Why “a page becomes a prompt”?

As mainstream search engines and ad platforms are integrating more LLM-based technologies, it is pretty important to understand why a page(a blog p…

As mainstream search engines and ad platforms are integrating more LLM-based technologies, it is pretty important to understand why a page(a blog post, a landing page, a product detail) becomes a prompt. It will benefit you in optimizing your content and marketing in LLM-enhanced search engines.

conversation.png

How it works

Both search engines and advertising platforms are answering pretty much the same question: based on a user’s context and related content available, what should be displayed to the user?

The search keywords users typed in the search box are not the full content of the prompt for a search engine(or a chatbot) to generate the corresponding answers. Several hidden steps precessed background actually help the search engine to extend the user’s input into a full prompt for LLMs to generate answers:

  1. Search engines get the search result pages along with their content. This is almost the same as what search engines are always doing, and we should call those result pages SERPs.
  2. Construct a prompt based on the user’s keywords or questions with SERPs(or embeddings of SERPs).
  3. Feed the constructed prompt to LLM and generate final answers.

As we see in the second step above, the content of pages becomes part of a prompt generated. Here we can conclude that a page becomes a prompt.

The case is similar for online advertisement platforms. Consider an advertisement platform preparing an ad for a visitor. The content of the ad could be generated by a prompt composed of the visitor’s info and the content of a promoted landing page. Here a page becomes part of a prompt for generating the ad content.

How does it affect us?

In the past, search engines handled how to index and lookup keywords in a very large dataset. But when the search results are presented to users, search engines depend pretty much on website owners’ work to determine how the data would be represented, and this is where the metadata on the pages take effect.

With the help of LLMs, the content of a page could generate the meta tags, the description, the text for ad display, and different kinds of JSON formatted meta information…, all by itself. Every human-curated metadata in the past decades is becoming less human-dependable every day. With the help of cloud computing, the machine can do it on a very scale of pages. Eventually, everything that matters is only the content of a page.

The grand shift powered by LLMs will take us back to the very beginning location of marketing — the content itself.