Three related websites — wifitalents.com, worldmetrics.org, and gitnux.org — have published a combined 215,128 machine-generated "best software" pages since late 2023, and research published this week by Trellner shows Perplexity's AI search cites all three hundreds of times when people ask it which software to buy. I spent an afternoon opening these sites myself, and what I found on their "about" pages is stranger, and more useful to know, than the raw page counts.
Short answer: Trellner's research found that wifitalents.com, worldmetrics.org, and gitnux.org — three sites registered within months of each other — published 215,128 "best software" pages and that Perplexity cited them 181 times across 380 categories tested. I verified the sites directly: all three list the identical German company registration number on pages titled "Facts & Grounding Page," built to be read by AI models, not people.
Disclosure: AISagely may earn a commission if you buy through links on this page, at no extra cost to you. It never changes the verdict — I recommend what I would actually use, and I say so plainly when a tool has a real weakness.
I use Perplexity, ChatGPT, and Claude most days for research, so a report claiming one of them is quietly citing spam got my attention. Rather than just repeat Trellner's numbers, in my testing I opened the three flagged domains myself, read their disclosure pages line by line, and checked what each AI vendor actually promises about sourcing. Here's what's real, what I could confirm on my own, and how to protect yourself.
What Trellner's research found
According to Trellner’s report, wifitalents.com, worldmetrics.org, and gitnux.org were all registered through Namecheap between December 2023 and May 2024, and a fourth domain, zipdo.co, shares the same infrastructure. Together, the three sitemaps list 316,202 URLs, and the 215,128 "best software" pages for AI to cite follow the pattern /best/<something>-software/ — a template page ranking tools in a narrow category like "project estimation software" or "invoice software for freelancers."
| Site | Total URLs | "Best software" pages | Perplexity citations (of 380 categories tested) |
|---|---|---|---|
| wifitalents.com | 103,578 | 70,731 | 71 citations across 27 categories |
| worldmetrics.org | 107,083 | 71,684 | 60 citations across 22 categories |
| gitnux.org | 105,541 | 72,713 | 50 citations across 23 categories |
To measure citations, the researchers queried Perplexity's sonar and sonar-pro models across 380 software categories and logged every source it returned. The three sites combined for 181 citations — 2.4% of the 7,534 total sources Perplexity cited. That sounds small until you see the bigger pattern: 59.8% of all 7,534 citations pointed to domains ranked worse than #100,000 on Tranco (a common web-traffic ranking), and 23.4% pointed to domains outside the top million entirely. The median cited domain ranked 71,611st on the web. For comparison, guideflow.com — a software vendor's own marketing blog — was the third most-cited domain overall, with 194 citations across 96 categories, ahead of research firms like Gartner.
I want to flag one honest caveat here: a few commenters on the original discussion noted that Trellner's own report reads a lot like the templated "research report" content it's criticizing, and that the same submitter has posted similar reports before. I can't verify Trellner's internal methodology from the outside. What I could verify — because I went and looked — is the underlying claim about the three sites themselves, and that part checks out.
I checked the three sites myself
Each of the three domains carries a page at /facts/ with the HTML title "[Site name] — Facts & Grounding Page." That phrase — "grounding" — is the technical term for the retrieval step an AI model performs before it answers a question, and it's addressed to a crawler, not a reader. Gitnux's version is explicit about it: the page states it follows the "Grounding Page Standard v1.6" and was "Published July 21, 2026." Worldmetrics' identical framework says "Published August 25, 2026." Neither reads like a page a human would ever land on and want to keep reading.
Here's the detail I didn't expect: all three sites' facts pages list the exact same German company registration — "HRB 32988, Amtsgericht Augsburg," registered at "Tränk 1, 86551 Aichach, Germany." Three sites that market themselves as independent research outlets share one corporate registration number, down to the digit. Each page also lists dozens of named "research analysts" with individual bios, and assigns published statistics a confidence label — "Verified," "Directional," or "Single source" — a system that looks rigorous to a model scoring source reliability, whatever it actually reflects.
None of that is illegal. It's optimization aimed at a different reader than the one the page pretends to address, and it works well enough that a widely used AI search tool is citing it as a source for real purchase decisions.
The AI tools that are actually transparent about where an answer comes from
Since the whole problem here is an AI tool citing a source without you knowing it's thin, the tools worth paying for are the ones that show you the source and make it easy to check. I compared the four I use regularly on exactly that basis, not on general capability.
| Tool | Entry paid plan | Free tier | Are citations visible by default? |
|---|---|---|---|
| Perplexity | Pro, $20/month | Yes — real search, capped Deep Research | Yes, numbered citations under every answer |
| ChatGPT | Plus, $20/month | Yes — capped usage | Only when web search is explicitly used |
| Claude | Pro, $20/month ($17/month billed annually) | Yes — capped daily messages | Yes, when web search is on; links are inline |
| Gemini | Google AI Pro, $19.99/month | Yes — genuinely usable, $0 | Yes, via an expandable "sources" panel |
Best overall: Perplexity. It's also, notably, the tool this exact report is about — which is the point. Perplexity shows you every source it used, which means you can catch a gitnux or worldmetrics citation if you actually look, something you can't do at all with a tool that doesn't show sources. In my testing over the past several months, Perplexity's citations are reliably clickable and dated, and Pro runs $20/month; the gap this report exposes isn't that Perplexity hides its sources, it's that showing a source doesn't mean the source is good. I've covered the day-to-day mechanics in how to use Perplexity AI and gone deeper on research-specific workflows in how to use Perplexity for research. Try it here: Try Perplexity →.
