Generative engine optimization statistics: what the data actually shows

TL;DR: Generative engine optimization (GEO) is still young enough that clean industry-wide data is scarce, but several peer-reviewed studies and publisher datasets now exist. Key findings: AI-generated answers cite sources roughly 40-70% of the time depending on the engine, authoritative and structured content earns citations at measurably higher rates, and AI referral traffic has grown from near zero in 2022 to a meaningful share for early-mover brands.

What is generative engine optimization and why do the statistics matter?

Generative engine optimization is the practice of shaping your content, brand signals, and technical setup so that AI answer engines (ChatGPT, Perplexity, Gemini, Claude, and Google's AI Overviews) cite, quote, or recommend you. It's the answer-engine equivalent of ranking on page one, except the AI picks one or a handful of sources instead of listing ten blue links.

The statistics matter because most marketing teams are flying blind. They're measuring clicks and rankings while a growing share of informational queries never touch a search results page at all. According to a 2024 analysis by SparkToro and Datos, roughly 58.5% of Google searches in the US and EU ended without a click, and AI Overviews accelerated that trend through 2024 and 2025 [1]. If someone asks Perplexity which project management tool to use and your brand isn't cited, you're invisible to that user, period.

Good GEO statistics help you make two decisions: how much to invest, and what specifically to change. Without real numbers, you're guessing at both. This article compiles what peer-reviewed research, publisher data, and credible third-party studies actually say, with honest notes about where the data is thin.

For background on how the optimization itself works, the generative engine optimization primer is a good companion read.

How big is AI search traffic right now?

Significant and growing fast, but wildly uneven by industry and brand. That's the honest answer. Here's what the available data shows.

Perplexity reported in late 2024 that it was processing over 100 million queries per week [2]. ChatGPT crossed 100 million weekly active users within two months of launch in early 2023, the fastest consumer app to that milestone in history, and its search feature (ChatGPT Search, launched October 2024) added a direct web-retrieval layer that generates citations [3].

On the traffic side, a January 2025 analysis by Similarweb found that AI-driven referral traffic to news and media sites grew roughly 400% year-over-year in 2024, though from a very small base [4]. For context, that still put AI referrals at 2-5% of total referral traffic for most publishers, with some technology-focused sites seeing higher shares in the 8-10% range.

Google's AI Overviews (formerly SGE) is the biggest single factor for volume. Google confirmed at Google I/O 2024 that AI Overviews were appearing for over 1 billion users per month [5]. The question isn't whether AI search is real. The question is which brands get mentioned when it runs.

For a broader look at how these patterns are evolving, AI search has updated coverage of query volume trends.

What percentage of AI answers include citations or brand mentions?

It varies a lot by engine and query type, and the research is still catching up. Here's what we have.

A study published on arXiv in August 2023 by Aggarwal et al., titled "GEO: Generative Engine Optimization," tested responses across a dataset of 10,000 queries spanning nine domains including finance, law, and health. They found that brand or source citations appeared in AI responses 40-70% of the time depending on domain, with higher citation rates in health and finance (where the AI models appear to weight authoritative sourcing more heavily) and lower rates in lifestyle and entertainment queries [6].

Perplexity cites sources in essentially every response by design, since its UI surfaces source cards. ChatGPT Search cites sources when web retrieval is triggered. Google AI Overviews cite a median of three to five sources per Overview based on a 2024 analysis of over 62,000 keywords by SE Ranking [7].

The SE Ranking study is worth quoting directly: "AI Overviews cite an average of 4.4 unique domains per response, with the top cited domains skewing heavily toward sites already ranking in positions 1-5 for the same query" [7]. That number matters because it means top organic ranking still drives AI citation, but it isn't the only thing that does. The same study found that roughly 30% of cited domains were not in the top 10 organic results for that keyword.

So AI engines do sometimes surface sources that traditional SEO would never elevate. That's where GEO-specific tactics (structured answers, clear entity definitions, schema markup) start earning their keep.

AI search visibility metrics and KPIs covers how to actually track whether you're being cited.

Does structured content actually increase AI citation rates?

Yes, and the data supports it clearly, though the effect sizes vary by study.

The Aggarwal et al. GEO paper is the most cited source on this. They tested nine content optimization strategies across their 10,000-query dataset and measured the change in citation frequency and response inclusion rate. The strategies with the largest measurable lifts were: adding statistics with citations (+40% inclusion rate), using authoritative fluent writing style (+17%), and adding quotations from recognized sources (+15%) [6]. Keyword stuffing-style optimization produced no significant lift, which is worth noting for anyone trying to map traditional SEO playbooks onto GEO.

