Midway through adding the examples to this article, it became quite apparent that the collection of rogues is far bigger than I thought. It already features several examples, but additions will continue to be made and added at the end of this article. Anyone who would like the full set of screenshots can contact me on hello [at] extra time dot media and request them, please include how you intend to use them in your request.
Much like earthquakes, the Google Discover Feed remains an unpredictable mystery. There are some general content policies to comply with ? although some content pieces appear in feeds despite routine flouting these ? but much of it is along the lines of ?do the right thing and then hope for the best?.
There have been some analysis articles written about it, but with no publicly accessible data, it is near impossible to reach any definitive conclusions.
What I Found After Experimenting with Google Discover for Two Months written by Mordy Oberstein and published over on Moz, is a fascinating exploration of some of the factors that influence the algorithm. Some of that is definitely visible in the odd quirks many have observed recently.
When something goes awry, as it seems to have over the last few weeks, users and content creators notice, fast. Relying on Discover as a source of traffic is folly, of course, but with the whole Featured Snippets / Zero Click searches thing, it always seemed a fair trade for complying with the requirements to be seen as a good egg by Google.
The reports of Discover going rogue with old and completely irrelevant content is at a high point, but strange behaviour is something that has happened a few times in the recent past. Whether that’s down to testing or a glitch, who knows.
As somebody obsessed with knowing how things work and why they work (or don’t work, in this case), the Google Discover Feed has been a fascination of mine for a long time.
Obviously, it’s also in the site’s best interest to monitor changes, because it can be a great source of traffic for creators like extratime.
While algorithms are, for the most part, a closely guarded secret, there are at least some reasonable hypothesis’ we can draw from some datasets.
But with Discover, there is no reliable dataset available and aside from the general content policy, Google says there is nothing creators can do to make sure they appear in these feeds.
That’s not a bad thing. It reduces the chances of ?gaming the system? as is so often seen in search results. It is also a great way to find new and interesting publications that might not appear in the ?Top Stories? in search results.
But it comes with a slew of complications, the most notable being how quickly the algorithm gets confused and how badly it gets things wrong.
The most recent reports (which started around 2 or 3 March for most users and publishers) are by far the longest run of chaotic algorithms and it is definitely the most widely reported, including that upvoted thread on Reddit.
Ever since last year, I have kept a catalogue of chaos ranging from odd to confusing to outdated content (including from this site), a very confused geolocator and the inability to accurately filter interests directly from the Feed due to the Artificial Intelligence’s inability to understand what about the story I am not interested in.
To keep the catalogue as untouched for as long as possible, I limited the number of publications to exclude from the Feed (set from the Don?t Show Stories From option).
That has had to change recently because of the inability to offer anything useful through random selection. However, much like many other users have reported, recently even choosing to exclude a source no longer works all the time.
In case it needs mentioning, everything here is anecdotal and until Google Discover data sets are publicly available, it will always be anecdotal. The high-level of personalisation the feeds offer makes it very hard to draw a definitive conclusion.
Nonetheless, we can spot some recurring bugs ? which often match dates where there were some tremors in the algorithm tools.
Nothing from the screenshots that follow has anything to do with any specific publisher. And, in some cases, the logos and headlines have been obscured because those things are irrelevant. Content creators and publishers cannot control what or when their work appears in Discover ? but they can be issued with a manual action for something that appeared in the feed. Yep, great work, really helpful.
November and December 2020: Why am I seeing this?
This oddity had an extended but intermittent lifespan. It included suggestions for an actual catalogue – which would be fine in theory, but the specials for 23 and 24 November. The card appeared on 25 November. There was also medical advice pertaining to something I erm, do not have (I am a woman).
I am not interested in the options you are giving me to not be interested in
While this might not be a bug exactly, it’s a bone of contention. I prefer to not restrict topics entirely because I want to Discover (that’s what the thing is called after all) new and interesting sources.
The examples here are extremely varied. In some cases, for the sake of research, I selected the drop-down at the right of the card to see what the options would be. No bonus points for guessing they are almost always a complete mess.
In the first screenshot in the gallery below, note the LOCAL NEWS prefix in the headline. The publication here is from the other end of the country where I live – over 1600km away.
Another one suggests alternative university options for my teenager, which I do not have. Well, unless you count the kids in The Sims 4, I guess. But I do not have kids and I have not done research about university alternatives for anyone else’s kids. Sesame Street content from YouTube is equally perplexing (no, a random child did not get hold of my YouTube account).
The example where I can only choose to not be interested in December (my birthday month of all things) is laughable. Ditto for Help Desk which is not a publication, but was quite literally the help desk of an airline. My punishment for my lack of interest in a beauty pageant is to disavow (yes pun intended) all of my interest in my whole country.
For Butter Sugar Bread (which incidentally, I don’t even know what that is, nor had I ever engaged with Apron content), my only choice is that I am not interested in butter.
And remember that example from above? My choices for what exactly I am not interested in…well, you just see for yourself. I mean, can we get a bit of nuance here?
A more recent example which can at least be deduced somewhat is this card which offers a guide on hiking trails for dog owners. At first glance, if you’ve spotted the hiking section on this site, you might think this to be a perfectly reasonable suggestion.
Well, I do not own a dog and I live in Cape Town – in South Africa – nowhere near Cape Cod. If this was an advert (seeing as the content is from a hotel’s blog), it was not marked as such. What I did search for, which might have nudged the algorithm, is information about whether you need a permit for taking dogs on trails in Newlands forest – in Cape Town (yes, you do, by the way). But how does the algorithm get the location that wrong – seeing as the information in my search included the very specific trail details?
