Do AI platforms share sources? How Bing powers more than you think

Partly, yes. Several AI assistants reach the web through search infrastructure they didn’t build, some of it shared, and the arrangements are only partly public and change more often than the industry’s confident diagrams suggest. The practical takeaway survives the murk: platforms overlap in what they read, so one well-placed source about your business can surface in several answers at once.

The shared plumbing, hedged and dated

AI assistants need a search index to read the live web, and building one is so expensive that several platforms borrow. That’s why index-sharing exists, why it’s mostly arranged in private deals, and why any specific claim about who powers whom should carry a date. The stable fact is the structure, not the pairings: shared indexes mean shared reading lists, and shared reading lists mean your sources travel.

What’s on the public record

Kept deliberately short, each item dated. Microsoft’s Copilot runs on Bing; that one’s structural, since both are Microsoft products. OpenAI’s search announcement described ChatGPT search drawing on search partnerships alongside its own crawling (October 2024)1, an arrangement widely reported as Bing-backed; independent analyses in 2025 disputed what its results actually reflect2, which we cite not to settle the question but as proof the pairings churn. Perplexity describes its own real-time search of the internet3, and Google’s surfaces run on Google’s index, the one pairing nobody needs to leak.

That’s thinner than the confident source-map graphics circulating in this industry, on purpose. Where the record is thin, the honest post says thin.

What we refuse to claim

We can’t tell you the current percentage of any platform’s results that come from any index, we can’t verify private partnership terms, and we won’t repeat the map everyone else copies from everyone else. When our own cross-platform tracking data is big enough to say something about overlap in cited sources, it will say it with dates and a method attached; until then, the inference boundary sits exactly here.

Why this matters for a small business

The churn is the argument for indifference to it. If you’d optimized specifically for one index arrangement in 2024, the 2025 dispute would have obsoleted your strategy; the businesses that never knew about the arrangement lost nothing. Meanwhile the sharing structure quietly works in your favor: get accurately listed on a source that multiple indexes carry, and you’ve fed several platforms with one afternoon’s work. It’s also part of why platform gaps close from the sources side rather than the platform side. Our opinion, and this post is the argument for it: index gossip is the least actionable information in AI visibility, and the sources the indexes all read are the most.