Why landing page headlines fail

Eye-tracking studies have argued for two decades that visitors decide whether to stay or bounce inside the first viewport. Whatever happens in the body copy, in the case studies further down, in the pricing table — none of it matters if the H1 didn’t earn a second glance. So the most read sentence on any landing page is usually a phrase the founder wrote in fifteen minutes, never changed, and assumes is fine.

We wanted to know what separates the H1s that earn the second glance from the ones that don’t. So we pulled 1,200 H1 headlines from the roast.page dataset — every page our analyzer had scored 35 or higher (filtering out broken pages) across six industries — and hand-classified them along three axes. Two reviewers per page, third reviewer breaks ties, 87% agreement. Then we matched each H1 to its corresponding page score.

Four patterns emerged. The first is the one most landing-page writers think they already know.

We split the H1 sample into two buckets. Outcome headlines name what the visitor gets (“Get paid in 1 click,” “Ship dependable AI agents”). Feature headlines describe what the product is (“AI-powered analytics platform,” “All-in-one CRM for growing businesses”). Edge cases — “The fastest CRM” names a property of the product, not a customer outcome — went to the feature bucket.

The two groups score very differently. Pages with outcome H1s average 58/100 overall. Pages with feature H1s average 44/100. The 14-point gap holds when we control for industry: feature-led SaaS pages averaged 47, outcome-led SaaS pages averaged 60. Same pattern in e-commerce, agencies, and dev tools.

The H1 is the only sentence everyone reads. Most teams spend an hour on it once and never look back.

The gap isn’t small. 14 points is the difference between a page sitting in the bottom half of the distribution and one reaching the top quartile. And the change costs nothing except a rewrite.

What does outcome-driven copy look like in the wild?

The clearest examples are short, specific, and ruthlessly focused on what the visitor walks away with:

What does feature-dumping look like?

The pattern repeats so consistently across the bottom half of the distribution that we could write a generator for it: The [adjective] [category] for [audience]. Plug in “AI-powered analytics platform for growing teams,” “All-in-one HR software for SMBs,” or “Modern project management for engineering teams.” All three describe what the product is. None of them tell the visitor what changes in their life.

The replacement pattern is equally simple: [Outcome the reader wants] — [for whom]. “Close your books in three days, not three weeks — for finance teams at 50–500 person companies.” That sentence took us thirty seconds. In our regression, pages that made this exact swap moved an average of 11.4 points.

The top quartile’s average H1 is 6.8 words. The median page averages 7.2 words. That’s noise — neither group is meaningfully shorter than the other. So “keep your headline under N words” is the wrong advice. The right advice is to look at what fills those words.

We tagged every H1 for the presence of a concrete anchor — a number (“1,000 teams”), a timeframe (“in 5 minutes”), a dollar amount (“save $4,800/month”), or a named result (“close more deals” counts; “grow your business” doesn’t). The presence of any of these anchors split the dataset cleanly.

The specificity gap is the largest in the table. Pages that anchor the H1 in something countable score, on average, 9 points higher than pages that don’t. “Save 12 hours per week on reporting” outperforms “Save time on reporting” by 8 points in our within-industry comparison.

The simplest reason is that numbers force you to know your customer. You can’t write “close books in 3 days” unless you know how long it currently takes (10–14 days for most growth-stage finance teams). The vague version — “close books faster” — is what you write when you don’t want to commit. The specific version is what you write when you’ve talked to enough customers to know the answer.

In January 2023, “AI-powered” in a landing page H1 was a signal. In May 2026, it’s wallpaper. 42% of SaaS pages we sampled use either “AI,” “AI-powered,” or “AI-driven” in the H1 or hero subheadline. The phrase tells the visitor nothing they can’t infer from the year being 2026.

We compared the Differentiation dimension scores for pages that use the phrase versus pages that don’t. Pages with “AI-powered” in the H1 averaged 3.4/10 on Differentiation. Pages that named what their AI actually does — “Drafts your follow-up emails in your voice,” “Catches lookalike fraud at 2 ms latency” — averaged 5.9/10. That’s a 2.5-point dimension gap (the dimension is 0–10), which translates to a 6-to-8 point swing on the overall score depending on how heavily the rest of the page leans on AI claims.

The fix is structural, not stylistic. Don’t describe the technology — describe its job-to-be-done. “AI customer support” becomes “Resolves password resets without a ticket.” “AI sales assistant” becomes “Books your demo calls while you’re still typing.”

What about pages whose differentiator genuinely is the AI?

Even then, the pattern holds. Anthropic’s claude.com H1 doesn’t say “AI-powered.” It says “Claude is your thoughtful collaborator at work.” The AI is implicit in the brand; the headline does the harder work of describing the relationship the user gets to have with it. That’s the upgrade move. Don’t announce the technology — describe the experience it produces.

We expected fintech, dev tools, and enterprise SaaS to dominate the top of the distribution. They sell to readers who know what good copy looks like. Instead, those industries have some of the widest internal gaps — meaning the worst pages in those categories are dragging the average down, while the best are world-class.

Two patterns jump out. First, course creators are the only industry where the median page leads with an outcome — because hype-driven copy is outcome-driven (“Transform your life,” “Make $10K/month”) even when the proof is thin. That’s why their Copy scores are still mid-pack despite the outcome dominance: the outcomes are unbelievable. Outcome + believable beats outcome alone.

Second, agencies are the worst offenders in absolute terms. Only 21% of agency pages we sampled lead with what the client achieves. The other 79% describe the agency — “Full-service digital marketing,” “Award-winning creative studio,” “Performance-driven media agency.” Agencies, more than any other category, sell with their portfolio and their case studies. The H1 should set up the portfolio, not duplicate it. The strongest agency page in the dataset led with “We help SaaS companies grow from $1M to $10M ARR.” Score: 78. Specificity, anchor, named outcome, named buyer.

Don’t convene a workshop. Don’t hire an agency. Pull up your current H1 and run it through five tests. If your H1 fails three or more, rewrite it tonight.

1. The outcome test. Read your H1 aloud. Does it name what the customer gets, or what your product is? If it describes the product, you’re writing in the wrong direction.
2. The pronoun test. Does the sentence contain the word “you” or “your”? If not, you’re probably describing yourself when you should be addressing the reader.
3. The number test. Is there a digit, a dollar sign, a percentage, or a timeframe anywhere in the sentence? If not, can you add one without lying? “Faster invoicing” becomes “Invoices clients in 90 seconds.”
4. The “AI” test. Strike the phrase “AI-powered” from your H1. Does the sentence still make sense? If yes, the phrase was filler. If the sentence collapses without it, replace it with what the AI does — not what it is.
5. The competitor test. Could a direct competitor use the exact same H1 without anyone noticing? If yes, you’ve written a category description, not a product description. Rewrite until your competitor would feel weird stealing it.

Once you have a candidate H1, run it against your live page using the headline analyzer or score the full page against this dataset with a free roast. Expect the score to move somewhere between 8 and 14 points if your old H1 was feature-led. We watched that exact lift happen on hundreds of pages while building this study.