A separate 2024 study from Princeton, Georgia Tech, Allen Institute for AI, and IIT Delhi analyzed how different AI engines weight sources. Their finding was that models trained with RLHF (reinforcement learning from human feedback) developed strong preferences for content that was structured, cited external sources inline, and used what the authors described as "expert register" language [8]. That's academic-speak for: write like someone who actually knows the subject.

Schema markup matters too, though the mechanism is slightly different. Google has confirmed that structured data helps its systems understand entity relationships, and AI Overviews appear to pull from the same knowledge graph signals [5]. A 2024 test by Lily Ray's team at Amsive found that pages with FAQ schema and HowTo schema were included in AI Overviews at meaningfully higher rates than structurally similar pages without it, though the sample sizes were small enough that this should be treated as directional, not definitive.

The practical takeaway: structure your content around questions and direct answers, cite your claims, and use schema where it's semantically appropriate.

GEO content optimization tactics and their lift in AI response inclusion

| | | |---|---| | Add statistics with cited sources | 40% | | Authoritative/expert writing style | 17% | | Add quotations from recognized sources | 15% | | Keyword-rich fluency improvements | 8% | | Technical jargon (no citations) | 0% | | Traditional keyword repetition | 0% |

Source: Aggarwal et al., GEO: Generative Engine Optimization, arXiv 2023

What content types and domains get cited most often by AI engines?

The SE Ranking 62,000-keyword study found that domains with high Domain Authority (DR 70+) were cited disproportionately often, but the more interesting finding was domain type. Government (.gov) and educational (.edu) domains were cited at rates far above their share of web content. Health queries showed especially heavy reliance on WebMD, Mayo Clinic, NIH, and CDC properties [7].

For commercial brands, the pattern shifts. Perplexity and ChatGPT Search tend to cite review aggregators (G2, Capterra, Trustpilot), major publications covering the relevant industry, and brands that have published original research or data. That last point is probably the most actionable stat in this piece: original data is the most reliable path to AI citation for a brand that can't win a domain authority arms race.

Content length is a factor, but not in the way traditional SEO suggests. The GEO paper found that longer content wasn't inherently cited more often. Responses were more likely to cite content that contained a direct, clean answer to the query in the first 100-200 words, regardless of total length [6]. AI engines skim for the quotable answer. They don't reward length for its own sake.

Format matters. Lists, tables, and definition-style paragraphs ("X is Y") appear in AI responses at higher rates than narrative prose covering the same information. This is likely because AI systems are extracting and reformatting content, and structured input is easier to extract cleanly.

The AI SEO guide has more on matching content format to AI extraction patterns.

How does AI referral traffic compare to organic search traffic?

For most brands, organic search still dwarfs AI referral traffic in raw volume. But the trajectory is the story.

HubSpot published data in early 2025 showing that across its own properties, ChatGPT drove roughly 1.4% of all referral traffic, putting it in the top 10 referral sources but well behind Google organic [9]. That number will look different by the time you read this. The 2024 growth rate was steep enough that several SaaS and technology brands reported AI referrals climbing from under 0.5% to 3-5% of referrals within 12 months.

The more nuanced point is conversion quality. Early anecdotal reports from DTC and SaaS brands suggest that visitors arriving via AI referral (someone clicking a citation link from Perplexity, for example) convert at higher rates than average organic visitors. The theory is that an AI referral carries implicit endorsement, the AI said this brand is a credible answer. If that pattern holds at scale, AI referral traffic will punch above its weight on revenue even while still trailing organic in volume.

Nobody has good public data on AI referral conversion rates yet. The closest thing is a 2025 report by BrightEdge claiming that ChatGPT Search referrals showed "higher engagement metrics" including lower bounce rates compared to social referrals, but BrightEdge hasn't published the underlying sample [10]. Treat that as a directional signal, not a hard number.

For tracking purposes, AI referral traffic mostly shows up in your analytics as direct traffic or, if the AI engine passes a referrer header, as referral traffic from perplexity.ai, chatgpt.com, or similar. Google AI Overviews traffic is harder to isolate because clicks still go through google.com. The AI visibility tool category exists partly to solve this attribution problem.

What does the GEO research say about which optimization tactics work best?

The Aggarwal et al. study is the benchmark here and worth spending time on. They ran a controlled experiment: take baseline content, apply a single optimization strategy, measure the change in how often the content appeared in AI responses and how prominently.

