Media

    Hate to say I told you so

    I don't think I was the first person by a long way to draw the conclusion that AI posed an existential threat to the publishing industry. While far too many publishers took one look at large language models and saw a way to make the cost of content significantly cheaper, my background in audience development told me straight away that the ability of LLMs to ingest content and spit out a summary was going to fundamentally change search engines, to the long-term detriment of publishers. Why click on a link when Google can give you an answer?

    Ben Thompson – still one of the sharpest commentators on the relationship between tech and media – this the nail on the head when he talks about LLMs as an extension of aggregation theory:

    LLMs are breaking down all written text ever into massive models that don’t even bother with pages: they simply give you the answer.

    Google's announcement at Google I/O that it would roll out AI Overviews on search pages in the US seems to have finally got people's attention. Google is, of course, claiming this will increase clicks to publishers, which seems barely credible -- and the fact it's not giving publishers any way to see if a click originates from AI Overviews in Google Search Console suggests it really doesn't want anyone measuring that claim.

    And don't think as a publisher you can opt out. The only way to do that is to opt out of search entirely. While you can opt out of being crawled for use in the Gemini chatbot, you can't opt out of AI Overviews on its own.

    Gartner estimates the fall in traffic down to having AI-driven answers on the results page at between 35-60%, which would be a catastrophic fall for most publishers.

    I think this is only the start: AI answers are not conversational, and for many situations – especially purchases – a conversational interface is generally superior to a simple text search. Turning search into a conversation is a huge step and it's likely to further impact publishers as a new generation gets used to "talking" to a machine to get exactly the answer they want.

    So what should you do? The only way forward is to build direct audience, and focus on owning the relationship between you and the reader without being mediated by the large platforms. That may mean subscription - but it could also mean focusing on a smaller but highly valuable audience, wether that's a niche in the B2B market or focusing on building community rather than giving answers.

    Treat your existing ad revenue and affiliate revenue as a cash cow, not a growth market. Although my gut feeling is that AI Overviews won't impact on affiliate content as heavily as the kind of answers-based pages which have been big traffic drivers for the past few years, their time will come too.

    Be entertaining – remember that audiences don't just read content for information, they want to be entertained, provoked into thought, and to come away from an article or site feeling like it's really delivered. That probably means raising the quality of your writing – think New Yorker or The Verge.

    But most of all: be yourselves. Don't try and be anyone else. For as long as I've been in publishing one of its scourges has been the reverse of "not invented here": a kind of belief that competitors are somehow ahead of you, that they have their strategy straight while you're still stuck in some kind of technological or content strategy dark ages. It's never been true: everywhere I've worked has felt the same way about whoever their competitors were.

    You need to know your audience better than Google does, better than Facebook does. Can you do that? I think you probably can.

    The future of search-driven affiliate content

    This week, rather than focus on a single topic, I thought I would look at a couple of different things – and also point you in the direction of some other great articles I’ve read this week. First, though, something that’s a bit of a theme for me: the long-term unsustainability of the affiliate/SEO content model...

    This screen grab neatly illustrates why I think search-driven affiliate content has no long-term future:

    If you're a publisher, and you have an affiliate-led revenue strategy, you might want to start thinking about what your revenue model will be in a few years. Conversational AIs are going to massively impact on your ability to monetise.

    This is an image from a query performed using Google Gemini Advanced, and the answer to my question includes links directly to retailers for users to just buy the recommended products. No affiliate intermediary required. And remember: Gemini will be the default voice assistant on Android, eventually. I haven’t seen good data on how many affiliate purchases take place on mobile versus desktop, but it’s likely to be the majority.

    Conversations generative AI tools like Gemini are not good enough yet for purchasing (and that's true of most generative AI) but it will be, and probably sooner than you think. In nearly thirty years of covering technology, I don't think I have seen a new tech that's improved as rapidly as generative AI. It's gone from laughable to serious in 18 months, and I suspect that pace will continue given the money and resources pouring into it.

    For purchasing decisions, conversational interfaces will ultimately be better for users, as they allow them to hone down recommendations based on personal priorities. I could talk to Gemini and say, “OK, I have a mix of hard wood floor and carpets, what would you recommend?” I can change price ranges, asking if there's a really outstanding model just above what I'm looking to spend.

    It's going to take, I think, between three and five years to get to the point where the technology is good enough, but it will get there. If you have affiliate revenue now, I would strongly recommend you invest what you're making in a long-term strategy of building more direct traffic, and more direct revenue from consumers (in whatever way is appropriate to your market).


