Aravind Srinivas and Perplexity’s Answer Engine
A search engine that answers instead of listing is a small idea with very large consequences, and Aravind Srinivas has spent the last few years proving it.
The most interesting thing about Perplexity is not its technology. It is that a company of a few hundred people made the search industry’s incumbents visibly nervous, which I would not have believed before I watched it happen. Nothing alarms a large institution like a small one doing the obvious thing.
The product does something simple: it answers a question and cites its sources, instead of returning ten links and leaving the work to the user. That is a small interface change with an enormous business implication, because search advertising is built around the clicks that an answer removes. One answer is one chance to sell, and the answer has already been given.
Why the incumbents cannot simply copy it
The obvious objection is that Google can do this, and it can. The difficulty is that doing it thoroughly cannibalises the advertising model that funds everything else. A company whose revenue depends on clicks has a structural incentive to be less decisive than a company that does not — and structural incentives outlast a great many clever memos.
That is the classic innovator’s dilemma, and it explains why an insurgent with no legacy revenue can move faster. He has nothing to lose but his investors’ patience.
The business model problem
- Inference costs scale with usage, unlike classic search, which is mostly fixed cost.
- Publishers want compensation for content that the answer summarises.
- Subscription revenue, not advertising, has to carry the model.
The publisher fight
The most fraught part of the story is the relationship with the websites whose content makes the answers possible. Summarising a page is, from the publisher’s perspective, close to republishing it without sending the reader. Several large publishers have pursued legal action, and the licence deals that followed have become a template for the industry. The quarrels that end in a licence deal are the ones where both sides counted the cost of the alternative.
Srinivas has argued that citation traffic is more valuable than generic search traffic, and that revenue-sharing arrangements are the sustainable solution. Publishers are not uniformly persuaded, but the direction of travel is clear: the answer engine must pay for its inputs. Whether it pays in coin or in readers the courts are still working out.
The cost structure problem
Classic search is a fixed-cost business: index the web once, serve queries cheaply forever. An answer engine pays for computation on every query, and the more useful the answer, the more it costs to produce. That is a strange thing to say aloud in a business meeting — that your best work is your most expensive work.
That inverts the industry’s traditional economics. Scale reduces unit cost in search; in an answer engine, growth in usage increases the absolute bill unless prices fall faster. This is why subscription revenue matters so much more here. A man who sells answers had better charge for them.
The browser and the assistant
The company’s expansion into a browser and an assistant is best understood as an attempt to own the starting point. If the assistant is the first thing a user interacts with, the search box becomes downstream of it.
It is an ambitious move for a company of its size, and it puts it in direct competition with organisations that have far more resources. That is either a category-defining bet or overreach, and the difference will be visible within a couple of years. I have been wrong about this before and expect to be again, so I will keep my opinion cheap.
Image credit and licence details for every photograph on this site are listed on the credits page. This article is editorial content; it carries no sponsored material.