How “engagement” made the web a less engaging place
theatlantic.com
theatlantic.com
And as more and more ads are viewed and not clicked, (there's probably some adtech term for this) the price per view is going to go down, which will encourage more clickbait, which will drive the value of an impression down, etc.
The article struck me as a bit weird, it complains about yellow journalism, but it doesn't really dig into why it's happening. It just declares that shitty articles are written for the metrics, and leaves it at that. As if "metrics" had any intrinsic value. The mapping of metrics to money, as done by internet advertisements, is the actual problem.
While engagement is certainly a way to get more people to view more ads, it's but a tool and symptom of the system it's a part of, not the genesis of it.
I've felt for a while that I liked facebook when it was weirder and there was more diversity in the types of content, but that has really declined in their pursuit for optimizing engagement. Twitter still has this so I appreciate that network quite a bit more for it.
Im taking a genomics class, so I figured I'd look up to see if there are any videos for more information on some topics...
Turns out there seems to be a cottage industry of having the computer read a definition, and post it on YouTube: for example this user. These videos show up often in searches and crowd out some of the more interesting stuff.
When you're browsing around looking at office design projects and photos, you're seeing static advertisements for office furniture and services you might be interested in if you're in the industry. And if you're spending time on a website that consists only of office photos, you are probably are in the group interested in the advertisements.
I would be interested to know whether you have any stats to back this up; if it works this would be great.
To actually judge whether users wanted to see the advert the best metric would be conversions, but that's likely data the site owner wouldn't have (and even with that data, you'd need something to judge it against - for all the talk that most adverts are ignored, the fact remains that some adverts through Google/Facebook/whatever ad-network can lead to direct conversions with a positive ROI).
Not saying it wouldn't be interesting to hear what his typical CTRs are, just that nobody should look at the answer and think you can make significant judgements based on them.
Setting aside what's "average" - are you controlling for the type of advert (both what it's advertising and its format), the type of audience, etc...
Over the years I've seen plenty of examples where CTR and ROI just aren't linked at all. Right now my work doesn't involve a huge amount of this kind of advertising (though I do have experience overseeing 7 figure budgets in the past) so I don't have much great data to pull examples from, but here's one real example from a couple of months ago:
Two not-huge (sub $10k) media buys, both direct with websites (not through networks), both with websites that are well-respected by their readers and by their industry, both using the same adverts. One of the sites excitedly sent us a report about how well the ads had performed, a 5.63% CTR for banner ads, clearly we're popular! The tracking on our side showed that of the just over 15k clicks they sent, 12 converted into users for us. Meanwhile the other site was reporting a 1.32% CTR, sent just over 6k clicks (pricing and order size wasn't exactly identical between the two sites), and landed 700 new users. I know both companies well enough to not suspect click fraud, and I've seen the company that performed badly perform well on other campaigns. But if you just looked at this comparison, there's no argument that the campaign with a much lower CTR was actually hitting an audience that was more interested in what we had to sell than the one with the high CTR. There's lots of possible reasons, from accidental clicks depending on how sites are set up to the audience targeting each site did (both are big enough sites to offer specific targeting within their audience), etc.
(Edit to add that of the two campaigns I compared, both produced a negative ROI from my point of view, i.e. our CPA was far too high - just one was a lot worse than the other. I just pulled up the first spreadsheet I saw from the most recent obvious example I could think of for CTR != ROI.)
If CTR is the only metric you are able to look at, then you can do your best to make judgements on it. But the reason the adtracking solutions that annoys so many people exist is because without proper tracking, it's really hard to actually judge the value of any advertising you buy.
All that said, CTR does still have a big place in the industry, because generally speaking websites (or ad networks) are able to tell potential customers their typical CTRs, but aren't able to give data about further down the funnel (conversions, etc.) because this is data the ad buyers control, not the ad sellers. But it's really important not to think it tells the full story.
Also, by ~5% CTR in your example are you meaning that for every 100 impressions that campaign was getting 5 clicks?
Depending on the ad format/placement and many other variables, it's possible to have a high-ROI advert with a 0.1% CTR, or a low-ROI advert with 5% CTR (or the exact opposite).
The majority of companies which are the best candidates for advertising on Office Snapshots do not have e-commerce stores to point users to as most of the sales in this particular vertical happen via office furniture dealerships.
