The fabulous Lily Ray has been monitoring 220+ websites that publicly identified themselves as customers of AI content creation tools. Her findings, published this month, are worth reading if you’ve been sold on the idea that scaling content with AI is a low-risk growth strategy.
- 54% of the sites she tracked lost 30% or more of their peak organic traffic.
- 39% lost more than half.
- 22% lost more than three quarters.
The pattern across all of them is the same – rapid content growth, a traffic peak a few months later, then a cliff edge drop.
You can read Lily’s full analysis here.
As Lily puts it:
“The playbooks being sold as ‘AI-first SEO’ or ‘GEO-optimized content at scale’ look remarkably similar to the playbooks that got sites flattened by the Helpful Content Update and the March 2024 Core Update. The packaging is new, but the pattern is not.”
I’ve been involved in website promotion and SEO for more than 30 years. I can confirm: the pattern is very much not new.
Lily identified eight content templates she keeps seeing on the declining sites. I want to go through them, because I’ve got something to say about every single one.
Comparison pages at scale will tank your Google rankings
If you were around in the early days of affiliate marketing, you’ll remember this one. Affiliate site owners worked out that people searching “product A vs product B” were close to buying – they just needed a nudge in the right direction. So they built pages for every conceivable head-to-head matchup in a category. Hundreds of them, sometimes thousands, all following the same template. Same structure, same verdict, different product names swapped in. It worked brilliantly, right up until Google worked out what was happening and started demoting sites that had built their entire content strategy around it.
Fast forward to now, and AI has made the same tactic ridiculously easy to execute at scale. You don’t need a team of writers churning out comparison pages anymore. You can generate the lot in an afternoon. So that’s exactly what some businesses are doing, particularly in the SaaS space where there are dozens of competitors and every combination of “tool A vs tool B” gets search volume.
The problem is that the pages are almost always near-identical in structure, thin on genuine insight, and written for search engines rather than the person actually trying to make a decision. A real comparison – one that’s properly useful – comes from someone who has actually used both products, understands the difference, and can give an honest verdict based on real experience. What these pages deliver instead is a templated rundown that could have been written about any two products in the category, because it basically was.
I stopped working with a SaaS company last year because this became their entire content strategy after listening to some hustlebro on LinkedIn. They had comparison pages for every competitor combination they could think of, all following the same format, none of them adding anything a potential customer couldn’t have found by spending five minutes on each product’s own website. I told them it was going to tank their SEO. They disagreed. We parted ways amicably, but I check in occasionally. Their traffic is now practically zero.
Lily’s data shows this pattern across multiple sites in her dataset. The comparison page template is one of the most common she found on declining domains. It works initially – Google indexes the pages, some of them pick up rankings, traffic grows. Then Google’s systems gather enough signals to understand what’s happening, and the nosedive arrives.
“What is X” glossary pages worked until everyone spammed them to death
When Google introduced the Knowledge Graph back in 2012, something interesting happened. Clean, well-structured “what is X” pages started performing really well. Google was trying to understand entities – people, places, concepts, things – and pages that clearly defined a single term in plain language were useful to it. Businesses and publishers worked this out fairly quickly, and for a while, a well-written glossary page on a topic relevant to your industry was a solid piece of content worth having.
Then, as always, people spotted the pattern and pushed it too far. Definition pages started getting churned out at scale, one per term, thin on substance, heavy on keyword matching. The same template, hundreds of times over. Google got better at identifying pages that existed purely to rank for “what is X” searches rather than to serve the needs of the reader, and the value of the tactic disappeared.
Now it’s back, dressed up in new clothes. The pitch this time is that single-term pages with clear definitions, structured answers, and schema markup are exactly what AI engines want to cite. Which is true, up to a point. But the response from a chunk of the SEO world has been to do what it always does – because remember, SEOs fuck it up for themselves – take something that works in moderation and produce it at industrial scale, often programmatically, sometimes across multiple languages from a single template with the terminology swapped out.