    You might have noticed that online media is currently in trouble, with layoffs happening virtually everywhere you look. Dave Ruddock has written an excellent summary of why – and in particular, why now. It’s a quintuple whammy of Google, consolidation, affiliate, AI and the death, for publishers, of the distribution platforms they have come to rely on.

    Dave finishes with “get out of media while you can, folks. It’s a bloodbath” and I sympathise with that view a lot. Paul Newman, who is MD of tech at Future, points out in the comments that this may just be the end of a cycle, and he’s right to say that the media business goes in cycles. When I’m optimistic, I think the next cycle will be all about direct traffic, paid for via subscription, and a resurgence of quality over pageviews. But the pessimist in me notes that getting there is going to be very painful, for businesses and for individuals.

    The other major concern I have is the loss of experience in the industry. If you’re in your 50s now and someone dangles a decent carrot of redundancy money in front of you, you’re probably going to take it and go off to do something other than active journalism.

    But it’s deeper than that. I had a conversation recently with a freelance tech writer of many years standing, who told me he no longer bothers even pitching to write reviews. Publications now want a “best of” with 10 products in it, and want to pay £200 for it – and actually properly testing 10 products takes a lot longer than than half day that £200 should pay for.

    That freelancer represents decades of experience in technology and testing products. That’s all now effectively lost to the industry, never to return. Crafts which bleed knowledge like that rarely see an increase in quality over time, and if we are moving to a new era of quality content, we will need – somehow – to get that back.


    On a more “admin” note, I’m thinking about expanding what gets emailed out to subscribers, and would like to hear your views.

    At the moment, the mid-week newsletter is the only thing anyone gets via email. Other posts are all web-only.

    I have started a regular link post on Fridays called “Ten Blue Links”, which covers a broad range of everything that’s interested me. I am thinking of sending this out by email too. I also do occasional longer posts on things which I have found interesting, again mostly in tech. And on Sundays I write a weeknote, which is a more personal reflection of what I’m up to.

    As an experiment I’m going to send out Ten Blue Links via email, as well as the mid-week post. I’m hoping this will be of value to you. Ultimately, I want to move to an email system which allows you to sign up to one, or both – but for now it will have to be all or nothing.

    Let me know what you think either in the comments below or by email. I’m planning to start emailing out Ten Blue Links from next week.

    HouseFresh and the challenges of affiliate content

    You might have noticed a post from HouseFresh doing the rounds, especially if you have anything to do with creating content intended to generate affiliate revenue. It’s caused quite a stir, particularly among publishers

    My background is in product testing. My first job in publishing was in MacUser's testing labs, where we would regularly have 10–20 products in and – literally in some cases – take them apart to decide which one was best. Next door was the PC Pro labs, which did the same thing, on an even bigger scale. On a visit to New York a few years later, I went to the testing labs of a US publisher: even bigger, with people who looked like they should be wearing lab coats picking over the bones of machines. The product testers were real experts, often devising unique tests designed to stretch the products in ways which matched up the real-world pounding they would take. 

    But those were proper group tests. What HouseFresh is writing about is not those. Their focus is the “best” article, written specifically to deliver affiliate clicks and sales, and designed to hit a specific keyword. 

    The HouseFresh article rips the lid off some of the worst aspects of content written to deliver sales though affiliate links (I refuse point-blank to call it “comtent”, which has to be one of the worst words ever invented). Their biggest complaint is that a lot of the pages you will see which rank highly on Google for affiliate-led keywords are written by people who have never had the products in their hands, let alone tested them. They may have done desk research, which involves, at best, scouring spec sheets for hidden details and, at worst, just scouring the user reviews on Amazon. But that doesn't tell you all that much about a product and whether it's any good or not.

    Of course, this is really Google's fault because it is rewarding low-quality content by ranking it highly. This content, which is far cheaper to produce than a real group test, can be churned out quickly. A quick writer can do one or two a day, while a group test might take two weeks to organise, test and write. Use an LLM and you can probably make that process even faster. Just make sure to write your prompt to make it include phrases like our lab tests and our experts said to satisfy Google's pretty surface-deep view of how content based on real-world experience works.

    HouseFresh’s hope is that Google will improve its algorithms and start rewarding content which is of higher quality, but I have my doubts. I suspect that the company’s focus is on creating “answers engines”like Gemini, rather than the traditional ten blue links. And even if it can improve its algorithms to prioritise in-depth reviews, gaming the SEO system will often look like a better option to many of the kind of publishers HouseFresh is attacking: the ones who have bought well-known brands but now use them to churn out lower quality content. 

