Quick question before we get into this:
Has your organic content traffic dropped in the last 6 months?
Three weeks ago, a founder I mentor called me.
He runs a forty-person media company built entirely on Google traffic, which has dropped roughly 30% since July.
He wanted to know what to do.
So I spent the next few weeks going deep on how these systems work.
What changed
Think about how you used to search for something.
You typed a question, Google showed you ten links, you clicked through and read, maybe you subscribed to someone's newsletter.
The entire economic engine of the internet ran on that sequence of events.
Now:
You type the same question, a box appears labeled "AI Overview," you read three sentences, you have what you need, and the tab closes.
The information enters your head and the person who spent hours writing the source article gets nothing.
When AI summaries appeared, only 8% of users clicked through, compared to 15% without the summaries.
Users ended their browsing session entirely 26% of the time when an AI summary appeared, compared to 16% when it didn't.
What I see
My mentee was just the first in his cohort to feel the full weight of what had changed.
He had built on a contract the whole internet ran on for twenty years: create useful content, Google sends readers, monetize those readers.
What frustrates me about this is that it was a design choice.
The AI systems that could have been built to send traffic were built to retain it instead.
So how does this work?
Traditional search built an index.
Google crawled the web, ranked pages by authority and relevance, and handed you a list of links to choose from.
The content lived on someone else's server and you had to go there to read it.
Perplexity works differently.
This approach is called Retrieval-Augmented Generation, or RAG
Here’s how it works:
To do this at scale, Perplexity maintains an index of over 200 billion URLs.
And part of its retrieval infrastructure runs on Vespa, a search engine built to handle this kind of large-scale hybrid retrieval.
It uses two types of search running simultaneously.
The first is semantic search, which converts your question into a mathematical representation and finds content with similar meaning even when the exact words don't match.
Ask about "making money online" and it surfaces content about "digital revenue streams" because the underlying meaning is close.
The second is lexical search, the traditional keyword match, for when you need an exact term or name.
Once it retrieves relevant content, a language model synthesizes it into a single answer with inline citations.
Perplexity's CEO calls it the "strict grounding principle": the AI is instructed never to state anything unsupported by retrieved sources.
The AI can't function without sources, but it doesn't need yours specifically. Any credible source on the same topic will do.
So the new question is: why would the AI choose you?
The study
They tested nine content strategies for getting AI systems to cite your work, and three held up consistently across every domain.
Citing credible sources in your own content.
When you reference authoritative research, the AI treats your content as more authoritative in return.
This is essentially PageRank logic applied to AI retrieval: the system learned what authoritative content looks like by observing citation patterns.
Citing good work makes you look like good work.
Named experts with verifiable credentials.
Unattributed claims and anonymous opinions get filtered out. A quote from a named researcher with a traceable affiliation carries real weight in retrieval.
Real data.
It could be any number, percentage, or research findings.
There is also a technical requirement that is easy to miss.
PerplexityBot fetches the raw HTML of a page without executing JavaScript.
Most modern websites are built so that content loads after the page opens, using JavaScript running in the browser. The raw HTML the crawler first sees is often just an empty shell.
If your content loads dynamically, the crawler sees an empty page and your work becomes invisible to the model regardless of how good it is.
Server-side rendering is now an entry requirement for AI discoverability.
38% of AI citations come from pages already ranking in Google's top ten.
Getting cited by AI and getting found by Google are still related problems.
The trap
The platforms that built the web's information economy are watching AI scrape their value and redistribute it for free.
News Corp signed a $250 million licensing deal with OpenAI.
The New York Times is suing them. 79% of top news sites now block AI crawlers entirely.
I understand the instinct.
But blocking the crawler means losing citation visibility completely, and letting it in means your content gets used and your readers never arrive.
There is no version of this where the content creator wins cleanly.
The choice
The old playbook was keywords: find what people search for, rank for those terms, capture the traffic.
The new playbook is becoming a primary source.
If your content can be easily restated, it will be restated, and your ideas will exist in the AI's answer while you remain invisible.
What survives this is harder to produce and harder to copy.
Original research, proprietary data, analysis that comes from years of doing the work rather than synthesizing what others already published.
That is where LLMs index heavily, and a recommendation in a high-traffic thread can do more for your AI visibility than a well-optimised article.
Reply and tell me what you are seeing out there.
Nandini