Generative Engine Optimization
What GEO is, how it differs from SEO, and what one paper measured
“We used to optimise for a search engine to rank. Now we optimise for an answer engine to understand.”
01 — What is GEO
What is GEO?
GEO — Generative Engine Optimization is the practice of structuring content so it shows up, and gets cited, in what generative engines (ChatGPT, Perplexity, Gemini) actually say.
To be the cited source inside an AI answer, and the one it recommends.
SEO targets keyword rankings. GEO targets how a model understands meaning and relates entities.
02 — Why GEO matters
Why it matters now
Search habits changed
More people ask a question outright than type keywords.
Better-qualified traffic
Someone asking a specific question is usually closer to a decision.
Authority, demonstrated
When an AI quotes your position as the answer, that is the claim being made for you.
Reusable by construction
Content structured well for models turns out to be good social material too.
03 — GEO vs SEO
GEO against SEO, side by side
| Dimension | SEO | GEO |
|---|---|---|
| Goal | Rank | Be the answer |
| Audience | Google's algorithm | Large language models |
| Method | Keywords, backlinks | Structured data, semantic relationships |
| Output | A list of blue links | An answer, with citations |
04 — GEO research, KDD 2024
Nine methods, measured: the KDD 2024 study
A paper presented at KDD 2024 by researchers at Princeton University and IIT Delhi and two independent researchers, "GEO: Generative Engine Optimization" (Aggarwal et al.), tested nine ways of rewriting a web page and measured how much of a generative engine's answer each one earned the source. Eight raised it; one lowered it:
| Method | Change vs. baseline | What it means in practice |
|---|---|---|
| Quotation addition | +41% | Add direct quotations from relevant people or experts, and say who said them. |
| Adding statistics | +31% | Replace vague wording with specific numbers — percentages, counts, durations. |
| Fluency optimization | +28% | Make the sentences read more smoothly and the paragraphs connect. |
| Citing sources | +27% | Back claims with external sources a reader can check. |
| Technical terms | +18% | Use the field's accepted technical terms where they fit. |
| Easy-to-understand | +14% | Rewrite complex ideas in plain, direct language. |
| Authoritative tone | +10% | Use a more assertive, persuasive tone. |
| Unique words | +6% | Vary the wording instead of repeating the same terms. |
| Keyword stuffing | −8% | Repeat the query's keywords on purpose. In this experiment it did worse than changing nothing. |
The figures are this site's conversion of the paper's Table 1 into relative change: the overall Position-Adjusted Word Count (how much of the answer comes from the source, weighted by how early it appears) against the unmodified baseline of 19.3. The engine was GPT-3.5-turbo answering from the top five Google results, on the paper's own GEO-bench queries. Source: arXiv:2311.09735
What these numbers don't tell you
They are averages on one benchmark, not a promise. The paper itself notes that generative engines keep changing, so methods may need to change with them, and that the questions people ask change too; the effect also varies by domain. A check on Perplexity.ai pointed the same way, by different amounts. What was measured is a source's share of the answer — not clicks, enquiries or revenue.
05 — How GEO works
How we run a GEO engagement
Step 1. Build the answer container
Not a website so much as a library laid out for machines.
- Clean Schema.org markup
- Question-and-answer content structure
- Fast enough that nothing times out
Step 2. Fill it with something worth citing
Models reward a distinct position and discard filler.
- Answers to problems people actually have
- Expert judgement with evidence attached
- Kept current, with the dates to show it
Step 3. Build the trust graph
Get authoritative sources standing behind your answer.
- Relevant outbound links
- Social signals and brand mentions
Next step
If you want to know how your own site shows up in AI answers, the place to start is measuring where it stands now.
Or write to hello@taux.io