Just for fun, not being interested in Broken Heart (does the AI think it is a band or a TV show or some sort of brand?) after a story about somebody with a broken heart showed up.
And yes, Google, *begin scene* I am not interested in payment because I can’t pay anyone because of what you did to the Discover feed recently so thanks for at least noticing that *end scene*.
What are you even saying?
Another quirk that has been around for ages and randomly affects various publications is the way headlines are rendered. While there is guidance for which images work best in Discover, headlines often generate extremely poorly.
In some cases, headlines would cut off completely or randomly select what appears to be meta text as part of the headline. This might be a technical issue specific to some publications, which would leave a lot of questions about those content policies.
The example below shows two screenshots – one which appeared to have indexed the meta of a category feed (which was not being updated dynamically on articles on this particular site). The other shows the adding on of a published timestamp.
The next example has also seemingly tacked the metadata of an article into the title and displays it as such, but the author is AFP – a wire content service, which somewhat contradicts the frequently trumpeted tagline that original content is something the Googlebots (hello bot!) love.
To stress the point: publishers cannot control whether their wire content appear in the Discover Feed. So if they were to be penalised, it would not be their fault. The above screenshot is not the only example of wire content surfacing in Discover, it’s just the most useful example considering it literally includes the wire service in the title of the card and the algorithm could not even chuck it out that way.
That’s old news now
Recently, many users have been reporting issues of outdated content surfacing in their Discover Feeds. This has happened to this writer personally and to some articles from this site. While Google Discover does show older content intentionally, these appearances are different. Old content should be marked with “In Case You Missed It”. These older articles are not.
Most of the examples that follow here are taken almost entirely from my Discover feed in March 2021. It is not the first time things have gone rogue, but it is definitely the longest time.
Over the last few weeks, my Discover Feed would choose something (and that includes some In Case You Missed It articles) that is completely outdated – despite newer and more relevant content being available.
For publishers, this one is a particular bugbear because it can destroy the CTR, possibly impacting the algorithm selecting your content – and there is absolutely nothing you can do about it. Again, there is no special markup to indicate evergreen content and marking others as time-sensitive.
The next example is particularly interesting viewed in this context. They are both from our site – with the screenshot on the right from 12 March and the one on the left from 18 March. As you can see, both are older content pieces, with the same number of days since they appeared, but only one notes it as an “in case you missed it” article.
In March (the first traceable reports from users that their Discover Feeds had gone blank appeared on 2/3 March), there have been times where my personal Discover Feed worked perfectly fine – but this rarely lasted more than a few hours at most.
My Discover Feed has also sometimes told me that there’s nothing for me to see (after updating the feed) and in one instance, only offered a single article – which was a duplication of itself. In the gallery below, two screenshots taken at different times of the day both threw up the empty feed fiasco. A single suggestion, a YouTube on how to make Hot Cross Buns was the only eventual offering that day. Sidenote: I cannot stand Hot Cross Buns.
There was also a suggestion of a motorbike (I do not own a motorbike, I do not even own a car or have any intention of owning either of these in the near future. But hey, on the upside: at least it was tagged with ICYMI.
Also in the gallery below is an article that offers some advice on how to help my child with their Afrikaans homework. Now, my first language is Afrikaans (and I use an Afrikaans dictionary alongside an English one) so that my only choice is to not be interested in Afrikaans is somewhat bemusing. I do use all Google products in UK English, as a matter of interest, though.
Another exquisite example is the “vacancies” listing (note prominence at the top of my feed). Yes, I am interested in our national parks. But I have not been searching for a job with them.
A particular favourite, which deserves to be highlighted on its own, is the suggestion of a Runway Live Stream from six days ago. We’ll ignore the fact that my fashion style is very much “is this clean” and I have worn my pants the wrong way around out in public more than once.
This might seem like an elaborate indulgence and a bit of a strop over something that, as we have already noted, nobody has any control over. That’s fair, to an extent. But the issue is affecting several users and a number of publishers.
Speaking as a user, I (used to) find the Discover Feed valuable and informative – it’s even inspired a few ideas for hikes. But for many users (myself included), the feed has been a hot mess as of late with users on Reddit, Twitter and other platforms reporting issues. Speaking as a content creator, these glitches are devastating, especially when in a growth phase (one publisher on a forum reported a drop to almost zero clicks from up to 80 000 daily).
It is foolish, of course, to rely on something so volatile, but as publishers who have been affected have noted – when they are complying with content policies and best practice and get glitched seemingly at random with no explanation, it hurts.
And since the official Google support forums are usually manned by volunteers who do not work for Google and rarely offer any helpful insight – there is nothing to be done.
When I started gathering the screenshots, I had hoped to build it up to write something more substantial, something that can offer real value and offer insights – with reliable datasets to back it up. As it turns out, the only thing reliable is that it is unreliable and there is nothing you can do about it – not even if you get hit with a manual action.
Location, location, location
On 31 March, the chaos of my Discover Feed remained….chaotic. And in another sterling attempt to out Discover itself, despite having access to my location right there on the weather widget, the best it could muster for my morning news break is a story about something in Johannesburg, a city over 1000km from where I live.
Not interested, does it mean something else?
In addition to the feeds on my own personal device that I use daily, I have another device, with a “clean” account set up. The phone has been completely reset and the only thing that can possibly connect it to me is the fact that it lives in my house and that it uses the same Play Store account. We’ll get more into detail about this in the future, but for now, the latest outstanding performance is being served content that has been marked as not interested.
I had seen this reported by other users but it was the first time I witnessed it in the wild. What a moment. Almost as mesmerising as spotting my first King Protea while hiking Devil’s Peak (in Cape Town, please, just in case the Discover Feed gets any other ideas again). Hey Google, how do you like them internal links?