One chef's-knife brand gets 31% of its Reddit recommendations from 5% of accounts
Does Reddit have an astroturfing problem? What the data suggests
A data analysis of six knife subreddits found that 49 accounts — the top 5% by brand-heaviness — wrote 11.3% of brand mentions in buying threads, versus 7.9% expected by chance. The concentration is strongest for a chef's-knife brand that received 31.2% of its buying-thread mentions from this tail. However, full Reddit histories of 23 tail accounts look ordinary: they are 4.5 years old, post across 66 subreddits, and name seven knife brands, not one. The author notes the data cannot distinguish paid campaigns from loud fan bases.
That is not what a warmed, single-purpose account looks like. It is what a person who is on Reddit a lot, and who has strong feelings about one knife maker, looks like.
- nomilk
The biggest problem I have with reddit is mods can slowly but systematically ban users who express unfavourable opinions, and over time, anyone casually browsing could mistakenly believe those views are unpopular, whereas in reality everyone with that view (everyone who spoke it, anyway) was banned.
To give a simple example, one video game franchise subreddit bans any negative discussion on the most recent title. Over 5-10 years, this shielded them from negative feedback. But the franchise has tanked because core users weren't listened to, and they ended up making something nobody wanted.
- nomel
> Thin accounts: few comments, low scores, no real standing in the subreddit.
> Young accounts, or accounts with histories that are hidden or wiped.
Nope, this is no longer an indicator of a bot account. Many that I've looked into will have active posts to local town/region subreddits (often times multiple!) and different sports subreddits, sometimes getting unusually high amounts of karma in them (which suggests a network).
Turing test is looong dead, so no reason to assume the small fractions of a cent to make comments, to get a reasonable history, isn't being taken advantage of.
- robotresearcher
> This article was drafted with AI from my outline
Maybe just post your human outline and the data, where the data processing and viz is AI assisted if you like.
The synthetic prose is not a value-add over your outline. It takes time to read, and reduces the bit-rate over your intentions.
- DharmaPolice
There have been a few cases posted to the theoryofreddit subreddit where it's obvious that there were sets of accounts which were blanket recommending specific products a disproportionate amount of time. The user accounts involved were often banned shortly afterwards (or were deleted). These were (upon examination) really obvious though, often promoting extremely niche products using very exact terminology.
This I think leads to a sort of toupee fallacy - where because you can spot the really bad examples you feel confident that you can identify the much better, more subtle examples. If an account which has only made 50 comments (all of which are very generic) makes 20 comments in a single month explicitly recommending ImageGeneratorDeluxeProMax then sure, you can guess they might not be legit. But if an account runs for a longer period of time and makes a lot more general organic contributions then if 1 in 500 of their comments are in favour of a particular product then it's harder to say.
I think ultimately account age is the only thing that will be difficult to replicate at scale. I know people buy older accounts but there is still a fundamental difficulty in obtaining them in sufficient numbers. And yes, obviously that will mean disregarding the many millions of legitimate accounts that are under (say) 2 or 3 years old. There is also no easy way of hiding accounts under a certain age when searching Reddit via Google.
By allowing uses to hide user profiles Reddit has also […]
- zergrush
This is why Reddit is a low value high effort distribution channel. When your userbase is hostile and suspicious constantly to any commercial introductions, advertising there becomes just an exercise in futility. The demographic also is problematic on Bluesky. Same on hackernews. Astroturfing is certainly a response to this friction and in turn it only creates more of it.
Also specifically with Reddit they have allowed hiding comment history which makes it even harder, I imagine they did this for boosting ad business
There are much better platforms to spend your effort on like X. X requires many users to purchase a subscription account and thus acts as sort of a litmus test (people who can afford low amount of money per month).
- yayachiken
My favorite anecdote for an astroturfing suspicion I had was in German subreddits.
Whenever somebody was talking about rolled oats, without fail, somebody would chime in that Kölln brand rolled outs were strictly superior and they would gladly pay the 4x markup for them.
It is hilarious because rolled outs are probably one of the most fungible goods out there. Also the comments eventually just... stopped.
Human perception is weird like that so even identical oats may taste better if the packaging is different, and also it might have been a memetic thing, but I am still somewhat convinced that it was a marketing campaign.
- BatteryMountain
It's had a problem for 10+ years. Not just astroturfing, but proper propaganda pushing on various topics & countries. It quite obvious as an 'outsider'. Then there is all the ai bots/posts that are being fine tuned by humans (deliberate incendiary posts, logical fallacies, which then get corrected by humans). Then the yearly cyclical repost of the same content and comments.. after a few years it become quite clear that certain content gets reposted over and over, by bots or staff. That is unusable.
- alun
It was clear that Reddit had an astroturfing problem, even pre-ChatGPT.
Funnily enough, Reddit got so bad that I stopped appending "reddit" to my queries and I often find myself appending "hackernews" instead (or more commonly, I ask the LLM to find HN discussions on the topic), just because the signal is much higher.
It also makes me think about how much of HN might be astroturfing. Less than reddit for sure, but certainly not 0%.
- smcleod
"The corpus, and the refresh that changed the answer" ... hello Opus 5.
- asksomeoneelse
Niche topics like linux feel somewhat safe from bots. But the last time I checked the more popular subs.. yikes..
When I see 2-weeks old account "Incendiary-Leg-8592" responding to a post by another weeks old user "Pivotal-Kitchen-0498", I have a hard time believing they are humans behind them.