A headline is not a summary

What it is competing against

A headline is not read in isolation. It is read in a feed, an inbox, a search page or a poster site, alongside a dozen other things making the same bid, by somebody who has already decided to skip most of them. It has one job, and the job is not to describe the piece.

Its job is to get the next line read. That is the whole of it. The headline that describes the piece accurately and is not read has failed completely; the headline that gets the first line read has succeeded even if it left things out, provided the piece then pays what the headline implied.

Once you accept that, a lot of common headline advice reorganises itself. “Be clear” is not a value in itself, it is a tactic that usually works because confusion is a reason to move on. “Be intriguing” is the same tactic from the other side and fails when the intrigue is unresolvable.

Specific nouns do the heavy lifting

Compare:

Improving Operational Efficiency in Small Manufacturing

The £90 tool that cut our reject rate in half

The second is not better because it is punchier. It is better because it contains three things a reader can picture (a price, a tool, a defect) and abstractions cannot be pictured. “Operational efficiency” is a category that contains thousands of things, so the reader has to do work to find out whether any of them concern them, and they will not do that work.

The practical test: underline every noun in your headline and ask whether you could photograph it. If none of them can be photographed, you have written a category, not a headline.

The two failures at the edges

Too vague is the commoner one, and it usually comes from a fear of being wrong. “A new approach to storytelling” commits to nothing and therefore attracts nobody.

Too clever is the rarer and more painful one. A pun that requires reading the piece to understand is a locked door with the key inside. Wordplay works when it is readable both ways: funny if you get it, plain if you do not.

Between them sits the version that is specific and slightly incomplete: it names the thing and withholds one piece, and the withheld piece is in the article.

Front-load, because the end may not arrive

Text is cut off in more places than you expect, and always from the right. Search results truncate. Email subject lines truncate, more aggressively on a phone than on a desktop. Link previews truncate. Poster text gets read at an angle by somebody walking.

So the distinguishing word goes at the front, not at the end.

• “How we cut our reject rate in half using a £90 tool” → truncated, reads “How we cut our reject rate in...”
• “The £90 tool that halved our reject rate” → truncated, still reads “The £90 tool that halved...”

The second survives because its most specific element is in the first four words.

Write twenty, keep one

This is the part people skip and it is the part that works. Headlines are cheap to produce and expensive to get right, which is exactly the shape of problem that rewards volume. Set a timer for ten minutes and write twenty, deliberately including bad ones: the literal one, the question one, the number one, the too-clever one, the flatly boring one. The good one is usually somewhere around fifteen, after you have exhausted the obvious moves.

Do not do this at the start. Write the piece first. A headline written before the piece describes the piece you intended; a headline written after describes the one you made, and those are different documents.

The limitation, which is about incentives

Everything above optimises for getting read, and getting read is a measurable thing, which means it can quietly become the objective. This goes wrong in a specific way: the headline formats that test best in a single measurement are the ones that create an unresolved question, and an unresolved question can be created without having an answer. Once a piece is judged only by whether the headline was clicked, the cheapest way to improve is to promise more than the piece contains.

The cost does not appear in the click figure. It appears later, in whether anybody reads the second piece, and by the time it shows up in the numbers you have trained an audience to discount you. The discipline is simple to state and hard to hold: the headline may withhold, but it may not promise something the piece does not deliver.