What is clickbait? It is a headline or preview built to get a click by hiding the context a reader needs, exaggerating the claim, or promising something the page doesn't really deliver. Does clickbait work? The evidence is mixed. Some headline choices clearly change how often people click, but the studies don't show that misleading or overly withholding headlines reliably earn more useful clicks, and a few find no advantage at all or fewer shares. In 2026, the big distribution platforms also say in writing that they don't want it. We'll grade the evidence as mixed, and that grade is about the general claim that clickbait outperforms. It isn't a claim that curiosity is a bad reason to read something. The idea to hold onto through the rest of this piece is the gap between attention and value: a click is something you can count, but it doesn't tell you the reader found the answer, trusted you, or came back.
What people mean when they say clickbait
The word gets used in two ways, and a lot of the arguments about it come from mixing them up.
In everyday marketing use, a preview counts as clickbait when it plays on an information gap and also does at least one of three other things. It withholds what you'd need to understand what the content is about, it exaggerates how big or important the thing is, or it implies a payoff the page never provides. Google's guidance for Discover describes almost exactly this, listing misleading or exaggerated preview details and withholding crucial information as tactics to avoid. Meta's description of clickbait links is similar, and it centers on posts that create misleading expectations about the page they point to.
Researchers often use a wider definition. Some studies count any headline with a question, a list number, a superlative, or a word like "this" as clickbait, even when the article could honestly deliver what the headline says. That matters when you read the results. A study showing that one of those features helps or hurts clicks doesn't tell you whether deceptive headlines help or hurt.
It also helps to know what clickbait is not. An outrage headline leans on conflict or anger, and it isn't automatically clickbait. Engagement bait asks people to like, share, or comment, which is different from tempting someone to open a link. And a rule in an advertising policy isn't automatically a rule for your unpaid editorial content, which we'll come back to.
Clickbait vs curiosity headline: where the line sits
Curiosity is a perfectly good reason to open an article. The trouble starts with how the gap is built, so it's worth being specific about what a good one looks like.
A useful curiosity gap tells the reader what the subject is, so they can decide whether it matters to them, and then leaves a worthwhile question open. Because it points toward a real answer, the reader who clicks gets what they came for. Curiosity gap headlines go wrong in two ways. The first is the empty teaser, which gives so little context that no one can tell if the page is relevant. The second is the overpromising teaser, which sets up an expectation the article can't meet.
That's why the clickbait vs curiosity headline question is about the relationship between the preview and the page, and not about punctuation. A truthful question can make a fine headline. A flat, specific statement can still mislead if the page doesn't back it up. The title, the snippet, the image, and the landing page all add up to one promise, and the reader judges the whole thing when they arrive.
One more point on the other side. A headline that gives away its entire answer can leave people with less reason to click. That's not permission to hide the topic or the evidence someone needs to judge your claim. What you can leave open is the answer itself.
Clickbait examples, and how to tell them apart
Real clickbait examples from research are useful here, and so are a few we made up to show the range. The made-up ones are illustrations, not tested headlines and not real campaign results.
Two research examples are worth knowing. The Center for Media Engagement compared the restrained "Fed chair says banking regulations good enough" with the more conflict-heavy "Fed chair slams critics, says banking regulations tough enough." That was an outrage manipulation, so it doesn't isolate deceptive clickbait. Molina and colleagues used "Secret Deal Allows Iran to Expand Nuke Program" as the control headline they then rewrote in different ways. A headline used as a research stimulus isn't a verified news claim, so treat these as stimuli and nothing more.
Now four illustrations of our own:
| Preview | How to read it |
|---|---|
| "You won't believe the one change that transformed our results" | Risky clickbait. It doesn't name the subject, it inflates the implication, and you can't tell if it's relevant to you. |
| "Why did the signup-form test fail?" | Can be honest curiosity, as long as a real test failed and the article explains why. It names the subject and leaves the answer to learn. |
| "We tested two signup-form versions. Here is what changed" | More descriptive. It works if a real test and change are documented, and it doesn't hint at a specific outcome. |
| "One small edit doubled signups" | Misleading unless a real doubling was measured, the cause is justified, and the page gives the context. A specific number doesn't make an unsupported claim honest. |
If you want a quick test to run on your own headlines, ask two things. Can someone tell what the article is about from the preview alone? And would a reader who lands on the page agree that what was promised is really there? Then look at the implied size and certainty of the claim, any conditions you left out, and whether the image and snippet change the promise the title makes.
