When Search Engines Start Answering Questions Themselves, What Should Content Creators Do?

Out of habit, I opened Perplexity to look something up. After I asked my question, it gave me back a full paragraph — viewpoints, data, citations, with the sources listed neatly below.

I froze for a second.

Because I suddenly realized something: it had been a long time since I clicked through to the second page of search results. I couldn't even be bothered to open every link on the first page anymore.

The search engine was no longer handing me a "go find it yourself" list. It was serving the answer on a silver platter.

For users, of course, this is a wonderful thing. But have you ever stopped to think about what it means for the websites sweating over their content?

From SEO to GEO: The Rules Have Changed

For thirty years, the logic of search engines was simple: you type in keywords, it throws back a pile of links, and you click through them one by one. The ones at the top get seen; the ones at the bottom are left out in the cold.

So an entire industry was born — SEO, Search Engine Optimization. How to stuff keywords, how to buy backlinks, how to push your ranking up. It kept countless marketing companies in business.

But now, search engines have changed.

They no longer just hand you links. They use large language models to blend information from multiple sources together and directly generate an answer. These systems have a name: Generative Engines. Google SGE, Bing Chat, Perplexity all belong to this category.

So here's the question: if users no longer click through links one by one, but get all their information from a single generated paragraph, then how many times your website gets cited, where it sits in the answer, how long the citation is — those are what really count as "visibility."

A group of researchers from Princeton University and the Indian Institute of Technology (IIT) gave this phenomenon a name: GEO, Generative Engine Optimization. Published in 2024 at KDD (the ACM SIGKDD conference on Knowledge Discovery and Data Mining — a leading data-science conference).

In plain terms: the next generation of SEO.

The Old Tricks Don't Work Anymore

The first thing the researchers did was test whether the old SEO-era tactics still worked.

They built a generative engine similar to Perplexity, gathered ten thousand real queries from nine data sources, and built a test set called GEO-bench, covering 25 domains.

Then they started testing.

Keyword stuffing — the bread and butter of SEO, frantically cramming as many keywords as possible into a page. The result? In a generative engine, it performed 10% worse than doing nothing at all.

Yikes.

Think about that. For the past two or three decades, the single most core tactic of the entire SEO industry — in front of generative engines, not only does it stop working, it actively backfires.

Why? Because behind a generative engine is a language model. It no longer relies on keyword matching to understand a web page; it is genuinely "reading" your content. Stuff a pile of keywords in, and when it reads, it decides your page is low quality — and becomes less willing to cite you.

What Actually Works?

The researchers tested nine optimization methods. Three of them far outperformed the rest:

Cite sources. When your page mentions a statistic or makes a claim, annotate where that information came from. Just that one simple action can boost your visibility in an answer by over 30%.

Add quotations. Directly quote things said by relevant people inside your content. The single most effective method, delivering a 40% relative improvement.

Add statistics. Wherever you can make a point with numbers, don't use adjectives.

Here's the thing — these three tactics share one thing in common: they all increase the credibility and information density of the content itself.

When a generative engine produces an answer, it has a natural preference for content that is verifiable, quantifiable, and quotable. The more solid the information you provide, the more willing it is to fold you into the answer.

It actually makes a lot of sense. The engine's goal is to give the user a good answer. The higher the quality of your content, the greater the chance it gets selected. The rule shifted from "whoever has the most keywords" to "whose content is more credible."

The Underdog Sites Strike Back

But the finding that surprised me the most in this research was this:

GEO helps smaller, lower-ranked sites even more.

The researchers ran an experiment. They pulled out the sites ranked first through fifth in search results and looked at them separately. After the same GEO optimization, the fifth-ranked site's visibility jumped 115%. Meanwhile, the first-ranked site actually dropped 30%.

Why?

Because traditional search engines rank by overall strength: domain authority, number of backlinks, years in operation. Big sites have a natural advantage. But generative engines are different — they are reading and understanding the content itself. If your content is good, they use it, regardless of how old your domain is or how many backlinks you have.

This means that, for the first time, small creators have a chance to compete with the big sites on a relatively level playing field.

Of course, how long this "fairness" lasts is hard to say. As everyone starts doing GEO, the engines will evolve too. But at least in this moment, the window is open.

No One Tactic Wins Everywhere

The researchers also found something else: the best strategy differs by domain.

For debate-style and history-type questions, writing in a more persuasive, more authoritative tone works well. For legal and government-related content, adding statistics is the most popular. And for topics involving people and social issues, directly quoting relevant people beats everything else.

There is no one-size-fits-all method.

They also tested combination strategies. The best pairing was: first smooth out the writing, then add statistics. Combined, those two outperformed any single strategy by an extra 5.5%.

In plain terms, this is a process of constant experimentation and tuning. What your field is, what your readers care about — those determine which tactics work best for you.

Does It Hold Up in the Real World?

To make sure their conclusions weren't just a lab result, the researchers also validated against Perplexity.ai — an actual commercial generative engine already serving millions of users.

The results matched what they'd observed in their own system: citations and statistics still worked, keyword stuffing still didn't. The single best method delivered a 22% to 37% lift on Perplexity.

This means these methods aren't academic gimmicks that only work in a specific system — you can take them and use them directly.

Final Thoughts

I've been thinking about something.

In the SEO era, what everyone studied was how to game an algorithm. Stuffing keywords, manipulating backlinks, buying reciprocal links — at the end of the day, it was all about gaming a ranking system.

In the GEO era, the rules have changed. When a generative engine selects content, what it's essentially judging is: can this passage actually help the user answer their question?

What you need to do is make sure your content genuinely answers questions. Cite sources, provide data, quote valuable perspectives. These are exactly the things you should have been doing all along.

Maybe search engines have finally evolved to the point where they start rewarding people who take content seriously. You no longer have to figure out how to game the algorithm — you only have to figure out how to make your content solid.

The GEO-bench test set is already public. If you create content, this is something worth paying attention to early.

I opened Perplexity again, and asked the same question. The answer was crisp, citing five sources.

I counted. The fifth source got a lot more visibility than before. The creator of that site probably has no idea what happened.