    There are, and will be, exceptions, mostly from publishers who have a heritage in creating brands, rather than the ones that buy brands just for their heritage. But the sheer volume of content created by others could drown them out—especially as LLMs make it easier to generate entire sites within days.

    As I have pointed out, I believe businesses based on this kind of affiliate-led content will also be disrupted over the coming few years by conversational AI. Once people have the option of having a conversation with a smart recommendation engine to tailor buying advice to exactly their needs, “best XXX” articles based on desk research or mining Amazon reviews just won’t be good enough. 

    Google is a choke point for the affiliate content business, but it’s not the only one. The second is Amazon, where most publishers derive a large chunk of their affiliate revenue. Although reliable numbers are difficult to find, Datanyze estimates Amazon has around 48% of the market share in affiliate networks, and anecdotally, I suspect the amount of revenue that brings in for publishers is higher still. Every publisher I know has sought to reduce their exposure to Amazon, especially after the effective demise of its Onsite Associates programme (known internally as OSP), changes in policy from Amazon would have a massive effect on publishers. But the reality is that if Amazon turned off the taps, or even reduced them, publishers with big investments in affiliate content production would be in trouble. 

    Would Amazon do this? It depends if you believe that Cory Doctorow’s enshittification cycle applies to it: 

    First, they are good to their users; then they abuse their users to make things better for their business customers; finally, they abuse those business customers to claw back all the value for themselves.

    Are we at the point where Amazon starts to claw back the revenue it shares with its “business customers” – affiliate partners? Currently, probably not. But it’s worth thinking about the longer term, too. Already, 61% of US shoppers begin their buying journey on Amazon regularly. That’s traffic which Amazon has to pay no extra commission on, and so it’s something that it would love to do more of. 

    Plus, of course, Amazon has hundreds of millions of reviews of its own that it could tap to automatically create recommendations for users, including by layering conversational AI on top of it to allow users to get “intelligent” recommendations. The potential is there for Amazon to be a trusted source of reviews, as well as the retailer of choice for online.

    The “good” news is that presently, Amazon is struggling with its own grey goo of content in the form of fake reviews generated by AI. It’s responded with more AI to try to trim them out. But the key question is really what happens at the point it decides that the money it spends on delivering affiliate revenue would be better off spent on ads, or on-site AI, or whatever else.

    Ironically, the kind of content mills which HouseFresh is railing against would be less bothered if Amazon does ever scale back on its focus on affiliates: they are, most likely, pretty aware that the brands they are using are near the end of their life, and if the affiliate cash cow moves on the private equity companies will have long since made a sizeable return. 

    The relationship between Amazon and publishers, like that of Google and publishers, is some kind of symbiosis. Amazon gains revenue from the clicks that publishers drive their way. Publishers get a slice of that money, enough for them to survive and grow. But the key question is whether that symbiosis is obligate – where each depends on the other for survival – or facultative, where each benefits but could survive alone. If it’s the former, affiliate content has a long and profitable future. If it’s the latter, then eventually, publishers who go all-in on it may have a problem. 

    Rebuilding

    Back in 2002 (or thereabouts), I started writing up my thoughts on technology at a site that I dubbed "Technovia". About fifteen years later, after about a year's break from blogging, I went back and found that at some point the database holding all that content had become corrupt. Thousands of posts, disappeared, and despite some pretty exhaustive investigation I've yet to find a way to get it back. I could probably do some snazzy SQL export and import, some kind of magic incantation, but at this point, when the site has been down for over a year, it's just not worth it.

    Which brings me here: Technovia's successor, which I'll be constructing over the next few weeks and months.

    I actually feel like starting with a clean slate is actually appropriate. It's not that I'm ashamed of my old posts, or feel the weight of history on me (there's no way I could be that pretentious -- my one significant contribution to "knowledge" can sit in Archive.org and Wikipedia for the rest of time -- but that I feel there's a new age of technology now, and it deserves somewhere new to write about it.

    Hence, here.

    The aim is to write something everyday. I'm following the "500ish words" model that MG Siegler pioneered, but hopefully with a little more regularity. It's going to be rough -- think of this as first drafts -- but I think that's perfectly fine.

    And I might well stray from the world of technology regularly too. There days, I'm not an active participant in the world of technology journalism. I do more management and thinking about media and business models than I do about bits and bytes. So don't expect too much commentary on the latest tech events only.

    I think it will be fun, for me. I hope it will be for you too.