Ask an AI product where its web search actually comes from, and most of the time the honest answer is: not from them. The chatbot, the agent, the research tool — whatever sits on top — is usually a thin, well-designed layer over a search index someone else built. That's not a criticism; building and running a web-scale index is one of the harder infrastructure problems in software, and renting access to one is a completely reasonable way to ship a product. But it's also a fact worth knowing, because it quietly determines two things about whatever you're using: what it costs, and what happens to it if the rented index goes away.
What "owning an index" actually means
An index, in this sense, is a company's own crawled, stored, continuously refreshed copy of the web — or some slice of it — that it can search directly, on its own infrastructure, without asking anyone else's API for the answer. Very few companies have one. Google has one. Amazon built one, launching Bedrock Web Search on it in August 2026. Brave has run an independent one for years. Most other AI products you'd name — the ones with a "search the web" button — don't; they call an API in front of someone else's index, or in front of a service that itself relays a search engine's results page.
That distinction sounds architectural, but it shows up immediately in the product. A company that owns its index decides what to crawl, how often to refresh it, and how to organize what it holds. A company renting one gets whatever the index owner decided to expose through the API — no more, no less — and inherits every change the owner makes upstream, with no say in the decision.
Why it sets the price
Owning an index is expensive to build and comparatively cheap to query once it exists — the crawling, storage and continuous reading are a fixed, ongoing cost that doesn't move with each individual search. Renting one flips that: no infrastructure to build, but a per-query meter that runs for as long as the product exists, set by someone else's pricing page.
That's the actual reason search API rates cluster where they do. A company that must itself pay a margin to whoever owns the index it's calling has no room to price far below that cost — and a category of much cheaper services exists precisely because they don't hold an index at all: they relay a search engine's results page and charge accordingly, a genuinely different, thinner product priced for what it is. A rate under a dollar per thousand and a rate of five, eight, or twenty-five dollars per thousand aren't really disagreeing about the value of the same thing — they're describing different things, one with an index behind it and one without.
What happens when a rented index disappears
This isn't a hypothetical risk. In August 2025, Microsoft retired Bing's Search API — the budget option a large share of this market had quietly been built on. Every product calling it had to migrate, on Microsoft's timeline rather than their own, to a replacement that was a different shape of product at a considerably higher price. Nothing about those products' own code changed; the ground underneath them did, on a decision none of them made and none of them could see coming with any real lead time.
That's the ceiling a rented index puts on a product, whether or not it ever gets tested: the price, the terms, and the continued existence of the thing you depend on are all decisions that belong to someone else. Most of the time that ceiling is invisible, because the API keeps answering and the invoice keeps arriving on schedule. It only becomes visible the day the owner changes its mind — and by definition, that's the day it's too late to have planned around it.
How to tell which one you're using
There's a quick way to check, and it doesn't require reading anyone's infrastructure diagrams: look at what a search actually returns. A product built on its own index tends to return an answer with the specific passage it came from, because the content has already been read and is sitting there to cite. A product relaying someone else's index or someone else's results page tends to hand back a list of links for you to open yourself — it doesn't have the content on hand to quote, because it never held it in the first place. That single tell — a passage, or a link — is usually enough to know whether you're one hop from the source or two.
Where askFinz sits
askFinz runs its own crawled web index rather than calling someone else's — the reading, ranking and storage described in how the index works happen on infrastructure askFinz operates, not a licensed feed. That's the basis for the fuller, numbers-based version of this argument, comparing askFinz's own rates against the market's, at askFinz vs traditional search indexes.
Join the beta to see what an owned index returns, against a real question.