By Mehdi HassanWhat is generative engine optimization?
Generative engine optimization is the work of making your brand and pages show up inside AI-generated answers, as a cited source, a named recommendation or both. The engines are systems that write a response instead of listing links: ChatGPT, Gemini, Claude, Copilot, Perplexity and Google's AI Overviews and AI Mode.
The term is new but the work isn't. Most of it overlaps with SEO and with good reference writing, which is why we treat it as part of our answer engine optimization guide rather than a separate discipline. For a two-line definition, see what GEO is; for how it relates to the other labels, see SEO vs AEO vs GEO.
What did the Princeton GEO paper actually find?
The paper that coined the term, published in 2023 by researchers from Princeton and collaborators, tested whether rewriting web content could change how visible it was in answers from a generative engine. It built a benchmark of queries, applied different rewriting methods to source pages and measured how much of each answer drew on them.
Its headline result was that some changes helped and others didn't. Methods that made content more credible and specific, such as citing sources, adding quotations from relevant experts and including statistics, improved visibility in the test setup. Traditional keyword stuffing did not help. The effects also varied by topic, which is a useful caution against one-size rules.
What have 2026 industry studies added?
Field studies in 2026 looked at real citations at scale and mostly confirmed the paper's direction while adding two things it couldn't test: off-site reputation and freshness. They're correlational, so they show what cited pages have in common rather than proving cause.
- Mentions: Ahrefs' study of 75,000 brands found web mentions correlated 0.664 with AI Overview visibility, against 0.218 for backlinks and 0.392 for branded search.
- Freshness: AirOps reports over 70% of cited pages were updated within 12 months; Ahrefs found AI-cited content about 26% fresher than average.
- Position on the page: reported analyses of ChatGPT citations place about 44% of them in the first 30% of a page.
- Engine differences: Semrush's index of 126 million prompts found ChatGPT cites about 15 sources per answer, Gemini about 3.
- Rankings aren't required: Conductor says 57% of AI Overview citations come from outside the top 10.
The mechanism behind several of these, retrieval through many sub-queries, is covered in how AI search engines choose sources.
Which GEO tactics are proven, plausible or myth?
Sorting tactics by evidence is the most useful thing you can do with GEO advice. Proven means research and field data agree. Plausible means the logic holds and some evidence points that way. Myth means tests found no effect or the claim has no support.
Tactic | Evidence | Label |
|---|---|---|
Specific, sourced statements and statistics | Princeton paper plus field studies | Proven |
Answer near the top of the page | Citation position analyses | Proven |
Brand mentions on other sites | Ahrefs, 75k brands | Proven (correlation) |
Regular updates to key pages | AirOps and Ahrefs freshness data | Proven (correlation) |
Server-rendered, crawlable HTML | Many AI crawlers don't run JavaScript | Proven |
Expert quotations in content | Princeton paper only | Plausible |
Self-promotional best-of lists | About 4% of ChatGPT citations (Peec) | Weak |
AI-specific schema markup | Ahrefs 2026 test: no citation lift | Myth |
llms.txt files | Ahrefs: 97% got zero requests | Myth, for now |
Keyword stuffing for LLMs | Princeton paper: no help | Myth |
We go further into the myth column in AEO myths. The short version: if a tactic is mainly about adding files or markup rather than making content clearer or your reputation stronger, be sceptical.
Why schema and llms.txt keep getting recommended anyway
Both are easy to sell: they're concrete deliverables, they can be added in an afternoon and they sound technical. Google's guidance calls AI-specific markup and llms.txt unnecessary for its AI features, and Ahrefs' tests found no lift from schema and almost no crawler requests for llms.txt files.
That doesn't mean schema is useless. Organization, Article and FAQPage markup still help with rich results and clean parsing. It means they're hygiene, and nobody should promise citations from them.
How do you apply GEO to a real website?
Apply GEO through the pages you already need, starting with the ones buyers check before they talk to you. TripleDart's Q1 2026 study of 28 B2B SaaS companies found homepages took 31.9% of LLM citations and 81.7% of AI-attributed leads. Fix those pages before adding any new content.
Make sure engines can read the site
Server-render content and check robots.txt and CDN settings allow the AI crawlers you want. A client-rendered React marketing site is the most common blocker; see fixing a React SPA for SEO.
State facts plainly
Write what you do, who it's for, what it costs and how it compares in standalone sentences. Specific, sourced claims are the proven core of GEO.
Add evidence
Use real statistics with named sources, your own data and first-hand examples. Never invent numbers; an engine that repeats a fake stat makes you the liar.
Earn mentions
Publish something others will reference, show up in communities and comparisons, and keep your company description identical everywhere.
Measure per engine
Run a monthly prompt audit across ChatGPT, Perplexity, Gemini and AI Overviews and record citations and mentions separately.
The full fix list, ordered by evidence, is in the AEO checklist. If you're planning content from scratch, our SaaS SEO strategy shows how to sequence money pages before the blog.
Does generative engine optimization work?
Yes, in the sense that what appears in AI answers can be influenced, and the levers are known: clear content, real evidence, crawlability, freshness and reputation. No, in the sense that nobody can guarantee a citation. Engines change models and sources often, results vary by prompt, and correlation studies can't promise outcomes.
Set expectations around probability. A well-built, well-written, frequently mentioned site is more likely to be cited across many prompts. That's a sound investment because the same work also improves rankings and conversion. A GEO plan that only works for AI engines, and makes your pages worse for people, is a bad trade.