There are some exceptions, and I've tried some affiliate links in the past, but the types of people I'm marketing at are people who will at some point in the future purchase 100 chairs for a project they are working on and not people who see an ad for a chair and want to buy it immediately.
Wastage
If that is true, the problem will solve itself eventually.
As someone with more experience than most advertisers when it comes to native ads (i.e. outbrain, taboola, revcontent, etc...) I can assure you that clicks and impression costs are going up, not down. As people learn what works and the competition gets fiercer, the costs go up.
The race to the bottom you fear is actually, a race to the bottom of what advertisers can afford to continue advertising. Right now, Native ads are dominated by supplements, which is essentially snake oil salesmen, and also some casino, and other either mass appeal or shady enterprises. Most legit businesses are priced out of the game...Paid Search and Paid Social have done a good job keeping out these low quality advertisers...
Essentially, the race to the bottom of creating clickbait that you are pointing out, is actually a race to the top. A race to the highest margin, most mass appeal advertisers.
If you really think about publisher incentives, they are more likely to create clickbait when the value of clicks goes up, not goes down. At some point of time when prices drop, it becomes easier to make money in other areas...so why bother.
Edit: corrected spelling on a word
This is problematic for me. The more a site knows about me, the better and more interested are the ads. But at the same time, it gets more scary.
Look for an item on Amazon, and you will see it advertised in a lot of pages that you visit. It can be useful to remember that it is there, also tempting. But it is the same feeling that being at home and when you look back there is the guy from the gadgets shop with the camera that you were looking in his hand. Creepy. Scary.
I believed in the promise of the Internet to create a more informed, compassionate, open, and enlightened society and that hope is turning into disillusionment.
Somewhere along the line (probably with the demise of AOL) it was decided that the model of the internet wasn't going to be such that we paid a subscription for the content we consume, but that the model would be akin to freesheets and completely ad funded. It's natural that people who invest in that kind of advertising are going to want some kind of metrics in return, and probably also natural that publishers will try to optimize for that metric too.
The major change is simply economic: that the product stopped being the publication and started being the article. The public turning point for this granularization was when Gawker decided to play per click in March 2008. As this motivation was generally adopted industry wide, it separated a writers incentives from that of a the publication as a product.
We no longer had a quality bar set by brands, and instead a race to the bottom set by individual writers competition. You can see it everything from click-bait headlines to inflamatory articles, all the way to its final form: Fake news written only for views at the price of all other qualities.
As an example, in Spain, Jot Down. Web: http://www.jotdown.es/ and store: http://www.jotdown.es/store
There must have been attempts to do this.
I wonder what's been tried that didn't work.
If you take a look at the homepage through the Wayback archive as it used to appear in 2005[1], 2011[2], and today, you'll see how content disappears and click-baity headlines rise over time.
The Atlantic is very much a part of the problem of "the race towards the bottom" the author describes, and instead of having a discussion about how to fix it and maybe trying different revenue models, it continues to un-ironically have share and tweet buttons at the top of this article.
[1] https://web-beta.archive.org/web/20050210070148/theatlantic.... [2]https://web-beta.archive.org/web/20110731234233/http://www.t...
Writers may not be able to bite the hand that feeds them, but they can nibble.
And when I can see how many people liked what I shared, that seems like a reasonable metric to optimize what to share
It makes sense, on the other hand - an average Facebook reader is more prone to clickbait-y content than a user who specifically targets the front page.
Likes would be a lot more meaningful if randomly the user was required to express the reasoning behind why they liked something. If the user showed a pattern of not being able to express why they liked something, at the very least, other users would not see their vote as part of the count.
This would also likely increase the cost for "like spammers" - but the real value would be that likes were more meaningful.
It eventually becomes a circle-jerk. That is the problem with forcing replies. Voat is a good example of this.
Voting as a signal to quality is as important if not more important than the comment quality. For example, HN as a vast amount of high-quality submission votes and related comments, but their still are a lot of spammy, off-topic, low-value, etc. submissions and comments. Key is to insure in any system that there's a cost to expression and that cost is relative to the value.
Another related issue with voting is that users don't understand the intent of the interfaces use or misuse it to achieve an unrelated outcome.
