100M Posts Analyzed: What You Need to Write the Best Headlines
buzzsumo.com
buzzsumo.com
Or that means you're measuring the completely wrong things about headline authoring, because the data have no stationarity at all.
They allude to this, but it would appear that the only thing that matters is Facebook's editorial, laundered through an algorithm. So maybe a more valuable article would be hacking into Facebook and just finding out what it is they idiosyncratically value in a headline.
Unfortunately, tsundokuists won't read it. Meta.
PS: Earth - art = Eh.
I honestly instantly re-experienced this episode when reading that title; there's nothing wrong with analyzing data, in fact, I probably would and the article isn't suggesting a machine write headlines, but it feels like writing for humans should feel human and not be machine optimized...
I realize I've just been rambling, probably connected with no one else on the forum but I've typed all this, so I'm submitting it.
Individuals checking individuals: yeah, you can detect non-intelligence. But if I told you half of all buzzfeed headlines are machine-written... would you believe it?
Honestly, yeah. Buzzfeed headlines aren't exactly the most creative or hmmm...deep?...meaningful?
There's only so many ways you can write
'10 reasons why x does y and that's bad'
Or
'Omfg x just happened, better y right now.'
Or it could just be a fluke random AI achievement. It will still be humans judging it as "more human than anything humanity could write" or not.
Made me think of:
https://www.youtube.com/watch?v=OyO18QrJ3zw (White Zombie - More Human Than Human, referencing PKD's "Do Androids Dream of Electric Sheep?")
But I agree with your point. It’s not bad to write with computer consumption in mind, unless it loses the core focus which is that a human needs to read and connect with the content, and that probably take a human to do it (at least at the moment).
I've never been able to articulate how disappointed I was that the writers made Picard retire to the family vinyard. Nothing matches the thrill of preventing a galactic war but we got little pictures of other interests he might have and just sit around growing grapes. Literally going back where he came from after an incomparable career.
The holodeck episodes also hinted at other of his interests. I recall a sailing ship and something about being a detective?
They did something very similar with Kirk and Riker although in Riker's case they gave a plot device for him making that choice.
> The crew refuses to hand over control to a computer
Star Trek universe lives in this strange place of being on the very cusp of technological singularity, edging on the threshold and yet not crossing it. Both in TOS and TNG era, their computers are so advanced they keep accidentally creating sentient AIs[0]. And yet, whenever that happens, the AIs are met with apprehension; Starfleet, in particular, would like to see them gone. In some cases, the protagonists stand against the zeitgeist and fight for the rights of digital sentience, but in others, they're desperate to put the cat back in the bag.
In-universe, I explain it to myself like this: the society of Star Trek is afraid. They do not understand their technology at all. All the scientific and engineering knowledge people in Star Trek demonstrate so frequently[1], it's all in context of what computers tell them. They almost never look at raw data, they almost never deal with raw reality. They grudgingly accept this state, but do not want to make that last step and actually let computers run things.
Out of universe, it's obviously because Star Trek was meant to be a humanist story - the adventure of humanity, exploring the universe and becoming a mature, respectable, good people. Anything that threatens it, anything that would immediately turn humans into NPCs, dethrone them from the role of protagonists, is shunned in the series. That's why both artificial intelligence and genetic augmentation receive such negative treatment, a glaring exception in the otherwise inclusive and forward-thinking show.
Circling back a bit - I think our real-world society is approaching the level of "too smart technology", and we're going to become afraid too. And unfortunately, developments like this article aren't the kind that would happen in a Federation research facility - they read as something straight from Ferenginar. Our civilization is not proto-Federation, it's more of a weird blend of Ferengi Alliance and Cardassian Union.
--
[0] - See e.g. "Emergence" (TNG: 7x23) and "Elementary, Dear Data" (TNG: 2x03) for Enterprise D's computer forking off sentient AIs, "Evolution" (TNG 3x01) and "Quality of Life" (TNG 6x09) for Federation robots accidentally becoming sentient, "The Offspring" (TNG 3x16) for how easy it is to fork Data, "The Ultimate Computer" (TOS 2x24) for what happened when they let a too-smart AI run the ship for once - and that's not counting all the cases where sentient life was created with involvement of alien entities or objects.
[1] - One thing I love about this show is that competence is considered table stakes. Everyone, whether they're Starfleet, a Federation civilian or a member of non-Federation species, is educated, curious, good at what they do, and expects the same qualities of everyone else. It's a breath of fresh air compared to the real world.
In my head, the only true successors to Star Trek as it was up to 2005 are 1) The Orville, and 2) Star Trek: Lower Decks. Apparently it turns out you have to market something as satire to get a shot at exploring more thoughtful topics.
I wonder if the authors of The Expanse were consciously or unconsciously riffing on that. Writing truly original stuff is hard.
Anyway, nice in-depth article. The results will surprise you. Especially point 6.
oh.
I wonder how this line would fare in their headline analysis.
I see what you did there.
