Barry Adams: What Publishers Must Do To Survive - IMHO

Hello and welcome to IMHO. Today I'm speaking to Barry Adams about the issue with LLM hallucination. Barry is an award-winning SEO consultant specializing in technical and editorial SEO for news publishers via Polemic Digital. Along with John Shehata, Barry founded the News and Editorials SEO Summit, and Barry also publishes SEO for Google News newsletter, which is an outstanding read. To set the tone for today's discussion, I'm going to read a recent quote— sorry, I'm going to quote a recent LinkedIn post from Barry, which is, Google's AI overviews are terrible at quoting the right sources. LLMs are incredibly dumb. They're advanced word predictors, but people don't seem to care, and Google doesn't seem to care. Publishers are keen to say that generative AI could be an existential threat to publishing on the web, yet they fail to consistently point out GenAI's biggest biggest weakness. So Barry, I'm just going to step back from that. And we're obviously going to come back to that. We're going to refer to this through our chat. But first of all, I want to say that SEOs are already creating spam in LLMs by generating listicles to cite themselves as best sources, which has been recently picked up by Lily Rae on LinkedIn. We've been fighting appalling search results for the last few years. And now LLMs are also vulnerable. Big surprise. So why do you think that LLMs are so bad at citing sources?

It's because they don't know anything. There's no intelligence. I think calling them AIs is the wrong label. They're not intelligent in any way, like, like you just quoted. They're word predictors, very smart word predictors, but still just word predictors. They're probability machines. They don't have any reasoning faculties as we'd understand it. There's no fact-checking really, even with the best efforts to ground results. LLMs will just regurgitate an answer based on the training data they've ingested. And then we'll try to post hoc rationalize the answer that they've generated with grounding efforts and citing links. And even with the best will in the world, it turns out that no matter how you prompt an LLM to only use actual sources and real-world links, it'll still have a very good chance of hallucinating some of the sources that it quotes. Because it's just inherent in the nature of the system. Like I said, these machines don't know anything. They're not intelligent in any way. They are just your predictive text on your phone on steroids, and they will just make stuff up and very confidently present it to you because that's just what they do. That's the entire nature of the technology. And I think we're making a very big mistake, both as an SEO industry and as a wider society, to assume that these things are actually smart and help us, which, you know, granted, in some cases they're very useful, but we're trying to shoehorn these systems into specific scenarios they're just not suited for. And I think that's a very, very worrying and concerning trend, really.

I've actually got this theory that we're in danger of creating an AI spiral of misinformation. With AI content where it's basically AI content referencing AI content to the point where we're completely losing the source of facts and the truth. I picked up a quote that you said recently, which I love this. It's, the real threat is how AI is replacing truth with convenience. So, you know, what point do you think that users might actually revolt against this, or just think they're going to sleepwalk into a situation where there is no truth?

I'm quite pessimistic about this. I don't think people care as much about truth as maybe we believe they should. I think people will accept information presented to them if it's useful, and if they— if it conforms with their preexisting beliefs, and if they think it helps them get the job done. People don't really care about truth. I think the last 15 years of social media has definitively proven people don't care about truth. They just care about, you know, being affirmed in their beliefs and their, their feelings rather than actually accepting the facts of the real world. And LLMs will facilitate that process even more than social media has. I feel that is probably the biggest risk we have at the moment, that up till recently, search engines specifically, and Google in particular, was very focused on surfacing factual information, about presenting people with maybe what they didn't want to hear, but which Google felt was accurate. And the move that Google now has embraced towards adopting AI, adopting LLMs, is a very clear step away from that. And all the signals, all the usability signals so far show that users don't mind. In fact, users seem to embrace it. Users seem to be very happy with being presented information that they probably know on a very clear level might not be factually correct. But it's convenient and it's handy and they don't have to do any reading, they don't have to do any thinking, and therefore they're just like, yeah, I'll just go with this and just not accept any personal responsibility for the information that they ingest. And that's the trend that we've seen since the rise of the web and the internet. In fact, even before the rise of TV, the rise of radio, and even before that, people have always been very happy to accept propaganda and disinformation if it makes them feel better, if it aligns with their beliefs. In fact, it is something that the technology platforms that now basically control our information habits should monitor and probably should have a built-in moral and ethical framework to prevent people from ingesting disinformation on a large scale. But they seem to be doing the opposite. And rather than preventing us from ingesting disinformation, they're encouraging us because it helps their bottom line, it helps their advertising revenues, it helps their overall profit motives. And we're now in a spiral which I don't think we'll ever recover from, where everybody lives their own personal truth. Nobody knows what's real and what's fake anymore. More importantly, very few people seem to care, and very few people seem to have the will to go through the extra effort required to learn what may actually be factually true. And I don't think that bodes well for a society as a whole. I don't want to be too doom and gloom here because, you know, we've always had little echo bubbles. We've also always had a little information filters to ingest what we want and disregard what we don't want. I just really fear that we end up living in an environment where entire societies will live in parallel realities and will decry the other side as being fake news and just not real. And it's going to just create even more division in our society, which is not something we need any more of.

