▲ Now tracking visibility across Google, ChatGPT, Perplexity, Claude, Gemini & AI Overviews
Generative Engine Optimisation (GEO) is a methodology, originating from a 2023 academic paper by researchers at Princeton, IIT Delhi, the Allen Institute and Georgia Tech, for making written content more likely to be reused inside the answers that generative AI engines produce.
Where AEO covers the whole strategy (entity hygiene, schema, crawler governance, measurement), GEO is specifically about the words on the page. The researchers showed that small, targeted edits to a passage, adding cited statistics, expert quotations, authoritative tone, can lift the chance that an LLM selects and reuses it by up to 40% in their controlled tests.
In practice we treat GEO as the content-layer technique set inside a broader AEO program. The entity and schema work gets you in the candidate pool. GEO work gets your specific passages selected and quoted once you are there.
Aggarwal et al, 2023, "Generative Engine Optimization (GEO)" (arXiv:2311.09735)
Passage-level content patterns inside generative answers
Up to 40% visibility lift in researcher experiments (source-dependent)
BingChat-class generative engines; methods generalise to ChatGPT, Perplexity, Claude
Subjective + position-adjusted visibility per passage
01
Quotes passages directly inside generated answers, often verbatim. GEO writing patterns lift inclusion rate noticeably.
01
Quotes and paraphrases liberally with inline citations. GEO-optimised passages get pulled at meaningfully higher rates.
03
Anthropic
Prefers authoritative, well-cited passages. Statistic and quotation methods perform strongly here in our testing.
04
Pulls and summarises large bodies of source text. Fluency and authoritative tone matter most for inclusion.
05
Google Search
Generated answer block above the SERP. GEO methods overlap with classic on-page SEO but emphasise passage structure over page structure.
06
Where the original GEO research was tested. Methods proven on Bing-Chat-class engines; transfer well to Copilot today.
A page targeting a query
A passage targeting a generated answer
On-page SEO: title, headings, intent match
Passage writing: citations, statistics, quotations, fluency
Position 1-10 in a SERP
Selection rate and position-adjusted visibility inside answers
Backlinks, domain authority, brand
In-passage citations to authoritative sources, named experts
Longer pages often win for competitive queries
Focused, dense passages outperform long pages
3 to 9 months
2 to 8 weeks once passages re-publish and re-crawl
Thin content, duplicate content, slow site
Vague language, no citations, no statistics, low fluency
Decades of practitioner + Google patents
Aggarwal et al, 2023, "Generative Engine Optimization (GEO)"
Stage 01
Your lever: AEO-level work: crawl access, schema, authority
Your lever: Passage structure: tight, focused sections
Your lever: GEO methods: citations, stats, quotes, fluency
Your lever: Quotable phrasing + accurate facts
Method 01
Highest lift in the paper
Add in-text citations to authoritative external sources within your passage. The engine treats cited passages as more trustworthy candidates for inclusion.
Include specific quantitative data within the passage. Numbers are easier for an engine to cite as evidence than adjectives like ‘many’ or ‘a lot’.
‘Brands that publish original research are cited 3.2x more often in Perplexity than brands that don’t’ beats ‘cited more often’.
Quote named experts inline. Engines prefer to pass through verbatim, attributed quotes rather than synthesise from prose alone.
Moderate lift
Rewrite for clarity, brevity and active voice. LLMs are trained on fluent text and tend to select fluent passages over awkward ones, all else equal.
‘It is the case that the optimisation can be carried out by means of multiple methods’ becomes ‘You can optimise this in several ways’.
Method 05
Moderate lift
Write with confidence: definitive verbs, named entities, specific claims. Hedging language (might, perhaps, generally) reduces selection rate.
‘Schema markup helps AI engines understand your content’ beats ‘Schema markup may sometimes help AI engines understand your content in some cases’.
Method 06
Moderate lift
Use precise, domain-specific terminology where appropriate. Engines reward passages that demonstrate genuine expertise over generic summaries.
If you write good SEO content, you are most of the way to good GEO content. The shift is one of unit (passage, not page) and craft (specific citation patterns, not just keyword targeting).
Engines still need to believe you know the subject. Topical maps, clustered content, expert authors. All still essential.
Question-led headings, scannable paragraphs, lists and tables. Good for SEO, even better for GEO passage selection.
Author bios, real expertise, source citations. The same things that lift you in Google AI Overviews lift you in ChatGPT and Perplexity.
If you publish original data, you become the cite-able source for every prompt that touches your data. The most durable GEO move there is.
3,000 words is no longer "more authoritative". Engines pull 200-word passages. Dense, focused, cited passages beat long catch-all pages.
Hedging hurts. Write with authoritative tone where you have authority. Save the hedging for genuinely uncertain claims.
Listicles work for Google because they scan well. Engines prefer prose passages with embedded citations. Rewrite list-form pages with named, cited prose under each item.
Engines pick passages by meaning, not term frequency. Write naturally. Optimise for the prompt and the claim, not the keyword.
GEO lift typically shows up within 2 to 8 weeks of republishing, once the relevant engines re-crawl. The biggest near-term win is usually adding cited statistics to existing high-traffic pages; that is the fastest method to ship and the easiest to measure.
Senior strategist · No lock-in · Plain-English report
No spam. Unsubscribe in one click. We’ll never share your email.
