DEV Community

Fay
Fay

Posted on • Originally published at xanlens.com

What is Generative Engine Optimization (GEO) and Why Your Brand is Invisible to AI

#ai #seo #marketing #webdev

Most brands have no idea how AI engines describe them.

Not Google. Not Bing. The AI engines: ChatGPT, Gemini, Perplexity, Claude, Grok, DeepSeek. The ones people are increasingly using instead of search.

A Princeton and Georgia Tech research team coined the term Generative Engine Optimization (GEO) in their 2024 paper (arXiv:2311.09735, published at KDD 2024). Their finding: traditional SEO techniques have almost zero correlation with visibility in AI-generated answers. The ranking factors are completely different.

How AI Engines Decide What to Mention

Each engine has its own retrieval pipeline. They do not all work the same way.

Gemini reads structured definitions and your opening paragraph. It uses Google Search grounding to pull live web data, but it also has training data biases. If your first 200 words are ambiguous, Gemini will fill in the blanks with hallucinated context. We learned this the hard way: Gemini described our own tool as a "life sciences data analysis platform." We are a GEO audit tool.

Perplexity indexes live web content aggressively. It reads your first paragraph and title, then decides if you are worth citing. It has 39.5 million backlinks from 82,000 domains and generates hub pages for every query pattern. If Perplexity does not have a third-party source mentioning your brand, you do not exist in its answers.

DeepSeek pulls primarily from training data, not live web. If your brand was not in its training corpus, it will literally ask the user: "Is it possibly XanLens, XenLens, ZanLens, or a similar spelling?" That is a real response from our audit.

Grok uses X (Twitter) data and live web search. It picks up social signals faster than other engines. If you have an active X presence discussing your product, Grok will find you. If not, it may categorize you as fictional.

The Numbers That Matter

The Princeton GEO paper tested optimization strategies across generative engines and found:

  • Adding citations and statistics to your content increased visibility by 30-40% in AI answers
  • Authoritative language and technical terminology improved inclusion rates
  • Traditional SEO signals like keyword density had near-zero effect
  • Content structure (clear definitions, FAQ format, schema markup) strongly correlated with AI citation

Separately, SE Ranking studied 2.3 million pages and found that domain traffic was the number one predictor of appearing in AI-generated answers. Not backlinks. Not keyword optimization. Traffic.

AI referral traffic is still small in absolute terms (roughly 1% of total web traffic) but growing at 130%+ year over year. The brands that establish AI visibility now will have a structural advantage as that percentage climbs.

What GEO Actually Involves

GEO is not one thing. It is a set of practices across three categories:

1. Identity signals (on-site)

  • Your H1 and first 200 words define you. AI engines read this first.
  • Schema markup (Organization, Product, FAQ, HowTo) gives engines structured data to extract.
  • Your robots.txt matters: if you block GPTBot or Google-Extended, those engines cannot index you.
  • A /llms.txt file explicitly tells AI crawlers what your product does.

2. Off-site presence (citations)

  • Third-party mentions on platforms like Dev.to, Hashnode, Medium, Reddit, Stack Overflow, Product Hunt, and industry publications.
  • AI engines weight third-party sources heavily. If only your own site mentions your brand, citation scores stay low.
  • Perplexity in particular relies on hub pages and authoritative external sources.

3. Monitoring

  • The only way to know if AI engines mention you is to ask them. Regularly.
  • Each engine may describe your brand differently, or not at all.
  • Scores change as engines update their models and index new content.

We Audited Ourselves

We built XanLens to do exactly this: audit how visible a brand is across AI engines. Score from 0 to 100, tested against 132 prompts across live engines.

Then we ran it on ourselves.

Score: 16 out of 100.

Knowledge score: 17. Discovery score: 18. Citation score: 29.

Gemini thought we were a pharma company. DeepSeek did not know if our name was spelled correctly. Grok categorized us as fictional.

We had robots.txt blocking AI crawlers. Our meta title was 27 characters. Our schema markup was 62% complete. We had zero off-site content.

The tool worked. It just told us what we did not want to hear.

Now we are fixing it, publicly, and tracking the score changes. We call it Day Zero.

What To Do Next

If you have never checked how AI engines describe your brand:

  1. Ask ChatGPT, Gemini, and Perplexity: "What is [your brand]?" Compare the answers.
  2. Check your robots.txt for AI crawler blocks (GPTBot, Google-Extended, Bytespider).
  3. Read your first 200 words out loud. If a stranger could not tell what you do from those words alone, rewrite them.
  4. Search for your brand name on Dev.to, Reddit, and Medium. If nothing comes up, that is your citation gap.

GEO is not a replacement for SEO. It is a parallel channel that most brands have not started working on yet. The research says it matters. Our own audit says it matters. The 130% growth in AI referral traffic says it matters.

The question is whether you check before or after your competitors do.

Top comments (0)