Yeah, I mean, all I'll say is 1984. If you've read 1984 by George Orwell, and not to get too dystopian, but it is quite terrifying actually how prescient he was and just where we're heading as a society. But just to come back from going down that avenue, um, in your latest SEO for News newsletter, uh, you call out the media. So just touching on what you were just talking about, you call out the media for being passive in the face of mass hallucinations, uh, and you said quote, the media can save itself by loudly and factually reporting on AI's limitations, thus influencing public perception and user adoption of GenAI. And for the life of me, I don't understand why mainstream media aren't shouting this from the rooftops. Now, obviously, we just— you've just talked a little bit about that, but why do you think, apart from obviously monetary— that's probably going to be the main reason— why do you think mainstream media is just resisting and going with this?

Mainstream media has a long and distinguished track record of passively accepting whatever is happening in society. There are very few activist papers out there that, you know, take a moral stance and then try to fight for it. And it's also because sometimes taking a stand backfires, especially if it tends to be an unpopular opinion that loses out in the court of public opinion. In this case, you know, the risks are more involved with money, monetary gains. Google is such a powerful force in driving traffic and revenue to publishers that a lot of publishers are afraid to write too critically about Google because they feel there might be repercussions. Now, whether or not that's warranted is an entirely different conversation. I'm not sure it is entirely warranted. I do think Google is open to ranking and creating visibility for dissenting voices. But, you know, there is an inbuilt hesitancy there from the editorial workflow to bite the hand that feeds, as it were. And also, I think that a lot of journalists just don't genuinely understand what's happening, and they don't understand how these AI systems work. They just gobble up the propaganda being published by the AI industry, who's constantly in pursuit of more growth, more profits, more investments. And the technology journalists, sometimes the ones that actually know what's happening, will, you know, put question marks there. But the general journalists writing for the big nationwide newspapers, they don't know. They don't really understand how it works. So they just regurgitate the PR spin and think, oh, well, that's the news, because, you know, if OpenAI I says it's true. If Google says it's true, it must be true. Or at least it might not be entirely bold-faced lie. And I think that's a very dangerous position to take because now, especially in the last several years, we've come to realize that a lot of what these companies say are actually bold-faced lies, just not grounded in any form of reality whatsoever. When AI overviews were introduced, Google said they were going to send more traffic to publishers because people would be clicking more on the cited links. And it turns out, no, that's the exact fucking opposite. of what's happening, which everybody with 2 brain cells saw coming a mile away. But Google, you know, had that line that, oh yeah, it's going to send more traffic to publishers. And no, actually, it's going to drastically reduce traffic to publishers. And the more and more these lies we keep stacking on top of each other, again, we're in this scenario we just talked about, that people just don't know what's true anymore and don't take the effort to properly fact-check, except from a few dissenting voices here and there. And then it's very easy to you know, characterize those dissenting voices as the fringe lunatics or the Luddites. They don't want us to progress. They don't want new technology to come in. They're afraid of change. And they just, you know, discredit any real valid criticism that's being voiced about these technologies.