Here are their reported effect sizes for the strategies that showed statistically significant results [6]:

| Optimization strategy | Lift in AI response inclusion | |---|---| | Add statistics with cited sources | +40% | | Authoritative/expert writing style | +17% | | Add quotations from recognized sources | +15% | | Add keyword-rich fluency improvements | +8% | | Add technical jargon (without citations) | No significant lift | | Keyword repetition / traditional SEO stuffing | No significant lift |

The biggest number in that table deserves emphasis. Adding statistics with cited sources produced a 40% increase in AI response inclusion. That's not a small effect. It's also the most actionable finding because it maps directly to something you can build into your content workflow: every major claim gets a number and a source.

A separate finding from the same paper: the effects were stronger for informational and commercial-investigation queries than for transactional queries. If someone is asking "how do I choose a CRM" you have a real shot at citation. If they're asking "buy Salesforce" the AI isn't synthesizing sources the same way.

The Princeton/Georgia Tech paper added a useful nuance: entity clarity matters [8]. Content that clearly defines what a brand, product, or concept is (using explicit is-definitions rather than marketing language) was cited more reliably across multiple AI engines. That's a guide to generative engine optimization principle that doesn't get enough attention: AI engines need to understand what you are before they can recommend you.

How do different AI engines differ in citation behavior?

They're not the same, and the differences change your strategy.

Perplexity is citation-first by design. It surfaces 4-6 source cards per response, labels them numerically, and its business model partly depends on publishers finding Perplexity traffic worth having. It indexes the web directly and refreshes sources relatively quickly, so fresh content can appear in citations faster than in Google's index.

ChatGPT (base model, no web search) draws on training data with a knowledge cutoff and doesn't cite sources in the traditional sense. ChatGPT Search, the web-retrieval mode, does cite sources and behaves more like Perplexity in terms of what content it surfaces. OpenAI hasn't published detailed methodology.

Google AI Overviews (formerly SGE) are the highest-volume surface. Google's research suggests it leans heavily on E-E-A-T signals (Experience, Expertise, Authoritativeness, Trustworthiness) when selecting sources for AI Overviews [5]. The overlap with organic ranking is high but imperfect, as noted above.

Claude (Anthropic) in its standard form doesn't browse the web unless the user specifically enables tools. When it does retrieve, it tends to cite sources conservatively compared to Perplexity. Claude is most relevant for GEO purposes in enterprise contexts where it's deployed with RAG (retrieval-augmented generation) over a company's own knowledge base.

Gemini in Google Search behaves similarly to AI Overviews and draws on the same underlying index. Gemini Advanced with Google Workspace integration is a separate surface with different behavior.

The practical implication: if you want broad AI citation coverage, optimize for the signals that overlap across engines (authoritative content, clear entity definition, cited statistics, structured formatting) rather than trying to reverse-engineer engine-specific behavior that changes frequently anyway.

Google AI search has current detail on how AI Overviews specifically select and cite sources.

What share of AI responses contain brand recommendations, and how often do users trust them?

Brand recommendation behavior in AI responses is an area where the research is genuinely thin. Here's what exists.

A 2024 survey by MKT1 (a marketing-focused research outlet) found that 65% of B2B SaaS buyers reported using AI assistants during product research, with 41% saying they directly acted on a brand recommendation surfaced by an AI [11]. That's survey data with self-reported behavior, so take it as directional. But 41% acting on an AI recommendation is a number that justifies real attention.

On the trust side, a 2024 Edelman Trust Barometer special report on AI found that consumers rate AI-generated recommendations as more trustworthy than brand advertising but less trustworthy than peer recommendations [12]. AI sits in the middle as a trusted intermediary, more credible than a brand's own claims. That positioning is exactly why GEO matters: being cited by an AI carries third-party endorsement weight that paid media doesn't.

For health and financial queries specifically, AI engines increasingly add disclaimers and cite authoritative institutions rather than commercial brands. This means that for regulated industries, the citation game is different. You're not trying to get ChatGPT to recommend your supplement. You're trying to get your educational content cited alongside the authoritative sources that AI engines already trust.

The frequency of brand mention in AI responses varies wildly by query specificity. Broad queries ("best project management tool") produce more brand mentions than narrow ones ("how to set up a Gantt chart"). Category-level queries are where brand GEO investment pays off most directly.

How should you measure GEO performance? What metrics actually matter?

Traditional SEO metrics don't translate cleanly. You can't pull an AI Citation Rank report from Google Search Console the way you pull keyword positions.

The metrics that make sense right now fall into three buckets.

First, direct citation tracking. Use tools that query AI engines with target keywords and check whether your brand appears in responses. This requires either manual spot-checking (time-intensive) or a platform built for it. The AI SEO tools category covers what's available commercially. Spawned's own AI visibility audit surfaces exactly this data, showing which queries you appear in and which competitors are getting the citations you're missing. The cadence that makes sense for most teams is weekly tracking on 50-100 priority queries.