The evidence that curiosity can win clicks
Some headline choices do change how many people click, and the best data on that comes from Upworthy. Between 2013 and 2015 the site tested different headline and image packages on its own visitors, randomly showing each variant, and the Upworthy Research Archive kept the impressions and clicks. That's a stronger basis for comparing options than looking at which stories were popular across different publishers, because the same story was tested against itself.
A 2025 analysis by Le Quéré and colleagues used this archive to look at how concrete a headline is and how that relates to click-through. What they found depended on the comparison being made. It doesn't say vagueness always wins, and it doesn't say the most detail always wins. The same paper reviewed nine earlier lab and field experiments on curiosity-style headlines compared with summary headlines. Three found a statistically significant negative effect, two found a positive one, and four found no effect that could be told apart from chance.
Banerjee and Urminsky reported something more encouraging in a preliminary draft from January 2021. They saw forward-reference curiosity cues linked with better relative click-through in a subset of the Upworthy tests. It's worth treating that as qualified support. It's a draft, it covers a subset, and a headline and image package can differ in several ways at once, so you can't credit the gain to withholding information alone.
There's also a caveat on the archive itself. In June 2024 the archive team reported randomization problems in 22% of its A/B tests. They also said that working with the authors of six peer-reviewed studies showed those earlier findings weren't materially affected. That doesn't invalidate the archive, and it isn't something to ignore either.
So what can you fairly say? Certain curiosity cues can help a particular headline win a particular comparison. You can't say what fixed lift a marketing blog should expect in 2026, and you can't say deception was the ingredient that worked.
The evidence against a dependable advantage
The most direct check on the "it works" claim is a controlled experiment by Molina, Sundar, and Rony in 2021. It was one of three studies in their paper. They took one political headline and rewrote it systematically, then assigned 249 participants across eight conditions: seven that each featured a single clickbait characteristic, and one control. They found no statistically significant difference in "read more" clicks or in sharing, and none in how curious or credible people rated the headlines.
That's evidence against a dependable advantage from those isolated features, in that setting. It isn't proof that every headline performs the same. There was one topic, the participants included students and online recruits, and the study didn't test every combination of features or a real marketing campaign.
Sharing tells a related story. A study called "Did clickbait crack the code on virality?" combined an experiment with 150 respondents and an observational analysis of 19,386 articles from 27 online publishers, along with their Twitter engagement. Before adjustment, clickbait-labeled articles averaged 80.61 shares and non-clickbait articles averaged 155.65. After propensity-score matching, the gap was 48.58 fewer shares for the clickbait-labeled ones, on average. Matching accounts for measured differences, but it can't turn observational shares into a randomized estimate of what a headline does to clicks. The authors discuss perceived manipulative intent and lower regard for the source as explanations. That's a proposed explanation from their research, and it doesn't prove a chain from clickbait to lost trust to lost revenue. People can also share an article without reading it.
A third study sits next to this topic without being the same. A May 2019 report from the Center for Media Engagement described a May 2018 experiment with 1,535 U.S. participants who saw headlines and articles about immigration or banking regulation. Outrage headlines lowered people's stated intentions to click, comment, pay, or return, compared with less outraged versions. It didn't report a real click-through rate, and the tracked likes, shares, and comments didn't change significantly. Its finding that outrage articles led to more "fake news" perceptions is about outrage coverage, not every curiosity headline. It's useful for questioning the idea that stronger emotion always means better engagement, but it isn't a measured conversion loss from clickbait.
Here are the main figures side by side, with what each one does and doesn't tell you.
| Figure | What it measures | What it doesn't measure |
|---|---|---|
| 249 participants, no significant read-more or share difference (Molina and colleagues, 2021) | Response to seven single-feature clickbait variants against a control for one political story | A population-wide zero effect, a conversion rate, or 2026 feed performance |
| Nine earlier experiments: three negative, two positive, four null (as reviewed by Le Quéré and colleagues, 2025) | Direction of reported effects across different curiosity-style comparisons | One pooled clickbait uplift, or a single definition of clickbait |
| 19,386 articles from 27 publishers, 48.58 fewer matched shares on average | Association with Twitter shares after propensity-score matching | The randomized effect on clicks, conversions, or long-term trust |
| 1,535 participants (Center for Media Engagement, 2019) | Responses to outrage versus less outraged coverage | Real click-through rates, or effects of curiosity headlines that aren't outrage |
| Upworthy headline tests, 2013 to 2015 | Historic random exposure to headline and image packages, with impressions and clicks | Present-day search, social, AI-answer, or B2B content outcomes |
What Google and Meta say in 2026
The experiments mostly come from older news and social environments, so the platform documents matter more for current constraints. They don't give you a click-through benchmark. They tell you what the platforms say they want.