For example, some users on HN use flags as downvotes, but that is explicitly not what they're to be used for per the guidelines; those users happen to know the system and that the system weights flags in a way that's comparable to downvotes.
"in case it's not clear, this would be a control for moderation, posting it as a comment for non-moderators to view would be optional."
Whereas I believe a downvote function should at least encourage a user to explain why. Otherwise it's impossible to know whether the downvoter believed the comment was inflammatory, offensive, off-topic, or plain just incorrect, which one would assume would be a call to explain as to why it was incorrect...
I suppose it could be argued that in that situation a comment explaining a downvote (that you agree with) would already exist, and that an upvote on that comment would serve a similar purpose to downvoting the comment it disagrees with.
There are plenty of times I disagree with a comment that don't deserve a response, however. Less frequent here on HN, but trolls trolling within the rules (to preclude moderation-based solutions) are an easy example of something that'd actually be better to downvote without a comment.
It absolutely could be, no sarcasm, but it would somehow have to be detached from "this comment is good/bad and should be promoted/demoted". "I agree/disagree" and "I think should be promoted/demoted" are certainly going to be correlated for as long as you are dealing with humans, but hard-wiring that correlation to 1.000 right in the design of the system is not going to produce good results. There is a reason all the successful upvote/downvote sites at least have a culture of explaining to people that the votes are not for agreement, even if it breaks down after a certain scale.
And I've never seen a successful two-dimensional moderating scale. (I've been keeping my eye out, and I'd invite anyone who could show me one to reply and tell me about it, please.) Slashdot came closest with what was about a 1.2 dimensional system where you could upvote/downvote with a set list of one-word reasons, but there was still no "-1 Insightful" or "+1 Contrarian Opinion You Need To Hear" or anything. (Capping at +5 also limited things.)
Can you give an example in what situation this could be useful? When you disagree, there are many ways to disagree. If you just want to drive-by disagree, without making a proper argument, then perhaps it would be better (for everybody) if you just didn't engage at all, and let other people deal with the situation.
Think about who you're doing the moderation for: other people. It's your job, as a moderator, to explain and educate well-intentioned people that you disagree with why do you disagree with them, or what do you perceive to be wrong with their comments. Then, as an added bonus, community can judge also your moderation criteria, by moderating on your counterargument or explanation.
> There are plenty of times I disagree with a comment that don't deserve a response, however.
For obvious trolls, there should be flagging and eventual deletion.
Update: Actually, it could be useful to "disagree but have nothing to add" when you already agree with some of the responses. So, IMHO, the rule should be, you can downvote a comment only if you either (a) added a response or (b) upvoted an existing response.
some forums I frequent added various like/dislike/sad/funny counters just to get rid of the noise of hundreds of contentless like posts.
Agree, useful/helpful, disagree, not relevant, spam, illegal, inflammatory, etc...
Then you can apply different weights to different votes and maybe even allow people to sort by the different values. Even some "dark" UI patterns might work well like completely ignoring some values (like disagree) and leave them there as a way for some to express that without having it change anything.
I haven't given much thought to the details, but it seems like this could help on the surface.
I feel like a major problem would be getting people to actually vote. It's already hard enough IMO and I frequently forget to vote on comments which I very much enjoyed and we're very relevant (like your own until this moment), and needing to have many options to choose from and possibly multiple steps would make that bar even higher which means less votes overall. Also getting across the meaning of what each vote really is meant to represent would be tough as well...
Now limit the number of times I can vote within a given time period, and you've reinvented Slashdot.
Encourage people to vote as many ways as possible, as many times as possible, as much as possible.
Check off not-relevant, agree, inflammatory. Or maybe disagree and relevant.
No limits, maybe some weighting, but no limits.
Just because some of the ideas have been done before doesn't mean it's just a reinvention.
Here's the magic of the Slashdot system: not everyone has all day to click voting buttons, but some people do. So what you'll end up with your proposition is voting dominated by people that make a career out of hanging out on one web forum, with the conversation consequentially dominated by them.
The Slashdot system is more like jury duty: go in, do your civic duty by casting your five or ten votes, then go on about your day. I don't want a justice system dominated by people with nothing better to do than sit on juries every day.
