How to write headlines that work

What a headline has to do

A headline has one job: earn the next few seconds of attention from the right person. It does not need to explain everything. It needs to tell a stranger what this is, whether it is for them, and why to keep going.

Nielsen Norman Group (NN/g) calls headlines microcontent. In a 2017 article, Hoa Loranger and Jakob Nielsen note that headlines often appear without the page around them, in feeds, search results and emails, so the "headline text has to stand on its own". If your headline only makes sense after reading the page, it fails in the places most people first see it.

The same NN/g research that shows people scan pages explains why headlines carry so much weight. In its 1997 research, 79% of test users always scanned new pages. Headings and the first words of each line are what scanners see. See what is copywriting for the reading research.

Clarity beats cleverness

A clever headline asks the reader to decode a joke. A clear headline lets them recognise their own problem. When both are possible, choose clear. Cleverness works only when the reader already knows what you offer, which is rarely the case on a first visit.

NN/g's 1997 guidance asks for meaningful subheadings, not clever ones, and its later plain-language article argues that clear writing helps expert readers as well as beginners. The digital.gov guide on plain language, and the habit of writing scannable text, gives the mechanics: shorter words, active voice, and real verbs instead of abstract nouns.

A fast test: cover the rest of the page and show the headline to someone outside your business for five seconds. Ask what you sell and for whom. If they guess wrong, rewrite.

Be specific, and put the key words first

NN/g's headline advice is to be specific and give readers something useful, rather than broad, generic phrasing. "Better accounting" says nothing. "Monthly bookkeeping for freelancers" says who, what and how often.

The second rule is front-loading: put the most informative words at the start. The inverted pyramid works the same way at page level. People scan the left edge of a line, and NN/g says they notice the first words more than the last. "Tax returns for freelancers: filed by a named accountant" works better than "A named accountant will file your tax returns if you are a freelancer".

On length, NN/g does not give a limit for headlines. Its 40 to 60 character guideline applies to page titles in search results. Keep on-page headlines short because short text is easier to scan, and write the HTML title separately for search.

Formulas that work, with examples for six industries

A formula is a reusable shape. None of them guarantees results, and every one needs real content from your customer research. See voice-of-customer research for how to find the words to put in the blanks. The examples below are fictional businesses, and each assumes the claim is true for that business.

• Plain statement: [what you do] for [who]. Accounting: "Bookkeeping and tax for freelancers, from one named accountant."
• Outcome without pain: [outcome], without [the thing they dread]. Law: "A clear plan for separating, without having to decide everything at once."
• The reader's question: ask what they already ask. Health: "Back pain that keeps coming back? Book an assessment, no referral needed."
• Specific promise: say exactly what happens and when, only if it is true. Trades: "Roof repair quotes in writing before any work starts."
• Situation fit: name the situation the product is for. Ecommerce: "Backpacks sized for a one-bag week away."
• From and to: show the change in a line. SaaS: "From shared spreadsheets to one scheduling view for every clinic."
• How-to or number: only when the page truly delivers it. "Five questions to ask before hiring a roofer", if the page has five questions.

Lead with the outcome, not the feature

A feature is what your product has. A benefit is what changes for the customer. "24/7 online booking" is a feature. "Book a visit at 11pm and wake up to a confirmation" is the benefit, written as something the customer can picture.

The right outcome is the one customers name, not the one you assume. This is where jobs to be done thinking helps: ask what progress the buyer is trying to make, then write the headline about that progress. Your interviews will tell you whether the strongest outcome is saving time, avoiding risk, feeling in control or something you had not guessed.

Outcomes need care. Say what the work is designed to do, or what the customer can expect from the process, not a guaranteed result. "We prepare your return so you can file with confidence" is honest. "Never fear an audit again" is not.

Honesty is part of the craft

NN/g warns that click bait the content does not deliver erodes trust. A headline is a promise, and the page is where you keep it. A reader who clicks for one thing and finds another leaves, and remembers.

There is a legal side too. The US Federal Trade Commission says advertising must be truthful and not misleading, advertisers need a reasonable basis for claims before they run, and the commission looks at the overall impression, including implied claims. "Guaranteed results", "cure" and "best in the city" are the kind of headline claims that need evidence or should not be used.

Regulated fields such as law, health and finance have tighter rules on outcomes, testimonials and comparisons. Check your regulator's rules, and have a qualified person review headlines before they go live.

Testing headlines honestly

A split test (also called an A/B test) shows different visitors different headlines and compares what they do. It is the fairest way to settle a disagreement, but only when you run it properly. NN/g notes in a 2005 article that it can detect small differences and measure real behaviour, but cannot explain why one version won.

The most common mistake is stopping early. Evan Miller, an experiment-design writer, shows that if you check a test repeatedly and stop when it first looks significant, false winners become far more likely. In his worst-case example, stopping at a reported 5% significance level gives a false positive 26.1% of the time. His fix is to choose a sample size in advance and not stop early.

His rule of thumb for the sample needed is n = 16 x variance / difference squared. For a conversion rate of 5% where you want to detect a rise to 6%, the variance is 0.05 x 0.95 = 0.0475 and the difference is 0.01. That gives about 7,600 visitors per version. A site with a few hundred visitors a month cannot run that test in a reasonable time.

If you lack the traffic, do not pretend. Use the qualitative route instead: show headlines to five to ten target buyers and ask them what the page offers, in their own words. Test one change at a time, and change the headline, not the colour of the button at the same time.

A step-by-step process for a new headline

Write ten versions before you choose one, because the first idea is usually the most obvious. Use the process below, and stop when a stranger can explain your offer back to you.

• State the offer in one plain sentence, as if explaining to a friend.
• Pull the reader's own phrases for the problem and the outcome from your research.
• Draft one headline from each formula above, then cut every word that does not carry meaning.
• Move the most informative words to the front.
• Check every claim against evidence you can show on the page.
• Read each headline alone, with no page around it, and drop any that need context.
• Pick two finalists, show them to five target buyers, and test only if you have the traffic.