Opening: A 2023 Paper Becomes 2026's Marketing Arsenal
In July 2026, as another wave of GEO (Generative Engine Optimization) data landed, a 2023 Princeton paper was pushed back into the spotlight: GEO: Generative Engine Optimization (Aggarwal et al., published at KDD 2024). It matters because it delivers what everyone has been missing: reproducible, measurable experimental results on how to get cited by AI — not mysticism, but numbers run across ~10,000 queries and nine datasets.
Over the past 18 months, "GEO" went from nonexistent to a paid SaaS package, but most "guides" are just old SEO talking points reskinned. The market backdrop: the AI answer layer (ChatGPT, Google AI Overviews, Perplexity) has swallowed a large share of search. Latest figures show ChatGPT commands ~92.4% of standalone AI referral traffic, while Google's AI summaries cite three or more sources 88% of the time and only 1% cite a single source. In other words, AI answers aren't winner-take-all — they're a multi-source platter, which is exactly the gap small sites can slip into.
The same research surfaced a brutal fact: only ~37.9% of URLs cited in AI Overviews also rank in the organic top ten. Chasing traditional rankings and getting cited by AI are now two different games. Plenty of GEO tools (Profound, Scrunch, Evertune) sell "are you being cited" monitoring — but monitoring doesn't change the outcome.
The paper's value is that it directly tested "change what, and citation rate goes up." Below I unpack the three most effective moves, a no-SaaS DIY path for SMBs, and a contrarian call.
Event Details: Three Moves, 30–40% Relative Lift
The paper used a framework called GEO-bench, simulating a two-stage pipeline ("Google retrieves top-5 sources → an LLM synthesizes a cited answer"), and tested how a piece of content's visibility (Position-Adjusted Word Count) in the AI answer changed after adding different elements. The conclusion is tight: overall optimization can lift visibility ~22%–41%, and three moves contribute most, each ~30–40% relative improvement:
- Cite Sources: cite authoritative primary sources inline with links.
- Statistics Addition: replace "many people" with "~92.4% of people" — concrete numbers over adjectives.
- Quotation Addition: quote experts or source documents verbatim rather than paraphrasing everything.
Notably, the paper also tested "keyword density" and "buzzword stuffing" — near-useless for AI citation, sometimes counterproductive.
Immediate Actions for Three Reader Types
- Brand owners / SMB bosses: You don't need a monitoring subscription. Do one thing first — take your 5 most important pages and add "at least 2 linked authoritative sources + at least 1 concrete statistic + 1 verbatim expert quote" to each. Those three are the paper's validated levers.
- Marketers / SEOs: Add "citability" to your content QA checklist instead of only tracking keyword rank. Before publishing, ask: if AI wants to cite this, can it grab the number, the source, the quote?
- Developers / agencies: Turn those three elements into CMS template fields (source link, data, quote) so non-technical editors can fill them; then reinforce with JSON-LD
citationandFAQPage.
SaaS Tool Comparison
| Tool | Focus | Price tier | Best for |
|---|---|---|---|
| Profound | Cross-engine citation monitoring | Enterprise (high) | Mid-large brands with budget |
| Scrunch AI | AI visibility tracking + suggestions | Mid-high | Marketing teams |
| Evertune | Brand mention analysis in models | Mid-high | Brand / PR teams |
| DIY (self-built) | Manual spot-check + content spec | Near zero | SMBs / agencies |
What They Won't Tell You
- Monitoring ≠ improvement: most GEO SaaS sells a "are you cited" dashboard, but what actually lifts citations is the content's three elements — tools can't do that part for you.
- 40% is relative, not a guarantee: it's a relative lift under a controlled experiment and a specific metric, not "traffic up 40% tomorrow." The paper itself has noted methodological limits (GPT-3.5, simulated pipeline).
- It can be countered: when everyone "adds stats and quotes," engines will eventually discount deliberate stuffing. The window won't stay open forever.
The No-SaaS SMB Alternative
You can build a GEO checking loop yourself at near-zero cost:
- Use a Google Sheet to list the 20 questions you most want to be cited on (what your prospects would ask AI).
- Weekly, manually run those questions through ChatGPT/Perplexity and log "did it cite you, and which passage."
- Against the paper's three elements, add source links, statistics, and quotes to content that wasn't cited.
- Embed
FAQPageandcitationJSON-LD so engines parse you more easily.
FAQ
Are GEO and SEO the same thing?
No. Research shows only ~37.9% of URLs cited in AI Overviews also rank in the organic top ten, meaning "chasing rank" and "being cited by AI" have decoupled. SEO is about being findable; GEO is about being quotable by AI.
Do I have to buy a GEO monitoring tool?
Not necessarily. Monitoring tells you "were you cited," but what actually raises citations is the three content elements (sources, statistics, quotes). SMBs can use a spreadsheet plus a weekly manual spot-check and get the content right first.
Which of the three moves should I do first?
Start with adding concrete statistics — easiest to implement and easiest for AI to grab; then add linked authoritative sources; finally verbatim expert quotes. Stacking all three works best.
How long until I see results?
No guaranteed timeline. AI citations shift as models and indexes update, so track on a "20 questions weekly" cadence rather than expecting one article to move the needle instantly.
My Take
The mainstream narrative is "GEO is the new battlefield, buy tools to stake your claim." My call is the opposite: the GEO tool market will consolidate within 18 months because its "monitoring" moat is too shallow — anyone can build monitoring; what changes outcomes is a content methodology written in a free 2023 paper. The service opportunity for ScriptWalker is clear: instead of selling "GEO monitoring subscriptions," package those three validated moves into a one-off "content structuring + JSON-LD implementation" deliverable — reformatting existing content to be AI-citation-friendly. That's a one-off, deliverable, quotable, paper-backed service, far better suited to an agency than an open-ended subscription.
Sources
- Princeton — GEO: Generative Engine Optimization (paper page) (first-hand)
- arXiv 2311.09735 — GEO paper full text (first-hand)
- AEO Vision — Google AI Overviews GEO Statistics 2026 (first-hand data)
- Aristral — SEO and GEO/AEO Statistics 2026 (third-party)
- Blck Alpaca — The Princeton GEO Study: Methodology and Critique (third-party critique)