A report crossed my desk yesterday that confirmed something I'd long suspected but never had hard numbers for:
Just 3 domains produced 215,128 pages of long-tail content like "Best CRM Software 2026." Perplexity and other AI search engines are citing these pages directly in their answers.
Source: trellner.com report
The story hit 122 points and 72 comments on Hacker News. Here's the breakdown.
How a "Farm" Pumps Out 215,000 Pages
The three domains named in the report all run the same playbook:
- Templated titles: "Best CRM Software 2026," "Best Project Management Tools 2026," the same sentence structure with a different keyword slotted in.
- Near-zero information density: pages share roughly 80% identical content. Only the product name, a few screenshots, and a boilerplate "pros and cons" paragraph change.
- Full long-tail SEO coverage: every "best XXX software" search intent in the industry gets captured.
How do you get to 215,000? Because this isn't writing, it's running a script.
The pseudocode looks roughly like this:
products = load_top_1000_saas_list()
categories = ["CRM", "PM", "AI Writer", "Analytics", ...]
for product in products:
for category in categories:
if product.fits(category):
slug = f"best-{product.slug}-for-{category}-2026"
render(template, product=product, category=category, slug=slug)
The template ships with pre-written paragraph skeletons, SEO meta tags, and FAQ schema. Once rendered, it pushes straight to Cloudflare Pages. A thousand pages a day is trivial.
How's the quality? A glance at the meta keywords tells the whole story: AI-generated content, auto-published, no human review.
Perplexity Didn't Get Hacked. It Trusts by Default.
The key insight: this isn't a Perplexity bug, it's the design.
The AI search workflow is straightforward:
- Crawl (like Bingbot, but with its own frequency and page-selection strategy)
- Index (embed page content into a vector store)
- Generate: pull top-k relevant snippets into the prompt
- Attach citation links to the answer
That top-k selection in step 3 is still built on traditional SEO signals: keyword match, domain authority, content freshness.
SEO farms win on those signals by default:
- Volume: 215,000 pages covering virtually every long-tail query
- Keyword precision: titles literally say "best XXX software 2026," a perfect match for "recommend a CRM"
- Structured data: FAQ schema, list schema, all the markup crawlers love
When Perplexity surfaces these pages as sources, it's effectively stamping spam with an authority seal. Users see "Sources: bestcrmsoftware-2026.com" and assume the link passed some kind of editorial filter.
GEO: The New SEO Battlefield Built for AI
This gives rise to a new acronym: GEO (Generative Engine Optimization).
How it differs from traditional SEO:
| Dimension | Traditional SEO | GEO |
|---|---|---|
| Goal | Rank on the SERP | Get cited by AI answers |
| Target | Google crawler + ranking algorithm | AI retrieval pool + top-k selector |
| Content strategy | High quality + high backlinks | Mass volume + keyword match + structured data |
| Success metric | Ranking position, CTR | Cited by Perplexity/ChatGPT |
GEO isn't about optimizing for users. It's about polluting the AI retrieval pool.
Traditional SEO still required semi-difficult-to-fake signals: backlinks, content quality, dwell time. GEO collapses to a much simpler formula: whoever spins up pages fastest wins the citation slot.
Who's Responsible in a Broken Trust Chain?
The frustrating part: every player in the chain denies responsibility.
- Content farms: "I wrote software recommendations. I never said they weren't AI-generated."
- Perplexity: "I retrieve and summarize. Source quality isn't my problem."
- Advertisers: "I bought sponsored placements on these pages. It's standard affiliate business."
- Users: "I just wanted a quick answer."
The net result: sponsored posts get a second life as "objective AI recommendations."
It echoes the medical advice pollution Google faced around 2010, when "best hospital for treating X" results were dominated by SEO-driven clinics. Google eventually cracked down with Penguin, Panda, and Medic updates.
But AI search is harder to clean up:
- In traditional search, users still see the title, domain, and thumbnail, enough to judge.
- In AI answers, links collapse into a tiny "Sources" footnote. Almost nobody clicks them.
In other words, AI search has upgraded SEO farm content into authoritative answers, and downgraded it into a footnote.
Practical Defenses for Users and Developers
Some tactics that work right now.
1. Check Source Domain Age
The links AI answers cite are almost always fresh domains carrying bulk-generated content.
A quick whois check is enough:
whois bestcrmsoftware-2026.com | grep -i "creation date"
If the domain was registered after 2024 and the headline points to "2026," it's almost certainly a farm.
2. Reverse-Engineer the Source List
Perplexity puts its sources in a collapsible panel at the bottom of each answer. Expand it and click through every link.
Heuristics:
- Domain age under 2 years → high risk
- Same domain appearing N times → farm
- Content is a templated listicle → farm
- Real author byline + contact info → positive signal
3. For High-Stakes Questions, Skip AI Recommendations Entirely
Medical, financial, legal, and tooling-selection decisions shouldn't rely on AI answers. Go straight to:
- G2, Capterra (genuine user reviews)
- Vertical industry press (Wired, The Verge, etc.)
- Long Reddit / HN threads with real debate
AI search is fine for "explain a concept," "quick comparison," or "where do I start." It's not for final decisions.
Editor's Take
The scary part isn't "215,000 AI-generated pages." It's that Perplexity and its peers are quietly rewriting what "authoritative" means.
In traditional search, an SEO farm can at most push an ad to the top of the SERP, where users can still see what it is. In AI search, the same content gets woven into a coherent narrative, and the source links are easy to miss. Users don't just lose the ability to find information; they lose the entry point for judging whether that information is real.
My predictions:
- GEO services will explode over the next 12–18 months. Any tool that can mass-produce pages + nail keyword matches will get packaged and resold by SEO agencies.
- Perplexity will be forced to build source-quality tiers, something like Google's PageRank evolution. But the road is long, and they lack the equivalent of PageRank's "public voting" signal from external links.
- The only real defense is user education: teach people that AI-cited links still need second-level verification.
Unlike Google's medical pollution era, AI answers are continuous prose. The pollution source gets hidden by natural language, the way it can't on a SERP. What's being lost isn't the ability to find information; it's the ability to scrutinize it.
I'm not sure Perplexity itself will want to acknowledge this problem. Admitting it means admitting the entire retrieval pool's credibility is discounted. Under commercial pressure, they'll more likely look the other way, until regulators or competitors force the issue.
References:
- trellner.com original report
- Hacker News discussion (122 points / 72 comments)