.png)
The fight over AI search is turning into a fight over who owns the sources that AI answers cite. This month a federal judge weighed whether Reddit can sue to protect the conversations that feed engines like Perplexity and ChatGPT, OpenAI pushed its agents deeper into the work people used to do by searching, and the largest study yet of AI search put numbers on which brands get surfaced and which get left out.
On July 23 a Manhattan federal judge spent nearly three hours questioning whether Reddit has the authority to sue on behalf of its users, in a case that accuses Perplexity and three data-scraping firms of harvesting Reddit content from Google results to feed an answer engine. Reddit filed the suit in October 2025; it has licensing deals with OpenAI and Google and wants Perplexity to pay the same way.
The stakes reach past two companies. Reddit is the most-cited source in Perplexity and Google AI Overviews and the second-most-cited in ChatGPT, according to analytics firm Profound. The engines that increasingly answer questions about brands are built on a narrow set of sources, and the terms of access to those sources are now being drawn in court.
For a brand, the lesson is where visibility actually comes from. If a handful of platforms supply the sentences an assistant writes about you, being present and accurately represented on them is not a nice-to-have. The legal fight is a reminder of how concentrated, and how contested, the supply of AI answers has become.
OpenAI launched ChatGPT Work on July 9, an agent built on GPT-5.6 that pulls context from a directory of more than 1,400 connected apps and files to produce finished documents, spreadsheets, presentations, and even hosted websites. It rolled out first to Pro, Enterprise, and Education users.
The move matters for discovery, not just productivity. Every task an agent completes by reading connected sources and returning a finished answer is a task a person no longer performs by searching. Anthropic's Claude Cowork and Microsoft's Copilot Cowork are chasing the same ground, which points the whole category toward answers assembled for the user rather than links handed to them.
Semrush's 2026 AI Visibility Index analyzed 126 million U.S. AI search prompts from January through April across ChatGPT, Gemini, Google AI Mode, and Google AI Overviews. Its central finding for brands: appearing in an answer is not the same as being cited as the source behind it. On Gemini, the overlap between mentioned brands and cited domains ran as low as 30%.
Citation behavior differs sharply by engine. ChatGPT drew on about 15 sources per response and leaned on Reddit and Wikipedia; Gemini cited about 3, from a narrower pool. Teams that integrated SEO and AI-visibility work reported gains far more often (81%) than teams running the two separately (36%).
A practical look at which structured-data types actually help a page get read and cited by AI engines, and which ones do little once the basics are in place.
Suppression worked differently when the output was a list of links. This post covers what changes when the surface is a generated answer, and how to shift the sources a model reaches for.
Tucked into the ChatGPT Work launch: a hosted-websites feature that lets people build and publish a site straight from the chatbot. Handy, and also one more way the open web fills with machine-made pages. The more of the web is generated, the more it matters which pages a model decides to trust.
.png){kind=logo}
