2B+
Sundar Pichai said AI Overviews has over 2 billion monthly users across more than 200 countries and territories and 40 languages.
Sample: Company-reported. Period: Q2 2025
Part 6. Content
Specific, sourced and current content is what gets used
The best controlled evidence is still the 2023 GEO paper by researchers at Princeton, IIT Delhi and others, published at KDD 2024. On a benchmark of 10,000 queries, adding quotations, statistics and cited sources raised a page’s visibility in the generated answer the most, by up to 40%. Keyword stuffing scored below doing nothing. It was a lab engine, so treat the size of the effect with care, but the direction matches every later study.
Freshness matters. Ahrefs found AI assistants cite pages that are on average 25.7% newer than the pages ranking in normal results, and SE Ranking found recently updated pages earned more ChatGPT citations. Format evidence is weaker and sometimes contradicts itself. Listicles were the most cited page type in one study, longer and well sectioned articles did better in another, and FAQ sections showed no clear benefit in the raw data even though the study’s summary claimed one.
GEO paper: visibility score by tactic
Position adjusted word count, overall score on GEO bench. Source: Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024.
What it means: Put a real number, a named source or a direct quote where a reader would want proof, and keep the date visible. Do not rebuild pages into FAQ blocks because a summary said so. The raw evidence for that is thin.
Up to 40%
GEO methods can boost a source's visibility in generative engine responses by up to 40%, per the paper's abstract.
Sample: GEO-bench: 10,000 queries from diverse domains and sources
27.2 vs 19.3
Quotation Addition scored highest overall (27.2 vs a 19.3 baseline), followed by Statistics Addition (25.2), Fluency Optimization (24.7) and Cite Sources (24.6); the paper says the best methods improve on baseline by 41% (position-adjusted word count) and 28% (subjective impression).
Sample: GEO-bench, 10,000 queries; lab generative engine
17.7 < 19.3
Keyword stuffing, a traditional SEO tactic, did not perform well and scored below the unoptimized baseline (17.7 vs 19.3 overall).
Sample: GEO-bench, 10,000 queries
Up to 37%
On Perplexity.ai, a real-world generative engine, the authors report visibility improvements of up to 37%.
Sample: Perplexity.ai evaluation
25.7% fresher
AI assistants cited URLs averaging 1,064 days old vs 1,432 days for organic SERP results (25.7% fresher); ChatGPT preferred the newest content, while Google AI Overviews was the only platform citing older content than organic results.
Sample: ~17M cited URLs across 7 platforms
6.0 vs 3.6
Content updated within the past three months averaged 6.0 ChatGPT citations vs 3.6 for outdated content.
Sample: 129,000 domains, 216,524 pages
~900 days
Reddit posts cited by AI search averaged about 900 days old with a median of about 80 words; about 80% had fewer than 20 upvotes, and Q&A threads were more than half of Reddit citations.
Sample: 248,000 cited Reddit URLs
21.9%
Listicles were the most-cited content type (21.9% of citations), followed by articles (16.7%) and product pages (13.7%), together 52% of citations across ChatGPT, AI Mode and Perplexity.
Sample: 75,000 AI answers; 1M+ citations
5.1 vs 3.2
Articles over 2,900 words averaged 5.1 ChatGPT citations vs 3.2 for those under 800 words; sections of 120–180 words averaged 4.6 vs 2.7 for sections under 50 words.
Sample: 216,524 pages
3.8 vs 4.1
In raw averages, pages with FAQ sections averaged 3.8 ChatGPT citations vs 4.1 without, and question-style headings 3.4 vs 4.3; the study's own takeaways claim FAQs help, so the evidence conflicts.
Sample: 216,524 pages