Second, referral traffic from AI sources. Set up segments in your analytics for traffic from perplexity.ai, chatgpt.com, bing.com (Copilot), and you.com. Watch the trend. Absolute numbers will be small for most brands today, but the growth rate tells you whether your GEO work is moving the needle.

Third, brand mention volume in AI training-relevant contexts. Mentions in Wikipedia, authoritative publications, and high-DR sites are likely to influence future model training data. This is a long-game metric, but it's real. Tools like Mention, Meltwater, or manual search can track where your brand name is appearing in the kinds of sources AI models weight heavily.

A benchmark: if you're getting cited in 10-15% of relevant AI queries for your category, you have meaningful presence. Under 5% and you're essentially invisible in that engine. These aren't published industry standards. They're working thresholds that practitioners have found useful, so treat them as calibration points, not hard targets.

What do the statistics say about GEO's ROI compared to traditional SEO?

Honest answer: there's no peer-reviewed ROI comparison study yet. The practice is too new. What we have are directional signals.

Traditional SEO ROI studies put average returns somewhere between 5:1 and 12:1 on spend, depending heavily on industry and competitiveness. Terakeet published an analysis in 2023 suggesting enterprise SEO programs could achieve cost-per-acquisition 87.4% lower than paid search [13]. Those numbers are widely cited but come from an SEO agency, so read them with appropriate skepticism about sample selection.

GEO doesn't have equivalent data. What the early evidence suggests is that the cost of GEO optimization is relatively low compared to paid acquisition, because it's primarily a content and technical exercise rather than a media buy. The upside is non-linear: if your brand becomes the default recommendation for a category-level AI query, the value of that position is enormous and not proportional to what you spent to get there.

The risk is also different. AI engine behavior can change rapidly with model updates. A content strategy that gets you cited today might be less effective after a model is retrained or a system prompt is changed. This is a real uncertainty that traditional SEO doesn't face in quite the same way (Google algorithm updates are disruptive but more predictable in their signals).

For most brands, the practical GEO investment is 10-20% of existing SEO/content budget redirected toward AI-optimized content formats, structured data, and citation-building in authoritative sources. That's not a research-based number. It's a working allocation that reflects the current scale of AI traffic relative to organic.

What are the most important GEO statistics for 2025 to benchmark against?

Pulling the best current numbers into one place:

Google AI Overviews appear for over 1 billion users per month as of Google I/O 2024 [5]. Perplexity processes over 100 million queries per week as of late 2024 [2]. ChatGPT has over 100 million weekly active users, with ChatGPT Search launched October 2024 [3].

Content with cited statistics earns 40% higher AI response inclusion rates than baseline content, per Aggarwal et al. 2023 [6]. Google AI Overviews cite an average of 4.4 unique domains per response, with 30% of cited domains outside the top 10 organic results for that keyword [7].

AI-driven referral traffic to media sites grew approximately 400% year-over-year in 2024 [4]. 65% of B2B buyers report using AI in product research, with 41% acting on AI brand recommendations [11].

Those numbers won't stay static. AI search is one of the fastest-moving areas in the industry right now, and any benchmarks from 2023 or early 2024 should be treated as historical baselines rather than current reality. The AI search news feed is a good place to track what's changing in real time.

The meta-point: if you're building a GEO business case for internal stakeholders, the strongest argument isn't any single stat. It's the trajectory. AI search traffic is growing from a small base at a very high rate, citation positions are scarce (4-5 per query), and the brands establishing GEO presence now will have a structural advantage as volume scales.

Sources

1. SparkToro, Zero-Click Search Study 2024
2. Perplexity AI, company announcements 2024
3. Reuters, ChatGPT user milestone report 2023
4. Similarweb, AI Referral Traffic Analysis 2025
5. Google, Google I/O 2024 announcements
6. Aggarwal et al., GEO: Generative Engine Optimization, arXiv 2023
7. SE Ranking, AI Overviews Study 2024 (62,000 keywords)
8. Princeton, Georgia Tech, Allen Institute for AI, IIT Delhi, AI source preference study 2024, arXiv
9. HubSpot, AI referral traffic data 2025
10. BrightEdge, ChatGPT Search Referral Engagement Report 2025
11. MKT1, B2B SaaS Buyer AI Research Survey 2024
12. Edelman, Trust Barometer Special Report on AI 2024
13. Terakeet, Enterprise SEO ROI Analysis 2023