Don’t get me wrong, I’ve had clients get decent results from properly written glossary content – pages that define a term clearly, give real context, and connect it to something useful for the reader. That’s different to running off 300 definition pages with AI and calling it a content strategy. Google has seen this movie before, and so have I. Lily’s data shows these pages appearing repeatedly across the declining sites in her dataset, particularly the programmatic multilingual variants where the only thing changing between pages is the language the thin content is served in.
If you’re going to have glossary content on your site, make it useful. One well-written page that helps someone understand a concept relevant to your industry is worth having. Three hundred templated definition pages are a liability waiting to be noticed.
“Best X for Y” listicles at scale destroy the sites that rely on them
Before Google’s Panda and Penguin updates landed in 2011 and 2012, affiliate marketing in particular had a very reliable playbook. Build a site, populate it with “best X for Y” roundups, stuff them with affiliate links, and watch the commissions roll in. The pages followed a formula so consistent you could have recreated it blindfolded – intro paragraph explaining why this topic matters, a list of products with brief descriptions, a “winner” conveniently placed at the top, a call to action. Rinse and repeat across every product category with search volume.
It worked because Google’s systems at the time weren’t sophisticated enough to distinguish between a really useful recommendation from someone with real expertise and a templated roundup written purely to capture affiliate revenue. Panda changed that. Sites that had built their entire website strategy around this template saw their traffic collapse almost overnight. Some never recovered.
The template never went away though. It just moved. And now, with AI making it possible to produce this content faster and cheaper than ever, it’s back in force – this time on SaaS blogs, B2B service sites, and business publications that really should know better. The pitch is slightly different. Instead of affiliate commissions, the goal is search visibility and AI citations. But the pages look almost identical to what was getting sites penalised over a decade ago.
A different SaaS client came to me after their blog, which was almost entirely this type of content, had seen its traffic reduced to practically nothing. We audited everything, deleted and redirected over 600 posts, and consolidated a load more into proper case studies that reflected their expertise and client results. Their traffic recovered, and their enquiries recovered too. It took time and it wasn’t cheap, but the alternative was continuing to haemorrhage visibility with no floor in sight.
What pisses me right off is that companies are still falling for this. The people selling AI content tools are very good at showing the initial traffic bump these pages can produce. They’re considerably less fucking forthcoming about what happens six to twelve months later, which is exactly what Lily’s data documents. The “best X for Y” template appears repeatedly across the declining sites in her dataset. It works, until it doesn’t.
Self-promotional listicles are a short-term tactic with a predictable ending
There was a period, roughly around 2008 to 2012, when “top 10 X providers” pages were everywhere. Companies worked out that if you published a roundup of the best suppliers in your category and put yourself at the top, you could rank for searches that your actual service pages couldn’t touch. The pages looked editorial. They had the structure of genuine recommendations. They were nothing of the sort – they were thinly veiled sales pages dressed up as helpful guides, and the company publishing them had an obvious financial interest in the outcome.
Google got wise to it. The signals weren’t hard to spot – a company publishing a “top 10” list in their own category, naming themselves number one, with suspiciously thin write-ups of the competitors they’d graciously included to make the page look balanced. Sites that had built significant traffic through this approach took heavy hits as Google’s ability to identify self-serving content improved.
Fast forward to now, and Lily found this pattern on sites at a scale that would have been impossible to achieve manually. Hundreds of these pages, each one a “best of” comparison in which the publisher conveniently tops the rankings, produced with AI and published in bulk. Many of the sites running this strategy at scale got hammered in an unconfirmed Google update around January 2026, with some losing between 40% and 95% of their organic traffic in the space of a few months.
I’ve had enquiries from potential clients doing exactly this and I’ve turned them down. Not because I can’t see why it’s tempting – these pages can generate real traffic initially, and the people selling the strategy are good at pointing to the early wins. But they’re short-term gain tactics with predictable endings, and I’m not interested in helping someone build on a foundation I can already see cracking. Google has caught up on this before. It’s caught up again. It will keep catching up, because the underlying problem – content that exists to serve the publisher rather than the reader – hasn’t changed. Only the tools used to produce it at scale have.
