must (legal, brand). If you must pin, use loose pinning (2 or 3 headlines per pinned position).

Launch the new RSA into the same ad group as the existing one. Set ad rotation to “Optimize.” Do not change bidding strategy or budget in the same week.

Day 15 to 30: monitor and iterate

Check Ad Strength on day 1 of the new RSA. Aim for Good or Excellent. If it scores Average or below, audit which warning fired and address it (more headlines, more diversity, less pinning).

Monitor at 7 days, 14 days, 21 days. Do not act on asset ratings before day 14, and ideally not before day 21. Watch for policy disapprovals (rare if you followed the audit, but possible).

By day 30, you should have stabilized asset ratings and at least 100 to 500 conversions across both RSAs (depending on account volume). Compare conversion rate, CTR, and cost per conversion. If the new RSA outperforms by 10 percent or more on conversion rate, pause the old one. If it underperforms by 10 percent or more, pause the new one and write a fresh variant. If they are within 5 percent of each other, keep both running and revisit at day 60.

Day 30 onward: continuous iteration

Every 30 to 45 days, replace the bottom 20 to 30 percent of headlines (Low ratings) with new variants based on the Best performers’ patterns. Refresh descriptions every 60 to 90 days. Review pinning quarterly. Audit for policy creep (Google updates policy quietly; what passed last quarter may fail this one).

Quarterly, step back and ask: are the keywords in this ad group still aligned with the messaging? Has the audience or competitive landscape shifted? Refresh the asset library wholesale once per year, even if performance is stable. Copy decay catches up with even the best ads.

Common mistakes

Mistake 1: pinning everything to control output

The most common failure mode. The advertiser does not trust the algorithm and pins all three headline positions and both description positions. The RSA collapses into a fixed creative that scores Poor on Ad Strength and underperforms unpinned RSAs in the same account by 20 to 30 percent. Fix: unpin everything except true legal or brand requirements, and use loose pinning when you must pin.

Mistake 2: filler headlines to hit 15

“Learn More,” “Visit Our Site,” “Click Here Now,” “Browse Our Products.” These add nothing and dilute the asset library. Google tests them, wastes impressions, and the ad’s overall conversion rate drops. Fix: stop at 12 strong headlines. Do not add headlines 13, 14, 15 if they are filler.

Mistake 3: keyword stuffing

Repeating the primary keyword in 10 of 15 headlines. “Buy Coffee,” “Coffee For Sale,” “Cheap Coffee,” “Coffee Online,” “Best Coffee,” and so on. Google’s stemming and synonym matching make this redundant. The system flags low diversity, and the asset library wastes its testing surface. Fix: primary keyword in 3 headlines, related keywords in 2 to 3 more, the rest on benefits, CTAs, social proof, brand.

Mistake 4: declaring a winner too early

Day 7, 200 clicks, 4 conversions, and the advertiser pauses the “loser.” The decision is statistical noise. Fix: 30-day minimum, 500-conversion target per ad. If the account cannot produce that volume, accept that you do not have testing power and switch to incremental iteration without head-to-head testing.

Mistake 5: changing too many variables at once

The advertiser launches a new RSA, switches bidding from manual CPC to target CPA, raises budget by 50 percent, and adds a new audience segment, all in the same week. When performance shifts, attribution is impossible. Fix: change one variable at a time, with 14 days between changes minimum.

Mistake 6: ignoring asset ratings after 60 days

The advertiser launches an RSA, sets it and forgets it. Six months later, the asset ratings have stabilized, three headlines are flagged Low, but no one has acted. The ad continues underperforming. Fix: schedule a recurring 45-day review of asset ratings on every active RSA. Calendar it.

Mistake 7: chasing Excellent Ad Strength as a KPI

The advertiser adds weak headlines just to clear the “low diversity” warning and reach Excellent. Conversion rate drops because the new headlines dilute serving. Fix: treat Ad Strength as a diagnostic, not a target. Aim for Good with focused, high-quality assets, not Excellent with noise.

Conclusion

RSAs reward discipline and patience. The advertiser who writes 15 thoughtful headlines across five message buckets, pins nothing they do not have to pin, lets the asset library run for 30 days before judging it, replaces the bottom performers every 45 days, and does not chase Excellent Ad Strength for its own sake will outperform the advertiser who runs three RSAs with pinning everywhere and rebuilds them every two weeks based on noisy data.

Smart Bidding amplifies whatever signal you give it. A diverse, well-written RSA library produces a rich signal, and Smart Bidding routes high-intent users to your best message at the right bid. A thin or homogeneous library produces a weak signal, and Smart Bidding bids conservatively against averages.

The 12 percent uplift figure for moving from Poor to Excellent Ad Strength is real in aggregate, but your account is not the aggregate. Measure your own relationship between asset structure and conversion outcomes. After three to six months of disciplined iteration, you will know exactly which copy buckets convert in your vertical, which CTAs work for your audience, and how much pinning your specific compliance environment forces on you. That knowledge compounds. The advertiser who builds it has a permanent edge over the advertiser who treats RSAs as a settings screen.

Build assets like a craftsman. Test like a statistician. Iterate like a gardener. The platform rewards all three.