Does GEO Actually Work? What the Research Says
Generative Engine Optimization promises to get your content cited by AI. One controlled study (KDD 2024) actually measured what moves the needle — and a lot of the popular advice does not. A skeptical, sourced look.
TL;DR — "GEO" (Generative Engine Optimization) is mostly hype layered over a little real evidence. The one controlled study — GEO, KDD 2024 — found that adding statistics, quotations, and citations to reputable sources improved a page's visibility in AI answers by 30–40%. Keyword stuffing did nothing. Most of the rest ("Claude prefers long-form," "+71% from expert quotes") is vendor folklore. Write well, cite sources, and don't overthink it.
Every few months a new acronym promises to crack the algorithm. The current one is GEO — Generative Engine Optimization — getting AI engines like ChatGPT, Perplexity, and Google's AI Overviews to cite your page. There's a whole consulting industry forming around it. So let's do the unglamorous thing and ask: what's actually proven?
The one study that matters
Most GEO advice is people watching outputs and guessing. The exception is a peer-reviewed paper, "GEO: Generative Engine Optimization" (Aggarwal et al., KDD 2024), from Princeton, Georgia Tech, and the Allen Institute for AI. They built a 10,000-query benchmark, generated AI answers over real search results, tested nine content edits, and measured how prominently each source showed up.
The winners were not subtle:
| Edit | Effect |
|---|---|
| Add quotations from credible sources | ~+41% visibility |
| Add statistics (concrete numbers vs. vague claims) | ~+34% |
| Cite sources (inline links to reputable references) | ~+30% |
| Improve fluency / readability | ~+30% |
| Keyword stuffing | negative — it hurt |
So the causal, tested levers are: quotations, statistics, and citations. The thing traditional SEO spammers love — cramming keywords — was empirically useless for AI engines, and slightly counterproductive.
That's it. That's the robust finding. Notice this post is full of numbers, quotes, and links — not by coincidence.
What's probably true (but only correlational)
Beyond that study, the better advice is observational — patterns people have noticed, not controlled experiments. Treat these as reasonable bets, not laws:
- Put the answer near the top. A Search Engine Land analysis found ~44% of ChatGPT citations came from the first third of the content. A TL;DR or direct first-paragraph answer plausibly helps.
- Use question-shaped headings. Real FAQ-style
H2/H3s with self-contained answers correlate with more citations. (The writing helps more than the FAQ schema, which Google says you don't need.) - Write self-contained sentences. If a claim gets pulled into a retrieval chunk, it should make sense alone — name the subject instead of relying on "it" from three paragraphs up.
- Keep it fresh. Pages updated recently tend to get cited more than stale ones.
- Show your author and credentials. Bylines, bios, a named publisher — Google frames quality around E-E-A-T (Experience, Expertise, Authoritativeness, Trust), and AI answers lean on the same authority signals.
What's overhyped or made up
Here's where to keep your wallet closed:
- Oddly specific multipliers like "+71% from expert quotes" or "19+ statistics doubles citations." The direction matches the research; the exact numbers don't trace to any primary methodology. They're marketing.
- Platform personality lore — "Claude prefers long-form, Perplexity rewards freshness." These are vendor-blog vibes, not controlled studies.
- The fake-brand experiments. The widely-shared test that invented a brand ("Xarumei") and watched AI repeat its claims was criticized for testing content in an authority vacuum — it mostly proved that when there's no real authority, answer-shaped detail wins by default.
- "GEO is a whole new discipline." Google's position is blunt: optimizing for AI features is still just SEO. You don't need
llms.txt, content "chunking," or AI-specific rewrites.
So what should you actually do?
Almost embarrassingly old-fashioned things:
- Be genuinely useful and specific. Original data, real numbers, firsthand experience.
- Cite reputable sources inline, and quote experts. The proven lever — and it makes your writing better anyway.
- Structure for skimming. Clear headings, a TL;DR, lists and tables, self-contained sentences.
- Establish authority. Real author identity, a real publisher, genuine off-site reputation.
- Stay technically clean. Crawlable, fast, mostly-static HTML (AI fetchers often see only the initial HTML), and don't block the AI search bots.
The uncomfortable truth is that "GEO" is mostly just good writing plus real authority, with one genuinely useful research-backed nudge: stack your stats, quotes, and citations. Everything else is someone trying to sell you a framework.
FAQ
Does GEO actually work?
Partly. A controlled 2024 study found that adding statistics, quotations, and citations improved AI-answer visibility by ~30–40%. Most other GEO advice is correlational or unproven, and keyword stuffing was shown to not work.
What's the difference between GEO and SEO?
Contested. Google says optimizing for AI features is still just SEO. GEO emphasizes being cited by AI answers (via stats, quotes, citations, structure) rather than ranked in a list, but the fundamentals overlap heavily.
Do I need llms.txt or special AI formatting?
No, not strictly. Google says you don't need llms.txt, content chunking, or AI-specific rewrites. They're low-cost extras some tools use — see our AI-readable blog guide — but good, authoritative, well-structured content does the heavy lifting.
Sources: GEO paper, KDD 2024 (arXiv), Google AI optimization guide, Search Engine Land, Search Engine Journal.
Image: Muhammad Rafizeldi, CC BY-SA 4.0, via Wikimedia Commons.
← Back to all posts