AI Isn’t Going to Save Us

One of the thoughts I see regularly is that AI is going to save us from the deleterious effects of the attention economy. AI is somehow going to magically help us deal with the increased cognitive load that technology requires, it’s going to solve the problems with the attention economy, and it’s going to be the end of enshittification. That’s a lovely thought, but I don’t believe it for a minute.

If you’ve followed me on podcasts or Bluesky, you know that I’m extremely skeptical about the likelihood of replacing human artists and writers with generative AI. Sure, it’s bad that it makes people who use it stupider and yeah AI reverts to the mean by design, and no, I probably wouldn’t build any commercial product on something that’s unprotected by copyright unless you do substantial modifications. But I do think AI is quite good at a few things—translating work from one format to another, generating certain types of code, backstopping humans in difficult pattern recognition roles, and raising the for accessibility in video and audio.

But the thing I think generative AI is the absolute best at churning out content that sounds plausible. It can do that at a rate that’s limited only by your budget. That means the easiest way to make money with AI right now is spam.

Air Purifiers Were a Bellwether

This famously came to a head a few years ago with air filters. There used to be a handful of sites that did in-depth testing of them, but because each filter is a relatively high-dollar item, reviewing them was a profitable enterprise. People who are looking at air filter reviews are typically in the market, so a sizeable number of reviews led to clicks into online stores, which in turn generates sweet affiliate paychecks. I actually love using sites that are supported by affiliate revenue, because I know if they give me bad purchasing recommendations, I’m less likely to trust their reviews (and click their affiliate links) again in the future. It’s a powerful incentive to write good reviews.

The sites that specialized in reviewing household air filters had to spend a ton of money and time buying a bunch of them, establishing a solid testing methodology, spending untold hours testing them, then collecting the results into a meaningful, easy-to-read reviews. Doing this kind of meat-and-potatoes review is incredibly labor intensive, but it was worth the effort. Separating the great from the merely good products would make or break a site.

Pre-AI, these high-investment, high-return sites weren’t particularly good targets for spammers. Sure, people copied reviews from the originator’s sites, but Google is constantly scanning the web, and it was pretty good about prioritizing results from the original creators instead of some yahoos who just lifted another site’s reviews. Sometimes being the first poster has an advantage.

But generative AI upended that model. These days, if you want a site with a bunch of air filter reviews on it, all you have to do is ask ChatGPT to write a reviews of each of the most popular air filters that are on the market. Then ask it to make a best of list, and link them all together. Are the reviews accurate? Who knows? Are they fair and representative of the products? I don’t know. Do they provide actionable buying advice? Sure. Do the people who use this tech to generate a bunch of spam care if people spend their hard-earned money based on terrible machine-generated advice? Not a whit, at least as long as they keep whatever they buy past the return window.

When you use something like ChatGPT to generate a bunch of bogus reviews, it indiscriminately aggregates reviews from all over the web—editorial reviews, user reviews, comments on social media, YouTube comments, and random blogs. It’s untraceable and makes generative AI the steam engine that empowers a new Industrial Revolution of stealing other people’s content.

Once you’ve built out a toolbox of relatively simple scripts and prompts to generate a site, you can get ChatGPT to generate a list of high value search terms. Then you use that list to plant AI-generated seeds for each of those terms in the form of real looking sites, complete with outbound links to other sites, images, and all the things you’d expect to see. You can create a positive feedback loop of content generation that only stops when you run out of money. But when you pair the AI-generated sites with low-effort affiliate and display ads, you won’t ever run out of money! Instead, you’ll create a flywheel of spam—an infinite ouroboros of spammy sites that each generate a little bit of passive income. This is an actual business model, presumably from the same people who brought us the plague of ad-driven content that looks kind of like it belongs on this website, like Demand Media and Taboola.

And all you had to do to make it happen was make the web just a little bit more useless.

On a positive note, there is actually a happy ending to the air purifier story. The search results today for that particular term are quite useful. It’s because a small site made a big stink about Google ranking AI-generated slop that was being hosted on sites with general search ranking ahead of smaller specialist sites a few years ago. It went viral, and apparently Google made some tweaks, at least to the way the search algorithm handles air purifiers.

