AI answers appear in more places than anyone can optimize for at once, and the platforms differ in ways that change the work: who uses them, how their answers cite, and above all which search index supplies their live results. This chapter tours the ecosystem so you can pick the two or three platforms that matter for your situation and ignore the rest. The mechanics of how any of these products assemble an answer are in Retrieval and Grounding: How an AI Answer Gets Assembled; this chapter is about who runs what.
The platforms
ChatGPT
The largest chat product by a wide margin: 700 million weekly users sending 18 billion messages a week as of July 2025.The NBER working paper "How People Use ChatGPT", built on OpenAI's internal usage data. When ChatGPT decides a prompt needs current information, it runs web searches against what OpenAI describes as third-party search providers plus content supplied directly by partner publishers, and cites sources inline with links.OpenAI's "Introducing ChatGPT search" announcement. OpenAI names the partner publishers but not the search providers. OpenAI does not name those providers, which matters for optimization: you are targeting an index you cannot query directly.
Google AI Overviews and AI Mode
Google runs two generative products on top of its own index. AI Overviews is the AI summary above classic results, shown on a subset of searches; in Pew's March 2025 browsing panel, 18% of Google searches produced one.Pew Research Center's measurement of real browsing behavior, including which searches triggered summaries. AI Mode is the full conversational tab: it runs a custom version of Gemini and uses what Google calls query fan-out, breaking the question into subtopics and issuing many queries simultaneously against the Search index before composing the answer with links.Google's AI Mode announcement from I/O 2025, which describes the fan-out technique and the custom Gemini 2.5 model behind it. Because both sit directly on Google's index and ranking systems, classic Google SEO is the retrieval-side work for both.
Gemini
Google's standalone assistant, distributed across the Gemini app, Android, and Google Workspace. It grounds answers through Google Search when it needs live data. For visibility purposes, treat it as a third Google surface with a chat interface: the same index, reached through different product behavior.
Perplexity
An answer engine from the start: every answer is built from a live retrieval pass and carries numbered citations. Perplexity operates its own crawler, PerplexityBot, to build its index.Perplexity's crawler documentation, covering PerplexityBot and the user-triggered Perplexity-User agent. Its user base is small next to ChatGPT's, but skews toward research-heavy queries, and its citation-forward format makes it the easiest platform to audit: you can always see which pages built the answer.
Copilot
Microsoft's assistant, wired into Windows, Edge, Microsoft 365, and Bing. Its live retrieval runs on Bing, which makes Bing rankings, and Bing Webmaster Tools, relevant again for any audience that lives inside Microsoft products. Enterprise workforces are the practical reason to care: for many office workers, Copilot is the assistant their employer provides.
Claude
Anthropic's assistant, with usage skewed toward technical and professional work. Claude runs web searches when a prompt needs current information and cites the pages it uses. Anthropic does not document which search index supplies those results.
Meta AI, Grok, and the spreading edge
Meta AI ships inside Facebook, Instagram, WhatsApp, and Messenger, which gives it enormous incidental reach; Meta does not document its search grounding. Grok answers inside X with live access to X posts. And the assistant layer keeps spreading into browsers, phones, and voice devices, each new surface reusing one of the models and indexes above. New logos will keep appearing; the short list of underlying indexes changes far more slowly.
The indexes underneath
The model name on the product matters less than the search index underneath it, because the index decides which of your pages can enter an answer at all. Underneath the whole ecosystem sit roughly three live-retrieval supply lines: Google's index (AI Overviews, AI Mode, Gemini), Bing's index (Copilot, and OpenAI's unnamed providers are widely assumed to include it), and the independents, Perplexity's own crawl and Brave's index, which licenses search results to AI products. Rank well on Google and Bing and you have covered most of the retrieval layer in one move. The training-data side, which crawlers feed the models themselves, is a separate pipeline covered in Where the Machines Get Their Data: Crawlers, Indexes, and Training Pipelines.
Three behaviors, one ecosystem
Across all these products, three behaviors matter to a marketer, and every platform mixes them. Chat behavior answers from model memory, with no retrieval: your visibility there was decided at training time. Search behavior runs live queries and composes from the results: your visibility there is decided by rankings and extractability now. Agent behavior goes further, browsing and acting on sites on the user's behalf, and it is the newest and least settled; Future of AI Search covers where it is heading. When you evaluate any platform, ask which behavior your buyers trigger, because the same product can serve you a memory answer at 9am and a searched answer at 9:05.
Sizing the channel honestly
Referral traffic from AI platforms is real and growing, but small against the attention the platforms absorb: Pew found sources inside AI summaries clicked on 1% of visits, and Cloudflare measures AI platforms crawling thousands of pages per referral sent back.Cloudflare Radar's crawl-to-refer data; in the June 19-26, 2025 sample, ratios ranged from Anthropic's 70,900:1 to Mistral's roughly 0.1:1. Judge platforms by presence in their answers, and by what that presence does to branded demand and assisted conversions, rather than by sessions in a referral report; Attribution and Business Impact covers the measurement.
Choosing where to compete comes down to three questions. Where are your buyers already asking: consumer categories concentrate in ChatGPT and Google's products, technical audiences add Claude and Perplexity, enterprise workforces add Copilot. Which indexes do those platforms draw on, and how do you already rank there. And can you see yourself in the answers today: run your prompts on each candidate platform and look. Two or three platforms will account for most of your exposure, and the tracking system in Building a Visibility Tracking System assumes you chose them deliberately.