Article
I looked into GEO. It is real, but please do not get weird about it
A grounded introduction to generative engine optimization, AI search visibility, and why the market is real without being nearly as certain as the sales decks make it sound.
I started looking into GEO because I wanted to know whether there was a real business here, or whether people had simply found a fresh acronym to sell old SEO work again.
The answer is annoying, which usually means it is closer to true.
GEO is real. It is also early, noisy, and full of people talking as if they already have the machine figured out. I do not think they do.
GEO usually means generative engine optimization, a term that came out of the 2023 paper “GEO: Generative Engine Optimization”. Some people call the same thing AEO, AI search optimization, LLM visibility, or answer engine optimization. The exact label matters less than the behavior it is trying to influence: when someone asks ChatGPT, Perplexity, Gemini, Copilot, Claude, or Google AI Overviews a buying or research question, does your brand get mentioned, recommended, or cited?
That is a different target from traditional SEO. Traditional SEO mostly cares whether your page ranks and earns clicks. GEO cares whether the answer engine uses you as source material or includes you in the answer at all. Sometimes you get a click. Often you do not. The exposure may happen inside the answer and never show up as a normal referral.
This is where the hype starts, because it is easy to say “search is dead” and then sell panic. But the data I found does not support that version.
SE Ranking studied 101,574 websites across 16 months and found that AI answer engines drove only 0.32 percent of total website traffic in 2026. In 2024, the number was 0.02 percent. In 2025, it was 0.24 percent.
So the absolute number is tiny. Roughly one out of every 312 visits came from AI platforms in that dataset.
But it grew 16x in two years.
AI referral traffic as a share of total website traffic in SE Ranking's 2026 dataset.
Growth from 2024 to 2026. Tiny base, fast slope. Both parts matter.
Longer average session duration for AI visitors compared with organic search visitors in the same study.
That is the uncomfortable part. If you only look at the current traffic share, GEO looks overblown. If you only look at the growth rate, it looks like something you probably should not ignore. I do not think this is a “drop everything” market. I do think it is a “start learning now while the rules are still forming” market.
The traffic quality also makes the situation more interesting. The same SE Ranking study found that visitors from AI platforms spent 67.7 percent more time on site than organic search visitors. Digiday’s 2025 publisher-side overview is messier, but it points to the same market shape: AI referrals are still small, and they are changing where some discovery starts.
That matches the product intuition. A person who asks “best LLM observability tools for a seed stage AI app” is not casually browsing. They are already in comparison mode. If the answer engine gives them five names and yours is missing, you may have lost before they ever touched Google.
Still, I would be careful with any vendor claiming they can guarantee results. The sources are unstable. The prompts are unstable. The engines do not behave the same way. One tool may say your brand is visible, another may say you are absent, and both may be sampling from synthetic prompts that no real buyer has ever typed.
Primary data
Academic papers, large datasets, official docs, and studies where the method is visible.
Use for claims.Industry reporting
Funding news, interviews, media reporting, and vendor data repeated by credible outlets.
Use for market direction.Vendor posts
Agency case studies, marketing reports, and self-published experiments.
Use as leads, not proof.This is the part that made me more interested, not less.
In boring markets, the obvious dashboard wins. In messy markets, trust matters. If everyone is overselling certainty, there is room for someone who says: “We cannot measure this perfectly. Here is how we measure it anyway. Here is the error. Here is what changed. Here is what we would do next.”
That is a much more honest product surface than “we will get you into ChatGPT.”
The other thing I noticed is that GEO is not a separate magic discipline. A lot of it is still good SEO and good content work. You need clear pages, useful comparisons, original facts, third-party mentions, citations, documentation, and pages that answer the questions people actually ask. The difference is that you are writing for a retrieval and synthesis layer as well as a search results page.
That changes the shape of the work.
You still ask, “Can this page rank for a keyword?”
You are also asking:
- Would an answer engine have enough facts here to cite this page?
- Does the page answer the follow-up questions created by query fan-out?
- Is the brand mentioned on third-party pages that the engine already trusts?
- If the engine compares the category, does it know what box this product belongs in?
- If the user asks for alternatives, does the brand appear in the comparison set?
The best short version I have is this: GEO is the practice of making your brand legible to answer engines, especially for high intent questions.
Legible is doing a lot of work there. It does not mean stuffing keywords. It does not mean adding an llms.txt file and pretending the job is done. It means the machine can find clear, quotable, well-supported information about you, both on your own site and in places it already trusts.
If I were explaining this to a founder, I would not say “AI search is replacing Google.” That is too broad and too dramatic.
I would say this instead:
Your buyers are starting to use AI tools during research. The volume is still small, but the buyers who do this are often closer to a decision. If your competitors show up in those answers and you do not, you have a visibility problem that normal SEO dashboards may not catch.
That is enough reason to study it.
Not enough reason to believe every screenshot on LinkedIn.
The rest of this series is my attempt to turn the research into something usable. First, how answer engines choose sources. Then how to run a small visibility diagnostic. Then how to write pages that are easier to quote, how to think about third-party mentions, how to measure without fooling yourself, and why I would start a GEO business as a service before trying to build yet another SaaS dashboard.
I still have mixed feelings about the whole thing. That is probably healthy. The market is real, but the current certainty level is fake.
That is exactly the kind of place where a careful person can learn faster than the people shouting.