But the overall impact of garbage AI-generated content is everywhere and it’s going to continue to outpace the tools designed to combat it. Since I started working on the newsletter, I’ve been looking for decent stock image services, and the low end of that market is completely overrun. The top results in every category on all of the big stock image sites are low-effort AI slop, to the point that you can’t find interesting human-generated Creative Commons or public domain work outside of geriatric sites like Flickr. While the high-dollar services, like Getty, are uninfested so far, anyplace you can buy rights to a specific image for a few bucks has been overrun.

How Do We Combat Webspam?

I like to end these with something actionable that you can do in a few minutes to make your computing experience better. But unfortunately, I can’t do much to help you with webspam. I’ve started testing some alternative web searches, but I’m not thrilled with any of the results so far. (I’ve spent time with both Duck Duck Go and Kagi so far, so if you have other recommendations you’d like me to test out, hit reply).

My basic rule for generative AI is the same as it is for vampires—I don’t invite either of them into my house. That means while I’m continuing to use Google, I disable as much of its AI-generated business as I can using the UDM=14 trick. (Ernie Smith, over at the always awesome Tedium blog, clocked this filter last year, and found that it not only disabled AI, but also all the other crap they’ve been adding to search results over the last 15 years.) What’s the trick? If you append your Google searches with ?udm=14 then you get 2004-style Google results—no knowledge graph, no additional context from the search engine, no embedded news results, no image results, nothing. Just straight web results that take you to the website you were probably looking for in the first place. It’s really nice.

Appending some cryptic combination of letters and numbers manually every time you search is a pain, so there are a variety of tools that will help you. The fastest and easiest thing to do, at least for Chrome and Firefox users is to go to tenbluelinks.org and follow the instructions on the page to create a custom search for your browser and then set it as the default for location bar searches in your browser’s settings pane. It should take just a few minutes and it’s been a significant upgrade to the way I search. (If you’re worried about passing all your searches through a third party, don’t be. I tested it on Chrome and Firefox on PC, and the people running the site have just used the OpenSearch spec to create a custom search that goes directly to Google. They’re not going to know anything about whatever dark business you’re searching for.)

If you’re on Safari, it’s a little more complicated. You can head to this page and use this userscript and the Userscripts extension for Safari to replace all Google search links with the UDM=14 versions.

So while you and I can’t fix webspam, we can prevent you from getting hit with crappy AI-generated summaries of AI-generated pages every time you search for something.

What’s Next?

A few weeks ago, I wrote about turning off almost all of my notifications, and the results have been pretty impressive so far. A good day of work conversations in Discord or Slack can still blow out my daily average, but as a whole my time spent on the phone is more directed by me and less directed by apps interrupting me. The first few days after I trimmed the notifications way down were weird. I felt lonely, in the same way I do when my group chats and small Discord groups are quieter than normal. But after a few days with that sweet, sweet brand interaction and my endless thirst for dopamine was successfully being slaked with a handful of good mobile games and an ebook or two instead of ads for cheap chain pizza. (Shout outs to my friend Sandy Weisz, who just launched Raddle this week. It’s a kick ass daily word ladder puzzle game that you should check out.)

This week I also kicked off premium subscriptions for the newsletter. As is always the case, the main newsletter is and will remain free for everyone. But I’d love to spend more time on it, expand it to other platforms (like podcasts and maybe even YouTube), and have time to track down things like interviews, and all that takes cash. If you want to chuck a few bucks my way, you’d get my eternal gratitude, an occasional special issue, and some cool bonus content. On a related note, I got my first ad offer on the newsletter, and you’ll all be pleased to know that I turned down the opportunity to annoy a few thousand people in exchange for $12.

I have a couple of different potential topics in the hopper for next week, but I haven’t honed in on one yet. As always, feel free to hit the reply button and let me know what you’d like to see. I got a bunch of great suggestions last week, and really appreciate the general feedback on what I’m doing over here. Until then, please consider subscribing, sharing the newsletter with a friend, or saying something nice about it on social!