In the previous post we mapped the surfaces where AI now sells: ChatGPT, Google AI Mode, Perplexity, and the Amazon app. Being present on those surfaces is necessary, but it is not the same as being chosen. When a shopper asks an AI "what is the best running shoe for flat feet under $120," dozens of brands are technically eligible and only two or three make the answer. This post is about how to be one of them. The discipline has a name: GEO, Generative Engine Optimization.
Why AI discovery is a different game than ranking
Classic SEO produces a ranked list. The shopper still does the choosing, scanning ten links and clicking one. An AI engine collapses that. It reads across many sources, synthesizes them, and hands back one answer with a short list of named products. There is often no page of links to scroll, so "position 4" has no meaning. You are either in the synthesized answer or you are invisible.
That changes what you optimize for. The AI is not matching your keywords to a query; it is trying to decide which brands it can confidently and defensibly recommend. It rewards content it can extract cleanly and claims it can trust. If you are new to the term, our plain-English guide to GEO covers the basics, and the 7 Pillars of GEO breaks the framework into a full checklist. This post is the strategic layer above both: why the game changed, and where to spend your effort first.
What actually moves the needle (and what backfires)
The most useful evidence here is not opinion. In 2024, researchers from Princeton, Georgia Tech, and allied institutions published the first large GEO study (presented at KDD 2024), testing optimization methods across thousands of real queries to see which ones actually increased a source's visibility inside AI-generated answers. Their findings are counterintuitive if you come from SEO.
Add citations, statistics, and quotes
The single biggest lever in the study was adding evidence: citing authoritative sources, including relevant statistics, and quoting named experts. These methods raised a source's visibility in AI answers by up to roughly 40 percent. AI engines are built to produce trustworthy, groundable answers, so content that already looks sourced and factual is easier for them to lift and attribute.
- Put real numbers on your claims ("cuts setup time from 20 minutes to 3"), not adjectives ("super fast").
- Quote a named person: a founder, an engineer, a verified customer.
- Link to primary sources for any factual claim, the way this article does.
Why it works: evidence is the highest-leverage GEO pillar. Persuasive copy with no facts behind it does not get cited.
Stop stuffing keywords
In the same study, keyword stuffing was one of the weakest tactics, and for some content it made things worse. AI does not count keyword density; it models meaning. A page written naturally, that clearly answers a real question, reads better to a language model than one padded with "best wireless earbuds for running" eleven times. The old SEO reflex is now a liability.
Do instead: write the way your customer asks the question, once, clearly. Natural, conversational phrasing matches how people prompt AI ("best X for Y under $Z").
Make each claim quotable on its own
The study also found that lower-ranked pages had the most to gain, up to about 115 percent more visibility, when they added these evidence signals. The reason is structural: AI extracts self-contained passages. A tight paragraph that answers one question completely, with a fact in it, can be pulled into an answer even if the page as a whole is not the top Google result. Being extractable partly levels the field against bigger competitors.
Do instead: write short, self-contained sections (roughly 40 to 160 words) under question-shaped headings, so any one of them can be quoted standalone.
The uncomfortable truth: most of your AI visibility is not on your site
Here is the finding that reframes the whole strategy. An AirOps analysis in October 2025 found that roughly 85 percent of brand mentions in AI answers came from third-party domains, review sites, publisher articles, forums, and comparison pages, not the brand's own website. Ahrefs' December 2025 look at AI brand visibility pointed the same direction: what correlates with showing up in AI answers is off-site signal, brand mentions and reputation across the web, more than on-page tweaks alone.
In other words, the AI's opinion of you is mostly assembled from what other credible sources say about you. Your own product pages still matter (they are where the AI verifies price, specs, and availability), but they are not where most of the recommendation is earned. This is uncomfortable because it is the part you do not fully control, and it is exactly why it is worth the effort.
The listicle trap: how your own page can sell your competitor
One more warning, because it catches good merchants off guard. In June 2026, Lily Ray documented for Search Engine Land that Google's AI Overviews frequently pull from self-serving listicles, the "best 10 X" roundups brands publish, and then recommend a competitor named further down the same list. The AI reads every brand on the page, not just the one that published it.
So a "Top 10 running shoes" post on your own blog can get cited by an AI that then tells the shopper to buy number four, which is not you. If you publish comparison content, lead with your own differentiated strengths, be deliberate about which competitors you name, and do not hand the AI a tidy list that frames a rival as the winner.
Do this now: the evidence pass
Pick your three highest-intent pages (the ones tied to how customers actually search) and run a single editing pass focused on evidence, not keywords:
- Add one real statistic per key claim. Replace "fast shipping" with the actual number.
- Add one named quote. A founder line or a verified customer review with a name attached.
- Cite a source for any factual or comparative claim, and link it.
- Break walls of text into self-contained, question-headed sections that quote cleanly.
- Delete the keyword padding. Read each paragraph aloud; if no human would say it, cut it.
- Line up one third-party mention. A review, a roundup, a partner post. This is the 85 percent.
None of this requires new technology, just a different definition of "good content": sourced, extractable, and corroborated elsewhere. Next in the series: even the best-written page is invisible if an agent cannot parse it. We move to the data layer, making your catalog machine-readable with structured data, clean feeds, and product pages agents can actually read.
Frequently asked questions
What is GEO, and how is it different from SEO?
GEO (Generative Engine Optimization) is the practice of getting your brand and products cited inside AI answers from ChatGPT, Perplexity, Google AI Overviews, and Gemini. SEO optimizes for a ranked list of ten blue links a person then chooses from. GEO optimizes to be the one source the AI synthesizes into a single answer, often with no click at all. The tactics diverge: things that help SEO (like dense keywords) can hurt GEO, while things SEO ignores (clear citations, statistics, expert quotes, and third-party mentions) are what get you pulled into an AI answer.
Does keyword stuffing help me show up in AI answers?
No. It can actively hurt. The Princeton and Georgia Tech GEO study (KDD 2024) tested optimization methods across thousands of queries and found that stuffing keywords was one of the weakest tactics, while adding citations, statistics, and quotations lifted a page's visibility in AI answers by up to about 40 percent. AI engines synthesize meaning rather than match keywords, so evidence and clarity beat repetition.
Where do AI answers get their brand recommendations from?
Mostly from sites you do not own. An AirOps analysis in October 2025 found that roughly 85 percent of brand mentions in AI answers came from third-party domains such as review sites, publisher articles, forums, and comparison listicles, not the brand's own website. Your product pages still matter, but the AI's opinion of you is largely assembled from what other credible sources say about you.
Why does AI sometimes recommend my competitors on my own pages?
Because AI reads every name on the page, not just yours. Analysis by Lily Ray for Search Engine Land in June 2026 found that Google AI Overviews frequently pull from self-serving listicles (best-of roundups) and then recommend a competitor listed further down the same page. If your comparison content names rivals, an AI can cite your page and still send the shopper to them. Lead with your own differentiated strengths and be careful how you frame competitors.
Sources & further reading: GEO: Generative Engine Optimization (Aggarwal et al., KDD 2024), AirOps: the influence of offsite signals in AI search, Ahrefs: AI brand visibility correlations, and Lily Ray / Search Engine Land: AI Overviews cite self-serving listicles. Percentages are drawn from the cited studies and reflect the conditions those researchers tested; treat them as directional, not guarantees.