Responsive Search Ads: Why Most RSA Setups Underperform and How to Fix It

Google's Responsive Search Ads system is supposed to find the best-performing combinations of your headlines and descriptions automatically. In practice, most RSA setups underperform because advertisers treat them like expanded text ads with extra steps rather than using the combination testing for what it was built for.

This post covers the most common RSA mistakes and the practical fixes that actually move performance.

The Core Mistake: Headlines That Can't Stand Alone

Google can serve any combination of your RSA headlines in any position. That means every headline needs to make sense as a standalone sentence fragment as well as in combination with any other headline you've provided.

Most RSA setups fail this test because headlines were written as a sequence, not as independent units. "Get More Leads" works alone. "Get More Leads / From B2B Decision-Makers / At a Lower Cost" only works as a set. When Google serves headlines 1 and 3 without headline 2, the message breaks.

Fix: Write each headline so it functions as an independent value statement. Then test whether any two-headline combination still makes sense. Combinations that break your message will be served — Google doesn't know your intent, only the click data.

Under-Pinning: The Optimization Trap

Pinning a headline to Position 1 or 2 locks it there permanently, regardless of performance. The RSA system learns which combinations perform in which contexts. When you pin, you remove a dimension of that learning.

Advertisers over-pin because they want to control the message. That's understandable. But controlling the message prevents the system from finding which version of the message performs best with each user intent.

Fix: Pin only what is legally or compliance-required (disclaimers, brand name requirements). Everything else should float. If you must pin a brand name, pin it to one position and leave Position 1 open for Google to optimize.

The Redundancy Problem

RSA performance data is reported at the ad level, not the asset level (unless you use the Asset Report in Google Ads). Advertisers who can't see which headlines are performing often assume the setup is working fine when one or two headlines are being served 80% of the time.

The most common reason for this over-concentration: redundant assets. If you have four headlines that all say variations of "Free Trial Available," Google picks the best-performing one and stops testing the others. Your RSA now effectively has 11 headlines, not 15.

Fix: Check the Asset Report weekly. Identify headlines and descriptions marked "Low" in the performance bucket. Replace them with conceptually distinct alternatives, not just word variations of the same message.

Ad Strength Is a Prompt, Not a KPI

Google's Ad Strength indicator ranges from "Poor" to "Excellent." It measures diversity and quantity of unique assets, not performance. An "Excellent" RSA can underperform a "Good" RSA if the excellent setup has more keywords stuffed into it rather than more unique value propositions.

Many advertisers optimize for Ad Strength without knowing they're optimizing for a proxy metric that doesn't predict clicks or conversions.

Fix: Use Ad Strength as a checklist prompt (have you included enough distinct messages?) not as a performance benchmark. The Asset Report's "Learning," "Low," "Good," and "Best" labels are more directly tied to actual serving behavior.

Practical Setup for a New RSA

A RSA that tests effectively:

Headlines (15 maximum, aim for 12-15): Write 3-4 headline groups covering distinct angles. For B2B: feature/capability, outcome/result, social proof/number, objection handling. Two to three headlines per group. Each headline reads as a standalone fragment.

Descriptions (4 maximum): One for each stage of the consideration funnel: awareness (what it is), evaluation (why it's different), decision (CTA), reassurance (proof or guarantee). Most advertisers use 4 variations of the same CTA. Don't.

Match types: Pair your RSA with phrase and exact match keywords targeting the same intent. Broad match against RSA amplifies the combination testing but requires close negative management to prevent relevance drift.

How to Benchmark Your Current RSA Setup

Run the Asset Report and count how many of your 15 headlines are in the "Best" or "Good" bucket vs "Low." If fewer than 5 are showing performance signal (Best or Good), your asset diversity is insufficient for the system to learn. Add new concept angles rather than variations.

Compare CTR for the ad group against your account's historical ETA baseline if you have it. RSAs can underperform ETAs for 4-6 weeks while learning. If CTR is still below baseline after 6 weeks, the asset set needs revision.