Three sites, 215,128 pages, one goal: get cited by AI
As of September 2026, a report titled "Manufactured Sources Behind AI Recommendations" reached the front page of Hacker News with 72 points and 43 comments. The finding: three websites published a combined 215,128 "best [category] software" pages, and Perplexity was citing those pages when people asked it what to buy. The pages were not written for humans to read. They were written for language models to quote.
This is the part every local business owner needs to sit with. The people running those three sites understood something most Main Street operators still don't: when a customer asks ChatGPT, Claude, or Perplexity for the "best HVAC company in Orlando" or the "best family law firm near me," the model answers from a small set of sources it has decided to trust. Get into that set and you get named. Stay out of it and you are invisible, no matter how good your Google reviews are.
Why the page farm is a dead end anyway
Here is the second story from the same week. A developer shipped Weedout, a $1.99 Safari extension that hides every YouTube video labeled "Made with AI" from your feed, search, and recommendations. It hit 168 points and 73 comments on Hacker News in two days. It uses YouTube's own label rather than a detector, so it only catches what the platform already flagged, but the direction is clear: platforms are labeling machine-made content, and users are paying to filter it out.
Manufactured source farms live in the gap between "AI engines will cite anything structured" and "everyone starts penalizing content built at industrial scale." That gap is closing. Google has run manual actions against scaled content abuse since March 2024. Perplexity and the others will follow, because a recommendation engine that keeps citing 215,128 auto-generated pages loses users. If your AI visibility plan is to out-publish the spam farms, you are buying an asset with a shrinking shelf life.
What actually gets a real business cited
The page farms are gaming a real mechanism. You can use the same mechanism honestly, and it holds up:
- Consistent business facts everywhere. Name, address, phone, hours, and service list identical across your site, Google Business Profile, Yelp, and the top three or four directories for your trade. Models cross-check. One conflicting phone number and you look less trustworthy than the farm.
- Real reviews with dates and specifics. Twenty reviews from the last 90 days that name the service and the town beat 200 undated ones. Recency is a signal the models weight.
- A handful of genuine third-party mentions. A local news writeup, a supplier's "where to buy" page, a chamber listing, a podcast appearance. Three real citations from sites a model already trusts outweigh a thousand pages you published about yourself.
- Structured data and an agent-readable version of your site. Schema markup for your services, locations, and FAQs, plus a clean text version an agent can parse without fighting your JavaScript. We covered the mechanics of that in our piece on markdown feeds and your website's second audience.
Do this by Friday
Open ChatGPT and Perplexity. Ask each one the exact question your best customer would ask: "best [your category] in [your city]." Write down three things: whether you are named, which competitors are named, and which sources the answer cites. Then click into those sources. If they are directory pages, you need your listing fixed on those specific directories. If they are review aggregators, you have a review-volume problem. If they are manufactured "best of" pages, you now know your category is being gamed and you have a short window to plant real signals before the models tighten up.
That 30-minute check tells you more about your AI visibility than any vendor pitch. It also tells you whether this is a job you can run in-house with a checklist or one that needs someone watching the citation set every week.
What this means if you're weighing AI marketing or an agent build
The mechanism behind AI recommendations is real and gameable, which means honest, well-structured signals work, but the easy exploit is closing fast. The businesses that plant real citations now will hold the position when the spam gets cleared out. Whether you run that yourself or hire it out, start by finding out what the models say about you today.
Get a free AI Visibility Report and see exactly which sources ChatGPT, Claude, and Perplexity cite for your category, and where you stand in them.