The two ways an assistant reaches a vendor's site
Both ChatGPT and Perplexity operate two distinct kinds of access to the web, and the difference matters for a vendor being evaluated. OpenAI runs OAI-SearchBot to build the index behind ChatGPT's search feature ahead of time, and a separate agent, ChatGPT-User, that fetches a specific page live the moment a user asks ChatGPT to look at it (OpenAI, "Overview of OpenAI Crawlers"). Perplexity's structure mirrors it: PerplexityBot indexes ahead of time, and Perplexity-User fetches a page live "when users ask Perplexity a question" that requires reading it directly (Perplexity, "Perplexity Crawlers"). A vendor can be indexed well in advance and still fail the live fetch when a specific buyer asks a specific question — or the reverse.
What the buyer actually asks
Vendor-comparison prompts tend to be specific and stated in plain business language: who handles a particular kind of work in a particular place, who has a particular certification, whether a given company can do a given job at a given scale. The assistant reads what it can access about each candidate, weighs it against the buyer's stated need, and answers with a short list rather than every company that could plausibly compete. A vendor that never enters the conversation isn't rejected — it's simply never retrieved as a candidate in the first place.
Why the first visit is invisible to the vendor
None of this shows up in a vendor's normal analytics. A live fetch by ChatGPT-User or Perplexity-User doesn't look like a customer browsing a site; if the vendor never gets recommended, there's no click to measure and no session to attribute. The evaluation happens, and the vendor either never learns about it, or only finds out weeks later when a deal they didn't know existed goes to a competitor instead.
What separates the vendor that gets recommended
The businesses that show up consistently in this kind of comparison tend to share the same traits: their crawlers are allowed in, their services and credentials are stated as facts rather than implied by design, and there's a clear, verifiable way to take the next step. None of that is exotic, but it has to be true at the moment a specific buyer asks a specific question, not just true in general. A vendor that looks credible to a person browsing casually can still fail the narrower test an assistant runs when a real buyer is deciding who to shortlist.
The shift this represents for B2B specifically
B2B buying cycles have always involved a research phase before a human conversation starts. What's changed is who — or what — conducts that research phase now. A vendor that was previously found through a salesperson's outreach or a referral can now be found, or missed, entirely by an assistant a buyer trusted to do the legwork. Being ready for that conversation is a prerequisite for being in it at all.