Like most art, we commercialize it until our categories are diffusive.
But it's still there.
eg. More headlines may be using the number 10 than 4, so 10 is more likely to be the most trending headline.
Similarly, in the lottery, the frequency of winners who picked their own number is dependent on the frequency of people picking their own number.
That said, a boxplot with 25th and 75th percentiles would likely indicate there is a heavy skew, as tends to be the case with social media data.
The result for headlines of 65 chars - shared 50,000 more times than 60 chars or 70 chars - seems too incredible to occur at random and suggests instead that a popular news source has implemented a 65 chars policy.
[Edited to note: Yep. YouTube is dominant as the popular publisher in this review, and truncates headlines at 66 chars - that's what this article observes]
>>> headline = "100M Posts Analyzed: What You Need to Write the Best Headlines"
>>> len(headline.split())
11
>>> len(headline)
62
So close!These are not the same. The fact that they're so commonly conflated is a major problem.
You could have a "good" headline that catches the attention of a large number of people who don't really care, or a "bad" headline which catches the attention of a small number of people to whom the article is very relevant and really care.
Which do you want to optimize for?
>The point is getting someone to read the article doesn't make a headline "good", it just means someone read the article.
It actually does. The headline did its job.
>Which do you want to optimize for?
That is a false choice. Why are these the only two options?
Just because you don’t have a good metric for something doesn’t mean that what you can measure is better.
A metric can simply lead to bad results, and thefore be a bad metric.
You're conflating content targeting with headline writing. Those are two separate points.
>Just because you don’t have a good metric for something doesn’t mean that what you can measure is better.
Certainly, if a metric is 'bad' in that it is not producing results, nobody wants to waste their time and keep using it. However, the engagement metric is producing results for many folks. Do you disagree with that?
>A metric can simply lead to bad results, and thefore be a bad metric.
Anything "can" lead to anything. That doesn't really make for much of a discussion without data to examine.
No.
>Just because you don’t have a good metric for something doesn’t mean that what you can measure is better. Certainly, if a metric is 'bad' in that it is not producing results, nobody wants to waste their time and keep using it. However, the engagement metric is producing results for many folks. Do you disagree with that?
This is meaningless to agree with or disagree with since the value of the results is what is in question.
>A metric can simply lead to bad results, and thefore be a bad metric.
> Anything "can" lead to anything. That doesn't really make for much of a discussion without data to examine.
So you agree that the metric could be bad.
How are you judging the value of the results? I am not understanding your point here. Again, back to my original question, please propose alternate metrics, otherwise we're just arguing over minutia that misses the meat of the discussion.
I know.
> Again, back to my original question, please propose alternate metrics
That’s not actually necessary in order to understand what I’m saying. In fact it would be a distraction.
That’s not really how it looked earlier in the thread.
You seemed to be strongly defending the idea that engagement is good, and not even accepting that there could be a problem.
Perhaps that’s a misreading of your intention.
Well, Hitler was popular too.
Given that definition, I may not go read the article because it doesn't interest me. It is still a good headline. I didn't waste my time. On the other hand, if I am interested in the content, I would have read the article and would not be irritated that I had been mislead about the content. The metric being used in the article here in no way leads to this definition of a "good" headline. More likely the opposite.
See:
https://en.wikipedia.org/wiki/Campbell%27s_law
https://en.wikipedia.org/wiki/Goodhart%27s_law
https://en.wikipedia.org/wiki/Perverse_incentive#Cobra_effec...
“Man bites dog” is appreciated by Post readers.
"Prime Minister, what about the people who read The Sun?"
"Sun readers don't care who runs the country, as long as she's got big tits."
I’m waiting for the first paper to reinvent itself.
Meaning the digital, and the print, is so good it will be something I need to subscribe to.
Maybe those days are gone? I was just thinking would I pay for a HN subscription. A site I have viewed since day one. Right now—-no.
I wish news and mass-sharing were banned from social media, because it's too often low-signal noise like celebrity gossip, contrived/falsified outrage, or some new movie or product. Garbagé.
ShowHN: Mono(te): An offline first, Turing-complete, blindingly fast notes app written in 23 lines of (Rust) code. Oh, and it respects your privacy, and it's Open Source.
I was thinking of that post or similar. Thanks for finding. I struggled w/ the decision to use Rust or Elixir.
The null hypothesis is that people share more ten-word headlines simply because there are more ten-word headlines. If you want to show they like to share them more, you need something else, like data about how many headlines people looked at versus shared.
~~That length is just shorter than recommended length for a commit message (72 characters if I recall correctly).~~
Compare that to 50 characters for a commit message’s subject, and 72 characters for each line in the body.
But hey, I clicked.
One thing is if I could understand it and another thing is if the author's writing skills are good enough to advise others about writing.
BTW, the submitter (or a moderator) seems to agree with me (check the current HN-entry title).
Oh wait nm, there's some art and some science to it by evaluating reader interest fashions.