I feel making the voting process easier by creative UX would help, and then weighting can take care of the rest. If a power user votes 100X more than the average, have their votes count 100X less, or some kind of other method to weight them back into normalcy and keep the system from becoming a "most active dictates what the platform is" kind of thing.
Slashdot sets lower and upper bounds on comment scores and lets readers set a threshold. Promoting good posts has more impact than demoting bad ones, so you'd probably just ignore all but the worst off-topic posts.
Making someone's votes count less would strongly discourage voting more often.
But the nice part is you can control how you approach this weighting.
Perhaps every days votes is normalized to 10. So if I only voted once it counts 10x, but if I voted 100 times each would count .1x.
There are tons of ways to implement it so the feeling of uselessness doesn't apply, after all, qt the end of the day we are pretty gullible bags of meat.
That might not be a bad thing. I'm just saying it works against your goal of getting people to vote as many times as possible.
You owe it to yourself to at least study Slashdot's system before putting too much thought into creative UX, dark patterns, and complex weighting. It isn't the last word in comment scoring, but as someone else pointed out, you could probably fill a book with all the lessons that shaped it.
Willingness to converse is the prinary metric; upvotes fuck with that by also pushing the orthogonal metric of how much people like the post itself (regardless of how worthy of response it is).
And ofc, you're trying to tack on like 6 more orthogonal metrics. But whats wrong with letting people vote with their feet?
Not to mention that it encourages controversial content as it generates more arguing (read discussion).
It's just not a system that I believe works well, especially with threaded content like on HN or Reddit.
You're completely ignored, but only in the sense that people skim your post... since the posts all live on the same plane, the comment must still be skimmed over (in order of time). But its not the system, or some subset of users ignoring you (by hiding your post automatically, by downvotes or a lack of upvotes)... it's the users choosing, individually, to ignore you.
And AFAICT, upvote-based systems have a higher tendency to post bullshit, in the form of trivially community-acceptable statements, or simple nonsense like "I agree".
However, I'm also of the opinion that most interesting comments on both types of systems are arguements and refutations (by knowledgeable people), info dumps (by knowledgeable people), or histories (by experienced people).
The latter two are supported by upvotes, but the goal of upvotes is to utterly extinguish the former.
I also say this while happily wearing my Readertron t-shirt. He's right - the conversations we had on Reader were very, very good.
It's probably the best approach I've seen.
All of which is to say that the current design of Slashdot's system contains an enormous amount of valuable information about how to defeat trolls. You could write an entire book on all the lessons Slashdot had to learn the hard way along the road to making it what it is.
Which is why it's so tragic that their experience has been more or less utterly ignored by the community sites that came after it -- looking at you, HN! -- who universally went with simpler, naïve systems like the one Slashdot started with, with the completely predictable result that they were as much a green field for trolls as Slashdot used to be. So much heartburn, all of which could have been avoided if we as an industry had an institutional memory that ran farther back than two or three years. What a waste.
HN isn't bad, because of the technical measures that prevent comment spam and the admin moderation.
The issue is probably less in the interface presented to users and more in the way the backend accounts/tallies scores, and then uses that data to determine the placement of the content.
The "direct democracy" approach of just taking the raw counts and sending them through a pre-set formula results in groupthink and echo chambers, we know that. As you know, Slashdot's saving grace was not that they required posters to say "Insightful!", but that there was reasonable human curation supervising the system.
Reddit takes a laissez-faire approach to moderating content, leaving it in the hands of subreddit moderators, and afaik does not attempt to modify non-spam vote counts (except for a few exceptional subreddits).
While you could definitely write more objective algorithms for ranking content, the question is really "Content is better according to whom?" Because the fact is that most people do not want to hear from the other side. You will not have a successful community where one story is "Vote for Trump" and another is "Vote for Clinton". You will have factions that try to tear each other apart (reddit is also an example of this, but not because the ranking is trying to be fair).
Not only do most people not want to be "open-minded", aka "experience and suppress cognitive dissonance until it's rationally justified", most people are frankly incapable of doing so. They do not subscribe to the FB pages of both Local Dems and Local Reps because they like to know both sides of the issue. They subscribe to one or the other, or neither, because they've selected a tribe and completely intend to stay with it. Information contradicting or deriding their tribe is a vexation, and they won't tolerate it.
