Ad headlines under twenty characters produced a cost per acquisition of $9.35, against $18.27 for longer ones, across roughly twenty thousand Google Ads accounts. That is close to half, from a constraint you can apply this afternoon without changing anything else.
The same analysis found sentence case beating title case by a factor of 3.7. Both findings run against how most accounts are written.
This is for you if your headlines are carefully written, use every available character, and are not doing much.
What the twenty thousand account analysis found
Optmyzr analysed responsive search ads across roughly 20,000 active Google Ads accounts in April 2026.
Headlines under twenty characters: cost per acquisition of $9.35. Longer headlines: $18.27. Roughly double.
Sentence case against title case was starker. Sentence case produced $7.46, title case $27.47, a factor of 3.7.
On pinning, partial pinning performed best and full pinning worst, which follows from the mechanism. Pinning everything prevents the system combining assets, which is the thing you are paying it to do.
The caveat worth stating: Optmyzr’s dataset skews towards experienced practitioners running its software, so this is not a random sample of all advertisers. At twenty thousand accounts it remains a scale nothing you run will ever match.
Why short ad headlines probably win
The study reports the effect rather than the cause, so this is reasoning rather than evidence.
Length invites generality. Given sixty characters, most people fill them, and the filling is qualifiers, context and hedging. Given twenty, you have to say one thing.
One specific thing outperforms three vague ones, consistently, across every format I have worked in. The character limit is enforcing specificity that discipline alone rarely achieves.
There is a reading-speed argument too. A search results page is scanned, not read, and a short headline is comprehended in a glance while a long one requires a decision to engage.
Both explanations point the same way, which is that brevity is doing something real rather than being a quirk of the dataset.
Why sentence case probably wins
The 3.7 factor is large enough to need an explanation, and the plausible one is register.
Title Case Reads Like A Headline. Sentence case reads like a person writing a sentence. On a page full of adverts all shouting in title case, the one that sounds like somebody talking is the one that stands out.
That also fits everything else in this cluster. Adverts that do not look like adverts outperform, and capitalisation is one of the cheapest signals that something is an advert.
Worth testing rather than adopting on faith, and it is close to free to try. Changing capitalisation costs nothing and breaks nothing.
Where Google’s own advice on ad headlines disagrees
This is the part that makes the topic interesting rather than just a formatting tip.
Google’s guidance emphasises ad strength, and states that moving from Poor to Excellent is worth 15% more clicks and conversions.
A separate Optmyzr analysis of over a million ads across more than 22,000 accounts found no correlation between ad strength and conversion performance. Their April 2026 study found ads rated Excellent had the worst cost per acquisition in the dataset at $28.68, while Average had the best at $12.43.
Both sources have an interest. Google benefits from advertisers adopting broad, unpinned, high-strength assets. Optmyzr sells software to people optimising accounts.
The honest position is that Google’s advice and the largest independent dataset available point in different directions, and you should know that before optimising towards either.
What to do with your ad headlines this month
Four changes, in order of how cheap they are.
Convert your headlines to sentence case. Free, reversible, and supported by the largest comparison available.
Then write shorter. Aim under twenty characters on at least several assets rather than all of them, since the system needs variety to combine.
Then unpin most of what you have pinned, keeping pins only where a legal or brand requirement genuinely demands a fixed position.
Then supply more assets rather than longer ones. Quantity of options helps the combination engine. Length within each option is where the cost appears.
What this does not mean
Two overcorrections worth avoiding, because both follow naturally from the findings.
It does not mean every headline should be under twenty characters. Supply a range. The system needs assets that work in different combinations, and a set of uniformly tiny fragments limits it differently from a set of uniformly long ones.
And it does not mean length is the only variable. A short vague headline is worse than a long specific one. The constraint helps because it usually forces specificity, not because character count has magic in it.
If shortening a headline makes it generic, you have removed the wrong words.
Why ad headlines evidence is worth more than most creative evidence
One thing separates this from nearly everything else in this cluster, and it is worth saying.
Most creative benchmarks have no source. Hook rate, thumbstop rate, fatigue frequency thresholds and the UGC premium all trace to vendor content citing other vendor content, with no sample size anywhere.
This does not. It comes from a named analysis, of a stated number of accounts, published with the figures attached, by a company whose interest is disclosed and whose dataset bias is known and stateable.
That does not make it beyond question. It makes it checkable, which is a category apart from most of what circulates about creative.
So when you are weighing whether to act on a creative finding, the sourcing question matters more than the size of the effect. A modest effect you can trace beats a dramatic one you cannot, every time.
Whether ad headlines behave the same on Meta
Honest answer: this evidence is from search and I would not transfer it as fact.
Search headlines are scanned alongside competitors on a results page, with high intent and no image. Meta headlines sit beneath an image that has already done most of the stopping work, in a feed, with much lower intent.
Those are different jobs, and a finding about one does not automatically hold for the other.
The underlying reason is plausible on both, since brevity forcing specificity is not platform-specific. Test it on Meta rather than assuming it, and be clear in your reporting which platform your evidence came from.
That distinction matters more than it sounds. A great deal of paid social advice is search evidence that migrated without anybody noting the change of setting.
What to change this week
Three steps.
Convert your search headlines to sentence case, which takes minutes and costs nothing.
Then count how many of your headline assets exceed twenty characters, and write shorter alternatives for at least half of them.
Then check what you have pinned, and unpin anything without a legal or brand reason to stay fixed.
Ad copy formulas that are not cliches covers the structures these headlines sit inside, voice of customer research covers where the specific words come from, fund the ad set and read what the algorithm chose is the position underneath, and why your landing page converts at two percent covers what happens after the click. On Meta the same brevity discipline is worth testing, which my Meta Ads work does routinely. To review your assets against this evidence, book a teardown.