Here is a thing almost nobody outside a newsroom knows, and it explains about half the arguments people have about the media: in most newsrooms, the person who wrote the headline is not the person who wrote the story. * The reporter files copy. Somebody else — a copy editor, a subeditor, a homepage editor, a social media editor — puts the words at the top. Sometimes the reporter suggests a headline. Sometimes they see the published one for the first time when their phone starts buzzing with people yelling at them about it. * So when you read a headline that seems to contradict the article underneath it, you are usually not catching a journalist in a lie. You are watching two different jobs collide in public. * **Why the split exists at all** * It is a print habit that survived into the internet. Headlines had to fit a physical column, in a specific typeface, at a specific size. That is a typesetting problem as much as a writing problem, and it needs someone looking at the page, not at the notebook. Reporters, meanwhile, are famously bad at compressing their own work — everything feels essential when you spent three days on it. * So the craft split. The reporter's job is to be right. The headline writer's job is to be right *in eight words that fit*. Those two jobs pull in different directions, and the second one is where accuracy goes to get squeezed. * **The regulator literally wrote a rule about it** * This is not a conspiracy theory — it is codified. In the UK, Clause 1 of the Editors' Code of Practice, the rulebook enforced by press regulator IPSO, opens like this: "The Press must take care not to publish inaccurate, misleading or distorted information or images, including headlines not supported by the text" (Editors' Code of Practice, Clause 1(i), via the Society of Editors). * Read that last bit again. "Including headlines not supported by the text." A regulator does not add a specific clause about headlines unless headlines are, as a category, a recurring problem. * And it gets enforced. In ruling 10895-22, *Various v The Daily Telegraph*, IPSO upheld complaints about a print article headlined "Doctors and nurses may join wave of strikes," published 20 July 2022. The complaint was that the framing around a 9.3% pay figure was misleading — doctors and nurses were not in line for that rise. IPSO's Complaints Committee found the inaccuracy significant, because it sat in the sub-headline and the first line and went to the heart of what the article was about, and ruled there was a breach of Clause 1 (IPSO, ruling 10895-22). * Note what the ruling turned on: not the reporting further down the page. The bit at the top. * **Online, the headline stopped being one headline** * In print, a story had one headline forever. Online, a single article can be wearing four different hats at once: * The **page headline** you see when you land on the article. The **SEO headline** buried in the HTML, written to match what people type into Google — often flatter and more literal. The **social headline** in the Open Graph tags, written to survive a scroll. And the **homepage headline**, which changes depending on where the story is placed and how much space that slot has. * They are frequently different sentences. That is why the version of a story your friend screenshotted from Twitter sometimes does not match the headline you see when you click the link. Nobody swapped it to trick you. There were always several. * **Then came the A/B test** * Here is the part that should genuinely change how you read a homepage: some headlines are not chosen by a human judgement call at all. They are chosen by measurement. * The clearest public evidence is the Upworthy Research Archive, published in *Nature Scientific Data* in 2021 by Matias, Munger, Le Quere and Ebersole. Upworthy ran a randomised experiment on essentially everything it published, and the archive records the lot: 32,487 experiments, 150,817 experimental arms — that is, individual headline-and-image packages — and 538,272,878 participant assignments, between 24 January 2013 and 30 April 2015 (Matias et al., *Scientific Data*, 2021). * Half a billion people, quietly serving as the test group for which sentence made them click. Upworthy was unusually open about it. It was not unusually eager. Testing headline variants is standard practice at large digital publishers now. * That does not automatically make a headline dishonest. It does mean the headline was optimised for a behaviour — clicking — and the article was optimised for a different thing, which is being accurate. When those two goals disagree, only one of them has a live dashboard measuring it in real time. * **And most people never get past it** * The reason this matters more than it sounds: for a large share of the audience, the headline *is* the article. * A study by researchers at Columbia University and the French National Institute (Inria), presented in 2016 and widely reported as the "Social Clicks" study, analysed around 2.8 million shares of links to major news sites on Twitter and found that roughly 59% of shared links were never clicked by anyone — including, often, the person sharing them. It also found a brutal concentration: about 9% of shared links accounted for around 90% of the clicks. * So the sentence written by the person who did not report the story is, for most of the people who encounter it, the entire story. * **How to read a headline properly** * None of this means headlines are lies. Most are fine. But you can defuse the bad ones in about ten seconds: * **Check whether the headline's claim appears in the article.** If the strongest word in the headline — "slams," "admits," "forced," "secret" — never shows up in the text, it was added upstairs. **Watch for the question mark.** A headline that asks "Is X about to collapse?" is usually a headline that could not stand up the sentence "X is about to collapse." **Look for the attribution gap.** "Migrants cause crime wave" and "Minister claims migrants cause crime wave" are the same story with wildly different truth commitments; the second one is the honest one, and it is the first one that fits the column. **Compare the tab title to the page title.** Genuinely — look at the top of your browser tab, then at the top of the page. If they differ, you have just caught the SEO headline and the human headline disagreeing about what happened. **Read the third paragraph.** The nuance the headline dropped is usually sitting right there. * **The useful reframe** * When a headline enrages you, the productive question is not "how could this journalist write that." It is: *which job produced this sentence?* * Sometimes the answer is the reporting job, and the story really is that bad. But very often the answer is the packaging job — a person under deadline, fitting a complicated story into a slot narrower than the truth, with an analytics dashboard glowing at them. * That is not a defence. A misleading headline misleads people whether or not the reporter meant it. IPSO writing a rule specifically about headlines is proof the harm is real. But it does tell you where to aim: the outlet's editing culture, not the byline. The byline is often the second-most-annoyed person in the conversation.