Adding statistics, quotations and named sources to your content is the one tactic from the 2024 Generative Engine Optimization (GEO) paper that rests on measured results. It also happens to be what good editors have always asked for: claims with evidence, attributed to someone.
What did the GEO research find?
The paper “GEO: Generative Engine Optimization” comes from researchers at Princeton University, IIT Delhi and independent researchers, and was accepted to KDD 2024. The authors built GEO-bench, a set of diverse user queries, and tested nine ways of rewriting a web page to see how visible it becomes in generative engine answers.
What the paper reports:
- Rewrites can boost visibility by up to 40% in generative engine responses.
- The strongest methods added quotations from relevant sources, added statistics in place of vague claims, and cited sources inside the text.
- Keyword stuffing, the old SEO habit, scored below the unchanged baseline (17.8 against 19.5 on the paper’s visibility measure).
- Gains differed by topic, so a tactic that helps a history page does not carry over to every query type.
Two limits matter. The study used a research setup and the Perplexity.ai engine, not Google AI Overviews. And it measured how much of an answer drew on your page, not clicks or sales.
Why does it matter for AI search?
Language models summarise, and a page with concrete facts gives them something specific to quote. A sentence like “many shoppers abandon carts” can be replaced by any other sentence. A sentence with a figure, a date and a named source cannot.
Google’s own guidance points the same way. Its AI optimization guide says AI Overviews and AI Mode run on Google’s normal ranking and quality systems, and it favours content with a first-hand perspective over a summary that restates what is already out there. So the tactic is the same quality signal that E-E-A-T describes, applied sentence by sentence.
Does any special markup help?
No. There is no quotation schema for GEO and no tag that marks a number as a statistic. Google’s guide says you do not need special markup or new machine-readable files to appear in its AI features. Use ordinary HTML:
<p>The 2024 GEO paper found that adding statistics raised visibility in
generative engine answers (<a href="https://arxiv.org/abs/2311.09735">Aggarwal et al.</a>).</p>
<blockquote>
<p>Quoted words go here, exactly as the person said or wrote them.</p>
<footer>Name, role, <cite>Source title</cite></footer>
</blockquote>Structured data stays useful for rich results (see structured data), but it is not a lever for AI citations.
How do you add statistics, quotes and sources, step by step?
- Pick the pages that matter. Start with the 5 to 10 pages that already rank or that answer your customers’ main questions. Rewriting the whole site gains little.
- Mark every vague claim. Read each page and highlight phrases such as “most companies”, “a growing number” and “studies show”. Each one needs a number and a source, or it needs to go.
- Find the original source for each claim. Use the original study, a government statistics office, a standards body or the company’s own data. Skip the blog that quotes a blog that quotes the study.
- Write the statistic with its context. Name the source, the year the data was collected, and what was measured: “The 2024 GEO paper found keyword stuffing scored 17.8 against a 19.5 baseline.” Link the source.
- Add real quotations. Interview a customer, an engineer or an outside expert, or quote published material with permission and attribution. Put the exact words in a blockquote, with the name and role under it.
- Add your own numbers. Your sales data, survey answers or test results are the facts nobody else can cite. The original research method shows how to turn them into a page others link to.
- Cite sources in the sentence, not in a footer dump. A link next to the claim tells both the reader and a crawler what backs it.
- Re-check dates and links every quarter. Figures go stale, and an outdated statistic damages trust more than a missing one.
The free digital marketing module of the Fe/male Switch startup school covers how to gather first-hand customer data you can use here.
How do you measure the result?
Visibility in AI answers is hard to measure, so combine three views:
- The Generative AI performance report in Google Search Console shows how your content performs in AI features in Search and Discover.
- Ask the same 10 to 20 questions in ChatGPT search, Perplexity and Google every month and log whether your page is cited. The AI citation tracking method gives a template.
- Check referral traffic from AI tools in GA4 before and after the edit.
Change one batch of pages at a time and keep an untouched control group, otherwise you cannot tell what caused a shift.
What mistakes should you avoid?
- Inventing or rounding up numbers. Fabricated statistics get repeated by models and traced back to you. Cite only what you can open and read.
- Citing weak sources. A vendor’s marketing page is not evidence for the vendor’s own claim.
- Fake quotes or “expert” names. Google’s spam systems and your readers both punish inauthentic content, and Google’s guide lists inauthentic mentions as something that does not help.
- Keyword stuffing. The GEO paper measured it below baseline, and Google’s spam policies treat it as a violation.
- Treating the 40% as a promise. It is the best result in one study, on one setup.
- Stuffing statistics into thin pages. A paragraph of numbers around no argument reads as padding.
Next steps: read the overview of generative engine optimization, then put the answer to each question first with answer-first content.