The implication for content ranking is that if you want an intelligent ranking system for interesting, challenging, or in-depth discussion, you want something completely different from what the public demands.
And that really gets at the crux of this whole issue. We look at "native ads" promoting fake dating sites, penis enlargement pills, and gambling, and we scoff. These work because many other people don't scoff, they say "Yes, I want sex and money" (and who can really blame them?). While they may be socialized to be embarrassed about it, they take advantage of the privacy the internet offers them and click anyway.
Now the question is, you have a business that has to make money. Like it or not, which market is the better target? The discretionary, measured, judicious reader, or the person who will impulsively click through to a site promising him an elongated penis?
The system was great! You could set a filter, and most of the trolls would disappear (and there were a LOT of trolls).
My crazy dream: Something like HN or Reddit, where hovering over the vote arrow brings up a small list of reasons.
^ (Agree/Correct/Insightful/Funny)
v (Disagree/Incorrect/False/Flamebait)
All votes are counted as normal and the algorithm everyone sees is otherwise unaffected, but the magic happens where you give users the ability to sort by score and reason, rather than just score.
This cancels out some of the hivemind suppression of unpopular or minority viewpoints, since most people are going to just click the bare button (which registers an agree/disagree vote). With one click, you can see insightful-rated posts which might be otherwise buried at the bottom of a thread.
Similarly, moderators can sort by flamebait and perhaps take action.
But that could just end up being a lot of noise, and it doesn't really address the social/psychological "content" creation problem the article focuses on. I usually want to read actual good things, not why someone thought something was good.
The article goes too far but there is something to be said about its core argument.
I don't think the Like button or emojis are good at recognising Value. A popular extrovert with 1000 fb friends uploads a selfie and gets 100 Likes.
Meanwhile a quiet but conscientious person uploads a before and after picture of a litter-pick they have undertaken and they get 2 Likes from their 50 friends.
There has to be a better way of recognising and incentivising the latter activity across social networks.
Also, I think a problem is one person can only send one like.
Returning to my example, I'd like to be able to give the litter picker 20 likes, and maybe the Selfie uploader 1 like.
I think there has to the option of another accreditation technology which runs across social networks.
It's an area I'm interested in and looking to do my next project on...
Slashdot hands out Moderation points sometimes, and asks people to meta-moderate sometimes -- both different from Likes.
I don't visit Slashdot anymore, but for the period between 2004-2010, I had virtually unlimited mod points.
This is why privacy matters -- deep consumer data leads to 'binge products'. This is the business netflix is in. It's possible to create cocaine using this model but it isn't a great way to make something awesome and new. 'What's new' will always come from the fringes.
I have limited time to browse, and it is much faster to "filter" content by outsourcing that to you guys.
I thought the article was going to be about how every web designer interrupts whatever it is you're there to read/see/etc. Multiple times. Instead we get yet another Medium-quality thinkpiece extrapolating someone's personal experience.
At any rate, the title here should match the article's.
So the blog version of "Stairway to Heaven" will become as rare as Bowie and substantive songs will be as tears in the rain.
Disagree. I think that is one of the main reasons of why podcasts haven´t become more popular. It is way harder to see people sharing just one episode or a piece of an episode in a conversation. Usually, they just say they listen to a particular podcast and you should go and listen to someone talk for 30 minutes before decided if you wanna follow him. And that particular episode that you chose could one of the worst that the podcaster has produced.
Individual episodes matter, especially to bring more people to your subscribed list.
A personal example it is how I discover the best, IMO, podcast: 99%invisible. I was reading the most upvoted comment about a shared video on Reddit about how Taiwan collect their trash and he was suggesting listen to this podcast episode http://99percentinvisible.org/episode/separation-anxiety/ After listening to the episode, and to his sponsor to this particular episode, I subscribed.
And brands cares way more about engagement than about the numbers of the subscriber. You could have a big number of subscribers that doesn't react to anything that you produce. And you could have a small number of subscribers that really like your content and, especially, trust in your endorsements. And the best way to metric this is seen the numbers of sharing, likes, and comments in the content that you produce. Engagement numbers is more important than plain numbers, for everybody.
http://www.adweek.com/digital/micro-influencers-are-more-eff...