Yeah, I mean, the traffic to news publishers, this is obviously a very key topic at the moment and something I just wanted to dig into a little bit more. I also picked up something that you said on LinkedIn So this actually might have been from your newsletter, and quote, you said, the theory is that extensive usage of LLMs like ChatGPT and Copilot drives increased Google usage. The likely causation is users wanting to verify the generative AI output with actual trusted sources, turning to Google to find those sources, which is interesting because when I spoke to Kevin Niddick recently, this is something he actually picked up and verified with his research that users are in fact clicking to check out sources. Uh, and, but Google is also not showing AIO on top news stories, which is obviously a lifeline for news publishers. And, you know, as we know, news publishers are really struggling. So how do we explain this overall reduction in traffic across news publishers when people are clicking on the sources and AIO is not showing, uh, sorry, is not showing on top news stories? So how is this, you know, how can we explain this?

Oh, it's, it's a conflation of different sources there. People do click on sources, people do double-check the citations, but not to the same extent as when they had to do their own research. You know, ChatGPT and Gemini will give you an answer. People will click 2 or 3 links to verify. Great, mission accomplished. Whereas previously, when people had to do their own research, they wouldn't click 3 to 4 links, they'd click 30 to 40 links and read them in detail and try to find the nuggets of information there. On top of that, yes, publishers that are focused on news output are less affected because top stories boxes and AI overviews on Google tend to be mutually exclusive to a very large extent. But a lot of publishers also were dependent on traffic to things like explainer content, background stories, analysis content, and that has been replaced with AI overviews. And where people just don't click through to the cited sources or click very minimally through to the cited sources. And then there's a third party, which is that Google has been diminishing traffic to publishers for years and years and years already, keeping people on Google's ecosystem for longer, generating more searches, and, you know, penalizing publishers with algorithm updates that Google feels are considered low-quality publishers, or in a lot of cases, especially when it comes to things like the helpful content update, false positives where publishers were just doing everything right or most things right. But Google still saw them as being on the wrong end of a particular metric in the HCU update and just, you know, destroyed half the traffic or more. And then there's also websites, even big, huge news publishers like The Guardian and The New York Times, you've seen steady decreases in click-through rate over the years and years. It's very hard to point at a specific source for that behavior. It's just, you know, Google sending less traffic overall and Google being more focused on generating ad revenues. And it's just all these things combined result in a landscape where most publishers are struggling to maintain traffic, let alone grow traffic, and are very dependent on other sources like the Google Discover feed to sort of make up the shortfall. And the Discover feed, by its very nature, is less of a news feed and more of a personalized interest feed. And that also doesn't lend itself well to the type of journalism that we should care about, truthful, fact-based journalism. Discover feed is very much susceptible to clickbait headlines and high engagement signals. So we're talking about the old days of Facebook-optimized content. That sort of content now does really, really well in the Discover feed. But again, it's not really news that you're then publishing. You're chasing after clicks in a different way. And I don't think that's something that journalism as a whole should be focused too much on. So it becomes very, very hard to have a viable publishing business on the web because, you know, It's easy to say that, oh, you should never put your eggs in one basket, but if all your audience goes to that one basket, you have to be there. Google is the monopoly on informational gateway on the web. So you could say, oh, don't be dependent on Google. That's such a flippant and ignorant thing to say. You cannot have a viable publishing business without very heavily relying on Google traffic. It just isn't really possible. So You know, we have to play these games, but at the same time, we are in a scenario where we're so dependent on whatever the Silicon Valley tech bros are dreaming up in the next ketamine high that we're just not really able to escape from these constraints that they built on us because these guys have been so good at capturing the entire world in their own dragnet that, you know, you can't exist as a business outside of that dragnet in any meaningful way.

I mean, I know it's interesting you're saying there about trying to find a publishing model. I mean, some publishers like have gone down that clickbait route, like for example Daily Mail, need we say anymore, um, whereas others have always tried to experiment with the subscription, New York Times. Um, I think for me Financial Times has always been the standout, um, of a superior publication. But I know that you work with the majority of the largest news publishers globally. So what are you actually recommending to your clients about, number one, how they can optimize for LLM inclusion to get some truth into LLMs, and number two, how to survive the current situation?

