We looked at the popular advice on writing Responsive Search Ad copy. Here's what actually holds up.

Responsive Search Ads (RSAs, the Google Search ad format where you supply up to 15 headlines and 4 descriptions and the platform assembles combinations to show) replaced fixed-text ads a few years ago, and a whole genre of "how to write RSA copy" content sprang up around it. Most of it reads as confident instruction: pin this, don't pin that, keep headlines short, use sentence case. Very little of it says where the claim comes from or how strong the evidence actually is.

That matters because the underlying mechanism, how Google's machine learning selects and combines your assets, is not fully published. Some of the popular advice is genuinely consistent with what Google has said publicly about how RSAs work. Some of it is plausible-sounding pattern-matching from marketers who ran a few tests on a handful of accounts and generalized. This piece goes claim by claim: what the mechanism would predict, what the published research and Google's own guidance actually say, and how confident you should be before you copy any of it into your own account.

RSAs let you "pin" a headline or description to a fixed position (say, always headline 1) so it always appears there, or leave it unpinned so Google's system can slot it into any position and test it against your other assets in combination. The two extremes people default to are pinning everything (treating the RSA like an old-style fixed ad, just with extra room) or pinning nothing (handing the whole thing to the algorithm).

Both extremes have a real mechanical cost. Fully pinning removes the platform's ability to test different combinations against each other, which is the entire point of the RSA format, so you're paying for a flexible ad unit and using it as a rigid one. Fully unpinning, on the other hand, means any of your headlines can land next to any other, including combinations that don't make grammatical or logical sense together (two headlines that both assume they're the sentence's subject, for instance, or a discount claim next to unrelated copy that reads as a non sequitur when stacked).

Published research and Google's own guidance point toward a middle path: pin only the handful of assets where position genuinely matters, most commonly a legally required disclaimer, your brand name if brand presence is non-negotiable, or a claim that only makes sense in a specific slot, and leave everything else unpinned so the system can keep testing combinations. The mechanism here is straightforward enough to trust: you're constraining only where constraint is truly needed and preserving the algorithm's search space everywhere else. That said, "partially pinning" is not itself a formula. How many assets to pin and which ones is account-specific, and the right answer for a heavily regulated industry (finance, health) will look different from a simple ecommerce account.

A commonly repeated claim is that shorter headlines, especially ones that display in full without getting truncated on smaller placements, convert at a lower cost than longer ones. The mechanism proposed is reasonable: a headline someone can read and process in under a second is less friction than one that gets cut off mid-sentence or takes longer to parse, and truncation can turn a coherent claim into a confusing half-sentence.

We'd flag this as a plausible, publicly discussed pattern rather than an established law. It's the kind of claim that's easy to state confidently and hard to verify at the level of "how much shorter, and by how much," because that depends heavily on what you're selling and who's searching. A technical B2B software buyer evaluating a considered purchase may read a slightly longer, more specific headline just fine; an impulse consumer purchase probably rewards brevity more. Treat "shorter tends to help" as a reasonable starting hypothesis for your first draft, not a hard ceiling you should compress every headline toward regardless of what the offer needs to say.

Another frequently cited pattern is that sentence case (only the first word and proper nouns capitalized, the way you'd write a normal sentence) outperforms Title Case (Where Every Word Is Capitalized). The theory is that Title Case visually signals "this is an advertisement" in a way that trained ad-blind users skip over, while sentence case reads more like organic content or a human wrote it, lowering the reader's guard.

That's a coherent psychological story, and it's genuinely testable inside your own account (case is one of the easiest variables to isolate in an RSA test, since you can write matched pairs of headlines that differ only in capitalization). But it's a claim about reader psychology, and reader psychology varies by audience and category. A finance or legal audience that expects formal, authoritative language might respond differently than a consumer lifestyle audience. Worth running as a real test with real headline pairs, not worth adopting as a blanket house style before you've checked it against your own account.

Common advice says to repeat the exact search keyword across several of your headline slots, on the logic that relevance drives Quality Score and click-through, and matching the searcher's own words back to them signals relevance clearly. There's real mechanism behind this: keyword-to-headline matching is a genuine relevance signal.

But Google's asset-diversity requirements (RSAs need enough distinct headlines and descriptions that the system can build genuinely different-sounding combinations, and it can suppress or deprioritize a set of assets that are too similar to each other) mean the same instinct taken too far can backfire. If eight of your fifteen headlines are near-duplicates of the same keyword phrase with minor wording changes, you're not giving the algorithm eight meaningfully different assets to test, you're giving it one idea repeated eight times, and the platform's own mechanics can work against you for it. The honest read: some repetition for relevance is fine and probably helps, but it needs to sit alongside headlines that do other jobs (price, differentiator, social proof, call to action) rather than being the majority of your fifteen slots.

Every claim above ranges from "mechanistically sound and consistent with Google's own guidance" (partial pinning) to "plausible but genuinely unverified for your specific case" (case style, exact headline-length cutoffs). None of it should be copied into your account as a rule. Treat each one as a hypothesis: write a real test with matched variants, run it long enough to get a meaningful sample, and look at what your account's actual data says.

This is the honest limit of any external advice on ad copy, including this article. Ad copy performance is genuinely audience-specific and offer-specific: what reads as natural and converts for a B2B SaaS buyer will differ from what works for an impulse consumer purchase, and no published research or industry analysis substitutes for your own account's test results. The useful posture is skeptical, not credulous: understand why a claim might be true, then go check whether it's true for you.