Generative Engine Optimization: How to Get Your Business Cited by AI

A new category of search is growing fast. ChatGPT, Perplexity, Google AI Overviews, and Claude are answering questions that used to require clicking through to a website. AI referral traffic grew 527% year-over-year between January and May 2025, according to Presence AI's benchmark report.

The question for businesses is straightforward: when someone asks an AI "who is the best accountant in Sydney" or "what does a website cost in Australia," does the AI mention your business?

That is what generative engine optimization is about.

What generative engine optimization actually is

GEO is the practice of structuring content so that AI-powered search engines are more likely to cite it. The term was coined in a 2023 paper by researchers from Princeton University and IIT Delhi, published at the ACM SIGKDD Conference (KDD 2024) in Barcelona.

Traditional SEO optimises for Google's ranking algorithm: keywords, backlinks, technical performance. GEO optimises for a fundamentally different system. AI search engines do not rank pages in a list of ten blue links. They read multiple sources, synthesise an answer, and cite whichever sources contributed to that answer. The content that gets cited is not necessarily the content that ranks highest on Google.

This distinction matters. SE Ranking found only 10.7% URL overlap between pages cited in AI search and pages ranking in Google's top 10 for the same queries. A page could rank number one on Google and never be cited by ChatGPT, or rank on page three and appear in every AI answer.

GEO is not replacing SEO. Google still processes an estimated 14 billion searches per day, roughly 210 times more than ChatGPT's 66 million daily search-like prompts. But AI search is growing fast, the visitors it sends are unusually valuable, and the strategies that help content get cited by AI also tend to make that content better for human readers.

The Princeton research: what actually works

The Aggarwal et al. paper is the most rigorous study of GEO strategies published to date. The researchers tested 9 optimisation methods across 10,000 queries using a benchmark called GEO-bench, then validated results on Perplexity.ai in real-world conditions.

Here are the results, measured by relative improvement in visibility over an unoptimised baseline:

Source: Aggarwal et al., KDD 2024, results summarised by The GEO Community

Three findings stand out.

1. Citing sources is the single most effective strategy

Adding authoritative citations to content produced the largest visibility boost across all domains. For sites starting at mid-ranked positions, the improvement reached 115% on some metrics. This makes intuitive sense: AI models are trained to value verifiable claims, and citations give the model a reason to trust (and therefore surface) the source.

This is also the most democratising finding. Lower-ranked, less-established sites benefited more from citation strategies than well-known sites. If your business is not a household name, adding proper citations to your content gives AI models a reason to cite you over larger competitors.

2. Statistics and quotations outperform presentation changes

Strategies that added new, verifiable information to content (statistics, data points, direct quotes from named sources) consistently outperformed strategies that only changed how existing content was presented (making it more authoritative-sounding, more fluent, or simpler).

The implication: you cannot GEO-optimise thin content by rewriting it in a more "AI-friendly" tone. You need to add substance, specifically data and attributed claims.

3. Keyword stuffing actively hurts

Traditional keyword stuffing, the SEO tactic of repeating exact-match keywords throughout content, performed 10% worse than the unoptimised baseline in the Princeton study. AI engines reformulate queries and retrieve content by semantic similarity. The exact search query appears in only about 5% of AI answers. Optimising for exact-match keywords is counterproductive.

The researchers also tested pairwise combinations of the top strategies. The best combination was fluency optimisation plus statistics addition, which outperformed any single method by 5.5% or more.

How AI selects sources (and why it differs from Google)

Google ranks pages based on hundreds of signals: backlinks, domain authority, keyword relevance, technical performance, user engagement. AI search engines select sources through a different process.

The front-loading effect

Kevin Indig's analysis of 1.2 million search results and 18,012 verified ChatGPT citations found that 44.2% of citations come from the first 30% of content. After that point, citation likelihood drops sharply in what researchers call a "ski ramp" pattern.

This aligns with how large language models process text. LLMs weight early framing heavily and interpret the rest through that lens. Content that buries its key findings at the bottom, saving the conclusion for last, is structurally disadvantaged for AI citation.

Freshness matters more than in traditional search

76.4% of ChatGPT's top-cited pages were updated within the last 30 days. Pages with substantive content updates earned 3.8 times more citations than those with timestamp-only refreshes. Adding a visible "Last Updated" date lifted citation rates by 47%.

For comparison, Google gives some weight to freshness but regularly ranks pages that have not been updated in years. AI search engines are more aggressive about preferring recent content.

Entity recognition over domain authority

A controlled test across 9 sites found that domain authority (the SEO metric measuring backlink strength) showed a weak negative correlation with AI citation rates. What mattered more was whether the AI could clearly identify the entity behind the content, through structured data like Organisation schema with sameAs links to Wikidata, Crunchbase, and LinkedIn.

This is a meaningful departure from traditional SEO, where domain authority is a primary ranking factor.

Why this matters: the conversion data

AI search traffic is a small channel. It accounts for less than 1% of total web traffic for most sites. But the visitors it sends are disproportionately valuable.

The conversion premium comes from pre-qualification. When someone asks ChatGPT "which agency builds fast websites in Sydney" and then clicks through to a recommended site, they have already been pre-sold by the AI's endorsement. They arrive with higher intent than someone scrolling through ten Google results.

The Ahrefs data is especially striking: 0.5% of their traffic from AI search drove 12.1% of their signups. Those visitors also viewed 50% more pages per session than organic visitors.

Practical implementation: how to optimise for AI citation

Based on the Princeton research, practitioner case studies, and platform-specific data, here is what to do.

