The question changed shape

"Best commercial roofer near me" used to produce a results page. Now it can be a question asked directly of ChatGPT, Perplexity, Gemini or Claude, and the answer that comes back is usually a short list — one or two names, with a reason attached. The buyer never sees the ten alternatives that didn't make the list. They see the ones that did.

This isn't a hypothetical future state. OpenAI, Anthropic and Perplexity all operate dedicated infrastructure for exactly this behavior today. OpenAI documents three separate crawlers — GPTBot for model training, OAI-SearchBot for building its search index, and ChatGPT-User for live fetches when a person asks ChatGPT to look at a specific page (OpenAI, "Overview of OpenAI Crawlers"). Anthropic runs the equivalent set for Claude: ClaudeBot, Claude-SearchBot and Claude-User (Anthropic, Claude Help Center). Perplexity does the same with PerplexityBot and Perplexity-User (Perplexity, "Perplexity Crawlers"). Each company built and maintains separate systems for indexing the web and fetching pages live on a user's behalf — infrastructure a company doesn't build unless people are actually using it that way, at a scale worth engineering for.

Why buyers prefer it

Comparing five vendors used to mean five tabs, five phone calls, and an afternoon. Asking an assistant to do the comparison collapses that into one conversation. The assistant reads what it can find about each business, weighs it against what the buyer asked for, and returns a short answer instead of a long list. For a buyer with limited time, that trade is an easy one to make, and once someone experiences it working, they don't go back to the old way for that kind of decision.

It also changes what "showing up" means. A business doesn't need to rank first on a page the buyer never visits. It needs to be the one the assistant actually recommends, which depends on whether the assistant could find the business, understand what it offers, and trust what it read — not on the buyer scrolling past it.

What this means for a business that hasn't adjusted

A site built purely for a human visitor to browse can be functionally invisible to an assistant doing this work on a buyer's behalf. Not because the business did anything wrong by the old rules — those rules rewarded a different kind of visibility, built around ranking on a page a person would scroll. The new reader behaves differently: it wants facts it can verify, structure it can parse without guessing, and a way to act once it has decided to recommend someone.

None of this replaces the old audience. People still browse. But a growing share of the buyers who used to start with a search are starting with a question to an assistant instead, and a business that is invisible to that assistant is invisible to that growing share — whether or not it shows up fine in a traditional search result.

The practical takeaway

This shift rewards businesses that treat "can an AI agent find, read, trust and act on my site" as its own question, separate from traditional SEO. It's a different reader with different requirements, and it's already the reader deciding some share of who gets called back today, not in some future cycle.