AI Doesn't See the Whole Internet. It Sees What It's Been Let Into.

Most people I know assume that when they ask an AI tool for the best plumber, the best brunch spot, or the best running shoe, it is reading the entire internet and giving them the honest average of everything it found. I assumed that too, for longer than I'd like to admit.

It isn't true, and the reason it isn't true matters if you run a business or a nonprofit that depends on being found.

Seeing Is a Permission, Not a Skill

The first thing to say is that this is not a story about what AI can't do. A model can read a review as easily as you can. What decides whether it reads yours is whether the platform hosting that review has opened the door.

Amazon is the clearest example. In late 2025, it updated its robots.txt file to block more of OpenAI's crawlers, including the ones that fetch live information to answer a ChatGPT user's question. Analysts read it as part of a larger effort to keep outside AI agents away from a business built on shopping traffic and advertising revenue. Amazon then took the same posture into court. In March 2026, it won an injunction against Perplexity's Comet shopping agent. In August, the Ninth Circuit vacated that injunction and sent the case back, reasoning that the agent runs on the user's own device and acts more like a browsing tool than a separate visitor. The court was careful to say it was not creating a new legal regime for agentic AI, and it left Amazon free to regulate access through its own terms of service.

So the legal question is still open. What isn't open is the practical one: Amazon decides who gets in, and it has been deciding.

The Other Side of the Door

Yelp made the opposite choice. In July 2026, it announced a licensing agreement with OpenAI so that ChatGPT can pull in Yelp reviews, photos, and business data when someone asks a local question, with Yelp's branding and a link back attached. The financial terms weren't disclosed, and the deal isn't exclusive. Yelp already supplies similar data to Apple Maps and Yahoo.

Put those two decisions next to each other. A person asks an AI tool where to find a good contractor, and the answer is shaped by a contract they will never see. Reviews from a platform with a deal can show up with confidence. Reviews from a platform that has closed its doors are, from the AI's side, a locked room.

Where "Pay to Play" Gets Complicated

I've been calling this pay to play, and I want to be careful with the phrase, because it's only half right. Businesses can't pay Yelp to make ChatGPT recommend them. What gets paid for, or negotiated, is the platform's access. The visibility belongs to the platform that holds the deal, and the business inherits whatever that platform's relationships happen to be.

That is a quieter kind of gatekeeping than advertising, and in some ways a harder one to notice. Nothing is labeled as sponsored. Nothing tells the consumer that a whole category of evidence was never in the room.

What This Means for Small Businesses and Nonprofits

If you've spent years collecting reviews on one platform, it is worth knowing whether the AI tools your audience uses can see that platform. If you haven't, you may be invisible in places you never thought to look.

A few things follow from that, none of them dramatic. Your own website still counts, and so does the language on it. Your presence across several platforms is more durable than strength on one. And your reputation, the real one, built out of how people describe you in their own words, is the thing that travels with you no matter which doors open or close.

For nonprofits, this carries one more layer. Many of the communities you serve do not leave reviews at all. Their voices were never going to appear in a licensed dataset. If AI is becoming the front door to how people find services, then the people least likely to be seen by it are, as usual, the ones already easiest to miss.

The Gap I Keep Noticing

This is, once again, a translation problem. Consumers assume AI sees everything, and businesses assume that doing good work will be enough to be found. Neither assumption holds. The distance between what a person thinks the tool knows and what it can actually reach is where good organizations quietly lose ground.

I don't think the answer is panic or a scramble to game a new system. I think it starts with knowing which doors are open, and keeping your story in enough places that no single contract decides whether you exist.