Front-load your data

Put the most important findings, statistics, and conclusions in the first third of every piece of content. Use a TLDR or key takeaway section at the top. 44.2% of citations come from this zone, so every high-value page should open with its strongest claims.

Structure matters too. 72.4% of pages cited by ChatGPT included an identifiable "answer capsule", a concise 40 to 60 word direct answer near the top of a section. This was the single strongest commonality among cited content.

Cite authoritative sources with links

Every factual claim should link to its source. This was the single most effective strategy in the Princeton study, producing up to 40% visibility improvement. In-text linked citations are better than footnotes or a sources list at the bottom. The AI model sees the citation in context and associates your content with the authority of the source.

Do not cite yourself or link to your own marketing pages as evidence. Cite independent, authoritative sources: government data, academic research, industry surveys, reputable publications.

Add statistics and quantitative data

Replace qualitative claims with quantitative ones wherever possible. "Most websites are slow" becomes "53% of mobile visitors abandon a site that takes more than 3 seconds to load (Google/SOASTA)."

The Princeton study found statistics addition produced up to 35% visibility improvement. Numbers are verifiable, specific, and give AI models concrete information to extract and cite.

Include attributed quotations

Direct quotes from named experts, practitioners, or research papers increase citation probability. The Princeton study measured up to 38% improvement from quotation addition. The quote must be attributed, naming who said it and in what context.

Update content regularly

Given that 76.4% of top-cited pages were updated within 30 days, a content refresh schedule is not optional. Substantive updates matter. Adding a new data point, updating a statistic with current numbers, or adding a recent case study all count. Changing a timestamp without changing the content does not: pages with timestamp-only updates received 3.8 times fewer citations than those with real updates.

Build entity clarity

Make it easy for AI to identify who you are. Implement Organisation schema with sameAs links to your Google Business Profile, LinkedIn, Wikidata (if applicable), and other authoritative profiles. Consistent NAP (name, address, phone) across directories helps too.

What NOT to do

The research is equally clear about what does not work.

Keyword stuffing. The worst-performing strategy in the Princeton study. AI engines match by semantic similarity, not exact keywords. The exact search query appears in only about 5% of AI answers. Repeating keywords degrades the semantic signals that AI engines actually use.

Rewriting for "AI-friendly" tone without substance. Controlled testing across 9 sites found that rewriting homepage copy to sound more "AI-friendly" produced no measurable improvement in citation rates when done in isolation. Content structure matters, but only after entity clarity and third-party authority are established.

FAQ schema alone. SE Ranking's analysis of 129,000 domains found pages with FAQ schema received fewer citations (3.6) than those without (4.2). Schema is not a shortcut.

llms.txt files. SE Ranking tested 300,000 domains and found zero measurable impact from llms.txt files on citation rates.

Ignoring traditional SEO. 76.1% of AI Overview citations come from pages ranking in Google's top 10. While AI search and Google search surface different pages for some queries, strong traditional SEO remains the foundation. Most AI-cited pages also rank well in organic search.

The scale question: how big is AI search?

It is important to keep perspective. AI search is growing rapidly, but the scale gap with Google remains enormous.

Google drives roughly 37 times more actual traffic to websites than all AI platforms combined. AI platforms account for 0.58% of web traffic versus Google's 21.37%.

The right framing is not "GEO or SEO" but "SEO plus GEO." The strategies that improve AI citation, citing sources, adding statistics, writing clearly, updating regularly, also improve content quality for human readers and for Google. There is no trade-off.

Getting started: a checklist

For businesses that want to start optimising for AI citation without overhauling their entire content strategy:

1. Audit your top 10 pages. Check whether they cite sources, include statistics, and front-load their key claims. If the most important finding is buried in paragraph eight, move it to paragraph one.

2. Add sources to every factual claim. This is the highest-impact single change. Link to the source inline, not in a footnote.

3. Add a "Last Updated" date. Make content freshness visible. Then actually update the content regularly with new data points.

4. Implement Organisation schema. Add sameAs links to your Google Business Profile, LinkedIn company page, and any other authoritative profiles.

5. Build third-party mentions. Reviews on G2, Capterra, Trustpilot, or industry-specific platforms create the corroborated consensus that AI engines look for when deciding which brands to recommend.

6. Track AI referral traffic. Most analytics platforms now identify ChatGPT, Perplexity, and Claude as referral sources. Set up separate tracking so you can measure the impact.

AI SEO: The Complete Australian Guide

A deeper look at AI search optimisation, including platform-specific strategies for ChatGPT, Perplexity, and Google AI Overviews.

Read more

Sources

• Aggarwal, P. et al. GEO: Generative Engine Optimization. KDD 2024 - Foundational GEO research, 9 strategies tested across 10,000 queries
• Ahrefs: Does AI Search Traffic Convert Better? - 23x conversion premium, 0.5% traffic driving 12.1% signups
• SparkToro: Google Search vs ChatGPT Volume - Google processes 210x more searches than ChatGPT
• Search Engine Land: 44% of ChatGPT Citations from First Third of Content - Kevin Indig's analysis of 1.2M search results
• ZipTie.dev: Content Refresh Strategy for AI Citations - 76.4% of cited pages updated within 30 days
• Sill Blog: 12 GEO Tactics Ranked by Scientific Evidence - Evidence-based ranking of GEO strategies
• The GEO Community: Princeton Paper Explained - Detailed breakdown of the original GEO study
• SERPs.io: AI Search Traffic Conversion Data - Multi-study analysis of AI conversion rates
• Presence AI: 2026 GEO Benchmarks Report - AI traffic grew 527% YoY
• The Digital Bloom: AI Citation Position and Revenue Report - 76.1% of AI citations from Google top 10