Competitor-vs-alternatives pages: templated, spammy, and painful to read
This is a close cousin of the comparison page tactic, with one key difference. Instead of pitting two products against each other in a neutral-ish “vs” format, these pages target a competitor’s brand name directly. “[Competitor] alternatives.” “[Competitor] vs everything else.” One page per named competitor, each one structured to capture people searching for the competitor’s product and redirect them towards yours instead.
The intent is obvious, and that’s part of the problem. A potential customer searching for “[Competitor] alternatives” is usually someone who’s already considered that product and wants to know what else exists. What they get with these pages is a barely disguised sales pitch from a company with an obvious interest in the outcome, dressed up with just enough structure to look like a genuine comparison.
When the pages are produced at scale with AI, they tend to be almost identical to each other apart from the competitor name. The same template, the same structure, the same verdict, the same call to action. Google’s systems are quite good at identifying when a site has published fifty near-identical pages that exist purely to intercept competitor brand searches, and Lily’s data shows these pages appearing consistently across the declining domains she tracked.
The SaaS company I mentioned in the comparison pages section was doing this at the same time, alongside everything else. The content was templated, spammy, and genuinely painful to read. When I suggested a different approach they weren’t interested, and I’ve already explained how that ended. Their traffic tells the rest of the story.
Programmatic location pages – nothing new, just faster to get wrong
This one truly is as old as SEO itself. The logic is simple enough – if you want to rank for “recruitment agency in [town]”, build a page targeting that town. If you want to rank in fifty towns, build fifty pages. If you want to rank in five hundred towns, build five hundred pages. And if the only thing changing between those pages is the town name, well, that’s a problem for future you.
About 15 years ago I worked with a recruitment company that had over 1500 location pages, all programmatically produced, all identical except for the place name. For a long time it worked. They were ranking in towns they didn’t even have offices in, picking up traffic from searches they’d never have reached with a handful of manually written pages. Then the Helpful Content Update arrived and took a significant chunk of that visibility with it, because Google had finally got sophisticated enough to identify pages that existed purely to rank in a location rather than to really serve someone in that location.
The pages hadn’t changed – Google’s ability to evaluate them had.
Location pages can work. A well-written page that speaks specifically to customers in a particular area, references genuine local knowledge, and connects to real work done in that region is a legitimate piece of content. A template with “[town name]” dropped in wherever the location needs to appear is not, and Google has been getting steadily better at telling the difference for years.
AI hasn’t introduced anything new here. It’s just made it possible to produce 1500 location pages in an afternoon rather than a week, which means the scale of the problem arriving on a site’s doorstep is larger, and the traffic drop when Google catches up is correspondingly steeper. Lily found this pattern across multiple sites, including multilingual variants where the same thin template was being served in every language a search engine would index. Different language, same problem.
FAQ farms miss the point of what FAQ content is actually for
The pitch for this one is seductive if you’ve been paying attention to how AI search works. AI Overviews and AI engines pull clean, structured answers from pages that make extraction easy – a clear question, the answer in the first paragraph, schema markup at the bottom. So the logical conclusion, apparently, is to build hundreds of pages each answering exactly one question, all optimised for extraction, and wait for the citations to roll in.
There are a few problems with this.
The first is that even if it works for AI citations, it generates almost no meaningful traffic. Someone asking a simple question in an AI Overview gets their answer in the Overview. They don’t click through to your page. The citation might technically exist, but the visit doesn’t. You’ve built a page that serves an AI engine’s answer rather than a human being’s need, which is more or less the definition of content that exists for search engines rather than people.
The second problem is that when someone does land on one of these pages, there’s nothing to keep them there. One question, one answer, a schema block, done. Bounce rate through the roof. No reason to explore further, no connection to your actual services, no demonstration of the expertise that would make someone want to get in touch. Thin content with nothing underneath it.
The third problem is the one nobody selling this advice seems to want to address – it fundamentally misunderstands what good FAQ content actually does. When someone asks a question, they don’t just have that one question. They have a cluster of related questions forming at the same time, and a page that genuinely serves them addresses those too. It gives context, anticipates the follow-up, and connects the answer to something useful. One question, one page, schema markup isn’t a content strategy. It’s a checklist exercise that produces pages Google is increasingly good at identifying as thin, and that real humans find almost completely useless.
