The New Gatekeepers: AI Chats with Secret Search Habits
ChatGPT search behavior is best described as a hidden, AI-driven layer sitting on top of traditional search engines, where models quietly run background queries, select ranking pages, and synthesize personalised answers that now compete directly with classic search results in shaping what users see first online. This matters because it shifts power from visible search rankings to invisible AI decisions about which sources are worth citing and recommending. When people ask ChatGPT or Gemini a question, the models fire off real-time web searches, pull in the pages that surface, and build a response from those citations. Unknown blogs and vendor sites can appear beside or instead of the expert review platforms that used to dominate research, so discovery is no longer tied only to human-entered keywords but to what the AI chooses to search and trust.
A recent study of ChatGPT’s software recommendations shows how stark this power shift has become. For each of 40 software categories, the researchers sent one buyer-intent query to ChatGPT with web search enabled and then classified all 188 cited sources across 105 domains. The result: vendors grading themselves accounted for 51% of citations, small anonymous sites made up another 23%, while the analyst firms, review platforms, and business press that have guided software buying for two decades contributed only 16% combined. In other words, the AI is quietly rewriting who counts as an authority, and most buyers have no clue this new hierarchy even exists.
Authority Inversion: When Unknown Blogs Beat the Old Experts
The most unsettling shift in AI content discovery is what DerivateX calls the “Authority Inversion”: the trusted middle of research is hollowed out and replaced by self-interested vendors and obscure blogs. When ChatGPT recommends business software, it leans far more on vendors’ own websites and on blogs almost no one has heard of than on established analyst firms or industry press. G2 and Capterra, two of the largest software review platforms, received zero citations across all 40 categories, while Gartner appeared only twice, and only via its user-review pages rather than analyst research. In several categories, sites with no public profile outranked household names, including a product-management tool’s blog that was cited more often than major news outlets. The quote that should make every marketer pause: “The institutions that vetted software for buyers have been replaced by vendors and sites no one has heard of.”
This is not a minor reshuffle; it is a quiet collapse of the old trust ladder. For twenty years, SaaS vendors were taught that credibility flowed from analysts, big review platforms, and the business press, earned slowly through coverage and reports. “AI quietly threw that hierarchy out,” said DerivateX co-founder Apoorv Sharma, and the data backs him up. For buyers, the practical impact is sharp: AI recommendations should be treated as a starting point to verify, not an independent verdict, because more than half of the sources behind those answers are vendors describing their own products and nearly a quarter are small, unvetted websites. The comforting idea that an AI assistant is aggregating expert consensus is, for now, mostly a myth.
ChatGPT’s Hidden Queries: AI as Universal Intent Decoder
Under the hood, the ChatGPT recommendations algorithm looks less like magic and more like a clever wrapper around standard search. When a user asks a chat model a complex question, the system breaks that conversation into subsets of solvable queries and runs them as traditional Google or Bing searches in the background. Those ranking pages become the raw material for the answer through Retrieval Augmented Generation (RAG), while the AI layer adds synthesis, personalisation, and conversational flow. Many users are unaware these traditional searches are happening, yet the optimization target has already moved: brands now need to rank for what the AI quietly searches for on a user’s behalf, not just for the visible query typed into a box. In essence, AI search is turning into a universal intent decoder that translates messy conversation into clean machine queries.
One incident with Reddit exposes how dependent AI outputs are on this hidden search infrastructure. Reddit had been enjoying meteoric visibility in Google and, according to citation tracking data from PromptWatch, its share of citations in ChatGPT responses ran as high as 15% before collapsing to below 2% within days. The trigger was not Reddit’s content or an AI training update, but a quiet change: Google removed the num=100 parameter from its search API, killing the ability to request 100 search results simultaneously. That alteration caused Reddit’s AI visibility to fall off a cliff, showing that an independent company’s AI visibility tracked the behavior of a search API it did not control and probably did not know existed. If your marketing strategy still assumes you are only optimizing for direct user searches, you are playing yesterday’s game.
SEO for AI Models: Ranking for What the Bot Searches, Not the Human
The rise of AI content discovery means brands must rethink SEO from the ground up. You are no longer optimizing purely for what a human types into a chat box; you are optimizing for what the AI agent quietly searches for on their behalf, in the background, without the user knowing it happened. Those hidden grounding queries, often very different from the original prompt, decide whether your content is cited or ignored. Tools like QueryFan exist precisely to expose this shadow layer: they generate persona-specific questions, run them through ChatGPT and Gemini, and capture the exact searches each model triggered, producing the list of queries you need to rank for to appear in AI-generated answers. Those are your new SEO targets, and until recently there was no free tool that surfaced them at scale.
QueryFan’s approach reveals how SEO for AI models will differ from classic keyword lists. Traditional keyword research is built around one-shot queries, but prompts to LLMs are multifaceted, conversational, and shaped by prior context such as a user’s stated preferences or identity. AI conversations branch, and tools now use question-proximity data to predict likely follow-ups, which in turn drive more background searches. This means modern gap analysis has two outputs: content to create on your own site and placements to earn on other people’s sites that the AI frequently cites. Most companies are still spending against a map that no longer decides anything, focusing on visible keywords while ignoring the intent trees and grounding requests that actually feed AI recommendations. The brands that adapt will own the AI answers; the rest will vanish from the new surface.
Why B2B SaaS Marketers Must Treat AI as a New Channel
For B2B SaaS, the implications are immediate and commercial. When buyers ask ChatGPT for software recommendations, more than half of the cited sources are vendors grading themselves, while trusted analysts and review platforms barely register. That means the AI is now a primary discovery channel where unknown blogs and niche consultancies compete head-on with long-standing review brands. Ignoring this channel is not caution; it is self-sabotage. Agencies such as DerivateX already specialize in helping B2B SaaS brands get found and cited inside ChatGPT, Perplexity, Gemini, and Claude, offering LLM SEO services and guidance on hiring AI-focused optimization partners. These services exist because the game has changed: you need to be visible to the hidden search layer if you want to influence what buyers see at the top of their AI chats.
The strategic takeaway is blunt. Treat AI chat systems as independent distribution channels with their own ranking logic, not as sidekicks to traditional search. Most companies are still spending against a map that no longer decides anything, pouring effort into familiar SEO dashboards while their competitors quietly secure the citations that power AI recommendations. The brands that will win this shift are those that ask a hard new question: “What does the model search for when my buyer talks to it?” Then they build content, partnerships, and tooling around that answer. For buyers, the advice is equally clear: treat AI outputs as draft research, verify sources, and be aware that the loudest voice in your chat window may be a vendor, not an independent expert.