Google's Discover guidance recommends avoiding clickbait that inflates engagement with misleading or exaggerated titles, snippets, or images, or by withholding crucial information needed to understand what the content is about. It suggests titles that capture the essence of the content. Being indexed can make a page eligible for Discover, but that doesn't guarantee placement, and the guidance doesn't give a numerical ranking weight or promise a site-wide penalty. Google's Discover content policy is a separate document, and it says preview content can't promise details that the underlying content doesn't contain.
On February 5, 2026, Google announced a Discover core update, and one of its stated aims was "Reducing sensational content and clickbait in Discover." It began with English-language users in the United States and was set to expand elsewhere. Google also cautioned that individual sites might gain traffic, lose traffic, or see no change. That announcement doesn't say every curiosity headline was demoted, and it doesn't name a specific penalty.
Meta describes clickbait links as posts that create misleading expectations about the linked page, including by withholding information or using sensationalist phrasing. Its examples include "You won't believe..." and "You'll never guess...". Meta says its feedback favors headlines that accurately reflect the article so people can make informed choices, without giving a percentage reduction in reach to quote.
Google Ads has its own policy, which doesn't allow ads that use clickbait tactics or sensationalist text or images to drive traffic. Its examples include "Click here to find out" and "You won't believe what happened" when you have to click to understand the ad. That policy covers advertising, so it isn't a blanket ban on every question headline in your editorial content.
A common piece of industry shorthand says the algorithm penalizes curiosity, and that's too broad. These sources separate real curiosity from previews that are misleading, exaggerated, or missing something crucial. At the same time, none of them shows that honest curiosity raises rankings, earns more AI citations, or produces conversions. So the grade holds at mixed, and the platform side adds a reason to be careful more than a reason to expect a lift.
What this means for your headlines
If you run a marketing team, the practical answer is fairly calm. You can keep honest curiosity, and you can drop previews that hide the topic, overstate what the evidence shows, or promise an answer the page doesn't hold.
When you want to know whether a curiosity version beats a plain one, test it against a specific, truthful alternative for the same article in the same channel. Decide first what you're measuring. Click-through rate is clicks divided by impressions, and engaged visits, relevant signups, return visits, and complaints each answer a different question. Keep the landing page the same and the audience comparable where you can. A higher click-through rate paired with worse downstream results isn't a clean win. These are ways to measure fairly, not published thresholds, and nothing here promises a particular test will show a lift. We cover the traffic side in more depth in our piece on clickbait and organic traffic. The wider set of numbers worth tracking is in a guide to measuring content performance beyond traffic.
Try this on your next few headlines. Write a curiosity version and a descriptive version, check each against the two questions from earlier, and only test the ones that pass. Then look at more than clicks when you read the results. If you want a structure for it, a 90-day content experiment is one way to run it. For what happens after a big spike, see how the traffic decay after a viral spike plays out. If AI summaries sit between you and your reader, what makes people click through from an AI Overview is a related question.
What would change this verdict
There are a few things we don't know. Nobody has published a credible, generalizable 2026 percentage for how much clickbait raises or lowers clicks across Google Search, Discover, social feeds, newsletters, and B2B blogs, so it would be wrong to make one up. Nothing measured here shows whether an honest curiosity headline improves AI-answer citations, and citation visibility shouldn't stand in for evidence about clicks or reader satisfaction. A universal causal link from a clickbait headline to lower trust or revenue also isn't established. Trust, clicks, stated intentions, shares, and platform eligibility are separate outcomes that were tested in different settings.
Date matters too. Much of the experimental evidence comes from Upworthy's 2013 to 2015 environment or from later news studies, and Google's documented 2026 Discover change makes it risky to carry an older feed result forward.
What would move the grade is well-powered, current, channel-specific randomized testing that compares truthful curiosity, misleading omission, and descriptive headlines for the same content, and that measures exposure, clicks, reader satisfaction, conversions, and repeat behavior separately. That's a description of what's missing, not a study that already exists.
If you'd like to see how your own pages show up when people ask AI engines questions in your space, DeepSmith tracks that, and you can start a free trial to see real data before you pay.