I haven’t seen this in client work directly, but the “one blog post per FAQ” advice is everywhere at the moment. Lily’s data shows FAQ farms appearing consistently across the declining sites she tracked. The pages exist. They just don’t do what the people who built them hoped they would.
Off-topic content at scale wrecks your topical authority
There have always been two versions of this tactic, and both of them have been getting sites into trouble for as long as I can remember.
The first is publishing irrelevant high-volume content to hoover up traffic that has nothing to do with your actual business. In the early days of SEO, you’d find plumbing companies with pages about celebrity gossip, accountancy firms with content about football results, and B2B service businesses publishing anything that got search volume regardless of whether it had the faintest connection to what they actually did. The logic was pure numbers – more pages, more traffic, more visibility. The fact that the traffic was entirely useless to the business was treated as someone else’s problem.
Google got better at understanding what a website was really about, and sites that had polluted their own topical authority with irrelevant content started paying for it. An accountancy firm that had spent years building genuine expertise signals around tax, finance, and business advice found that expertise diluted by hundreds of pages about things Google now understood had nothing to do with accountancy. The entity signals that help Google categorise and trust a site were muddied, and rankings for the content that relly mattered suffered as a result.
The second version is paid hosting of third-party content – most commonly gambling and casino pages, though it’s appeared in other forms over the years. A legitimate site gets paid to host pages that have nothing to do with their business, often buried in subfolders or on subdomains, on the basis that the site’s authority will give those pages a rankings boost. The site gets paid. The casino operator gets visibility. Google gets manipulated. Until it doesn’t.
I recently had an enquiry from a business publication that had hundreds of casino and gambling pages buried amongst their actual content, which they were being paid to host. When I told them those pages needed to go, I was told “everyone does it.” That was the end of the conversation. I can’t work with people who won’t listen.
Everyone does it isn’t a reason. It’s a description of how a tactic gets abused to the point where Google specifically targets it, which is exactly what has happened here repeatedly over the years. It messes with topical authority, dilutes your site’s entity signals, and sends Google confusing signals about what your website is about. Lily found this pattern across multiple sites in her dataset, including B2B service businesses with entertainment content that had no conceivable connection to their actual work. The sites doing it had no idea how much damage they were causing to the content that was supposed to be driving their business.
This cycle has played out before – AI has just made it faster
If you’ve read through all eight of those tactics and thought “this sounds familiar”, that’s because it should. The Britney Spears Naked era of SEO – stuffing irrelevant meta tags with celebrity names to hoover up unrelated traffic – felt cutting edge at the time, albeit completely fucking pointless. So did article spinners, producing thousands of near-identical pages of garbled content at a speed no human writer could match. So did programmatic location pages, and self-promotional listicles, and comparison page farms. Each one worked briefly, got abused at scale, and ended the same way.
AI hasn’t introduced a new cycle. It’s accelerated the existing one, because it makes the scaling part trivially easy. Tactics that might have taken a team of writers months to execute at damaging scale can now be deployed in an afternoon. The gap between “this works” and “this has been abused to the point where Google targets it” is getting shorter.
What Lily’s research adds to this is something that’s been missing from most of the conversation around AI content – actual data. 220+ sites tracked over months, consistent patterns across industries, real numbers showing how far traffic falls and how quickly after the peak. The people selling AI content tools are very good at showing you the early wins. Lily’s dataset shows you what comes next, and it makes for sobering reading.
If you’re currently scaling content with AI and any of the eight templates in this post sound familiar, it’s worth having an honest look at your site before Google does it for you. And if someone has sold you one of these approaches as the future of SEO, you now have the data to ask them some uncomfortable questions.
If any of this sounds a bit TOO familiar and you’re worried you’ve fallen for one of these tactics, I’m happy to give you some no-nonsense advice. Grab me on LinkedIn and I’ll take a look.