I'll start with number two first, because I think that's the more interesting question. I think publishers need to accept the fact search traffic will diminish. I don't think search is going to disappear. Don't get me wrong. I think search will continue to be one of the 3 or 4 main traffic drivers and subscription drivers to publishers for the foreseeable future. I think publishers need to be more confident about what they are and specifically what they're not. You just mentioned the Financial Times. I think that's a great example because nobody has any doubt about what the Financial Times is and what kind of reporting they're signing up for, which is one of the reasons why the FT can get away with a very hard paywall. They don't necessarily need to give out free samples for people to sign up because at some stage a lot of people are like, yeah, you know what, that headline that someone shared on LinkedIn or on, on, on, on Bluesky or whatever. And they— that's a really interesting article. I want to read that, but it's paywalled. And you see several of those pop up in the course of a week or 2 weeks because the journalism is just that good. At some stage, people are going to reach that, reach that threshold and say, yeah, it's the Financial Times, big known brand, fantastic reporting on economy, on business, and even on politics. I'm just going to suck it up and pay the subscription. And I do think publishers need to have that sort of brand power. I think too many publishers try to be everything to everybody and therefore nothing to nobody. You need to have a strong brand voice. And to be fair, the Daily Mail has also succeeded in this. The Daily Mail has a very strong brand voice. A lot, if not most, of the Daily Mail search traffic would probably be people specifically wanting Daily Mail news on specific sources. This was shared at a conference I was at recently by the global head of SEO for the Daily Mail, Carly Stephen, who said that they get a lot of traffic for searches like Meghan Markle Daily Mail, Prince Harry Daily Mail, where people specifically specifically seek out that topic and that website in combination. And if you get to that level as a publisher, you know, you're not too worried about Google's ups and downs. I mean, you're still worried to a certain extent because they send a lot of traffic, but Google cannot not show you for those queries because, you know, it's a very specific query where people want a very specific website to show up in their search results. If you can then convert that to even remove the middleman and just get people to install the app or sign up to your newsletters and go directly to your website, then you're in a very happy place indeed, which is one of the reasons why a lot of publishers are trying to invest money and effort into building that contained ecosystem where they're not dependent on the technology giants as intermediaries. But that does require a publisher to have a very strong brand voice, a very strong identity, and knowing what they're good at and sort of narrowing down what they're focusing on as a publisher. And I think that's a lesson that's hard learned for many publishers that are trying to be a bit of everything to everybody, or publishers that work in a specific niche where within that niche they don't have a strong enough identity either. I work with a lot of publishers in the technology industry, for example, in the video games or comic books and nerd culture, as I call it. But those websites are fairly interchangeable. You know, how many websites can there be about the Marvel Cinematic Universe and the DC superhero films? They're all reporting the same stuff with the same screenshots and the same set photos and pretty much the same content. If you replace one with the other, the reader isn't really losing anything. And I think that's very risky as a publisher to fall into that trap. You have to have something that makes you unique. You have to have something that makes you stand out. And that makes people want to go directly to you as a source of information for whatever the topic is that you cover without going through a middleman or trying to be tempted by alternative voices. And if you don't have that, if you are just one of many reporting on a specific topic, a specific area, you're going to have a bad time. And honestly, I think for companies like Google and OpenAI and even Microsoft, They don't see that as a problem. In fact, they see it probably as a betterment of the web because it does narrow down the web and concentrate journalism in websites that deserve it. And whether or not, you know, you agree with that assessment, it's just the way things are headed. You need to have a very strong brand identity as a publisher, and if you don't have it, you probably won't exist in the next 5 to 10 years.

Yeah, well, brands seem to be all that anybody can talk about in the last 12 months. And I have— for me, I'm quite interested in where everything is going. I think, you know, I'm quite happy with the direction things are going. I think this has been a long time coming, and I like to see the development of brands. I like to see this focus on trusted authoritative sources, which to a degree we are getting. So you know, I do feel that there's a lot of positives. I know it might feel quite negative and a bit dystopian right now, but actually I think there is a lot of positive happening. So Barry, that was fantastic to hear your opinion, and thank you very much for your time. Just before we go, I just want to ask you, do you want to give a plug for your newsletter please, which is one I never miss reading your newsletter? So please tell people where they can find it and subscribe.

Thank you very much. seoforgooglenews.com. I write very irregularly though. I don't adhere newsletter publishing schedule, once every 6 to 8 weeks if you're lucky. And I hope that it's useful and insightful, and I try to cover everything that's relevant to specifically SEO for publishing.

Yeah, as I said, I never miss your newsletter, and it is always well worth reading. So Barry, thank you very much for your time and hope to see you soon.

Thank you very much for having me.