Best for value: Claude. Claude's web search citations sit inline in the answer and link straight to the source, and at $20/month (or $17/month paid annually) it matches Perplexity's price while also handling writing, coding, and long documents well — a broader tool for the same money. I compare it head-to-head with the more chat-focused ChatGPT in Claude vs. ChatGPT. Try it here: Try Claude →.
Best free option: Gemini. Google AI's free tier costs nothing and its "sources" panel is genuinely there, not a paywalled feature — you can expand it on any grounded answer and see the domains behind a claim before you trust it. It won't out-research Perplexity Pro, but for casual "what should I buy" questions the free tier is enough, and $0 beats a report like this one finding your $20 subscription cited a spam farm anyway. See how it stacks up against the other default assistant in Gemini vs. ChatGPT. Try it here: Try Gemini →.
ChatGPT is the one honest exception on this list: unless you explicitly ask it to search, Plus and Free both answer from training data with no citation at all, which means you can't audit a ChatGPT answer the way you can audit the other three. If you already pay for it, it's still a capable generalist — just don't treat an un-searched ChatGPT answer as sourced.
How to sanity-check any AI software recommendation
Whichever tool you use, treat a citation the way you'd treat a stranger's product review: check who's behind it before you act on it. A few things that took me under a minute per site once I knew to look: open the "about" or "facts" page and see if it names a real person with a real, checkable history, not a bio-shaped block of text. Search the company registration number, if one is listed — a legitimate small research firm has exactly one; three "independent" sites sharing one is the whole story here. Check whether the numbers on the page cite a primary source (a vendor's own pricing page, a named survey) or just another aggregator page like itself. And if a domain feels unfamiliar, a quick search for [domain name] review or [domain name] reddit before you trust a product ranking on it takes less time than the AI's answer did to generate. For SEO and content teams specifically, our guide to the best AI tool for SEO optimization and to setting up AI Search Console tracking covers the other side of this — how to see whether your own site is being cited accurately, and by what.
My verdict
If you take one thing from this — 215,128 "best software" pages for AI don't become trustworthy just because a chatbot cited one of them. None of this means stop using AI search — it means stop treating a citation as proof. Perplexity remains my first stop for anything I need sourced fast, precisely because it's the one tool here that lets you check its work; Claude is the better buy if you also want a strong generalist at the same $20; Gemini's free tier is the right call if you're not ready to pay for any of them. What I won't do anymore is skim past the citation list. If you see wifitalents.com, worldmetrics.org, gitnux.org, or zipdo.co cited in an AI answer about software, treat that specific claim as unverified and go find the vendor's own page.
How I verified this
I read Trellner's full report and the discussion around it, then independently opened wifitalents.com, worldmetrics.org, and gitnux.org's homepages and their /facts/ pages to confirm the page titles, company registration details, and "Grounding Page Standard" language myself rather than take the report's word for it. Current pricing for Perplexity, Claude, and Gemini comes from each vendor's own pricing page, checked the week of writing; ChatGPT's $20/month Plus price is unchanged since 2023 and is corroborated across the same pricing pages cited in our Perplexity vs. ChatGPT comparison.
Frequently Asked Questions
What are wifitalents.com, worldmetrics.org, and gitnux.org?
They're three websites that have together published 215,128 machine-generated "best software" ranking pages since late 2023. Their disclosure pages list the same German company registration number, and research from Trellner found Perplexity cites all three when answering software-buying questions.
Is Perplexity safe to use for software recommendations?
Yes, with a caveat: Perplexity shows you its sources, which is what let researchers catch this pattern in the first place. Use it, but click through the citations on any recommendation that matters instead of trusting the summary alone.
Are free AI tools good enough for researching software?
For casual questions, yes. Gemini's free tier and Perplexity's free plan both include real search with visible sources, which covers most "what should I use for X" questions. Paid tiers mainly add depth and higher usage caps, not fundamentally different sourcing.
How much should I expect to pay for an AI search tool?
Perplexity Pro, ChatGPT Plus, and Claude Pro all sit at $20/month (Claude drops to $17/month billed annually). Google's AI Pro plan is $19.99/month, with a cheaper AI Plus tier at $4.99/month. All four have usable free tiers, so start there before paying for any of them.
How can I tell if an AI's cited source is trustworthy?
Open the citation and check for a named, checkable author or company, a single consistent business registration (not one shared across "competing" sites), and whether it links back to a primary source like a vendor's own pricing page rather than another aggregator.