SEO in the Age of Large Language Models

Intro – from keywords to answers

We’ve spent the last two decades obsessing over search engine optimisation (SEO): keywords, backlinks and playing around with Google's algorithm. 2025 is different, though – people are more and more often resorting to LLM Chats (ChatGPT, Perplexity, Claude). That behaviour is completely different and to some extent unknown to us, so we’re still guessing how to optimize for it but we have some ideas – websites need clear, concise structured text. Answer Engine Optimization is the new term for optimizing content to be preferred by LLMs, and as time goes, AEO will only become more impactful.

The rise of LLM chats

LLM (Large Language Model) chatbots exploded from a curiosity to an everyday utility in less than two years. According to data compiled by Similarweb, ChatGPT went from roughly 4 billion monthly visits in mid‑2024 to 5.4 billion visits in June 2025 [0]. The platform now handles over one billion queries every day [0]. Its user base more than doubled in the first half of 2025, reaching ~800 million weekly active users [1]. These numbers rival some of the world’s biggest websites: an April 2025 analysis by SimilarWeb [2] found that while traffic to Google, YouTube and Wikipedia fell year‑on‑year, ChatGPT.com’s visits grew 13% in a single month.

Other players are catching up. Glory Webs estimates ChatGPT still commands roughly 60% of the AI chatbot market, with Microsoft Copilot (~14%), Google Gemini (~13%), Perplexity (~6%) and Claude (~3%) trailing behind [3]. OpenAI’s rivals may chip away at its share, but the overall pool of users interacting with LLM‑powered search continues to swell.

The behavioural shift is most visible in news consumption. A TechCrunch summary of Similarweb’s data shows that ChatGPT referrals to news sites surged from under 1 million visits in early 2024 to more than 25 million by mid‑2025 — a 25x increase [4]. News‑related prompts to ChatGPT grew 212% between January 2024 and May 2025 [5]. Yet there’s a downside: as Google’s AI Overviews rolled out, the share of news searches resulting in zero clicks ballooned from 56% to nearly 69%, cutting organic news traffic from 2.3 billion visits to under 1.7 billion [6]. In other words, more people are asking AI tools for summaries, but far fewer are visiting the underlying websites.

An important nuance: generative AI may be popular, but it still represents a tiny slice of overall search behaviour. SparkToro’s study noted that only 0.1% - 0.25% of total website traffic came from AI search in 2024 [7]. ChatGPT processes about 37.5 million prompts per day versus Google’s 14 billion daily searches, a roughly 373x difference [8]. Surveys show 79% of web users have never tried ChatGPT, and even among those who do, 99.8% still use Google [9]. The LLM phenomenon is growing fast, but it hasn’t replaced traditional search yet.

SEO vs LLM traffic: a numbers game

So what does all this mean for site owners? Several studies put AI referrals into perspective:

Traffic share: Productive Shop tracked 20 SaaS firms and found that AI‑generated sessions accounted for just 0.3% of total traffic in Q4 2024, rising to 0.4% in Q1 2025, while organic search drove more than 35 % of sessions [10]. The same study showed that ChatGPT and Perplexity were responsible for most AI referrals, with Gemini and Copilot trailing far behind [11].
Conversion rates: On average, AI sessions converted at 3.96% versus 4.31% for organic search in late 2024 [12]. By early 2025 the gap narrowed, but organic still edged out LLM traffic [13]. Sessions from AI tools lasted 17% longer, suggesting engaged users; however, the tiny traffic volume means AI referrals still represent a rounding error for most businesses.
Zero‑click searches: Pew Research data (cited in the Similarweb report) suggests AI Overviews and other rich snippets reduce click‑throughs by 20 - 40% on informational queries. When the answer is presented directly in an LLM, the user has little incentive to click through.
Audience demographics: SparkToro reports that AI‑search users skew 70% male and primarily 18 - 34 years old, with adoption far lower among Baby Boomers [14]. From a marketing perspective, LLM traffic may represent a more technical, early‑adopter cohort.

Slicing the data by vertical

The impact of LLM referral traffic isn’t uniform across industries. SALT.agency analysed key event conversion rates (KECVR) across sectors by comparing LLM referrals with organic search for the first quarter of 2025 [15]. Some highlights:

Sector	Avg. LLM KECVR	Avg. Organic KECVR	Notes
Health	13.24%	12.88%	Users searching for health advice favour conversational answers; LLM traffic very slightly outperformed organic [16].
Careers	22.31%	16.58%	Job-seekers appear highly engaged when routed from AI tools [17].
Catalog/Publishing	2.34% / 10.49%	2.13% / 6.57%	For catalog sites and publishers, LLM referrals convert better, perhaps because AI recommendations lead directly to specific articles or products [18].
Real Estate & Property	6.25%	2.71%	Buyers looking for property may appreciate AI-generated summaries and direct links [19].
SaaS	6.69%	6.71%	Parity: LLM referrals convert almost identically to organic traffic [20].
Travel	24.25%	28.97%	Organic search still leads because users want rich descriptions, photos and pricing, though AI can aid early-stage inspiration [21].
Consumer Ecommerce	17.56%	24.12%	LLM traffic lags badly; shoppers need specs, reviews and comparison features [22].
Education	1.89%	3.97%	Students appear to rely on search for authoritative sources [23].

These figures reveal that certain verticals benefit disproportionately from AI referrals. Career, publishing and real‑estate sites see LLM traffic convert 1.2x - 3x better than organic search [24]. Travel, ecommerce and education remain dominated by organic search, suggesting people still want to explore options rather than accept a single recommended answer. SaaS lies in the middle: AI sessions drive similar conversion rates to organic [25], hinting that product‑comparison queries (e.g., “What are the best CI/CD platforms?”) can be answered fairly well by an LLM.

Deep dive: Real estate & publishing – small volume, big impact

To illustrate how LLM traffic behaves, let’s examine the Real Estate and Publishing categories. In SALT.agency’s dataset, real‑estate sites had a 6.25% conversion rate from LLM referrals versus 2.71% from organic search [26]. That’s more than double the effectiveness — albeit from a tiny traffic base. Why? Home buyers often have a strong intent and specific questions: “Two‑bedroom apartment in Vilnius with balcony under €200k?” A chatbot can triage the query, summarise listings and deliver a handful of relevant links. When someone clicks through, they’re already deep in the funnel.

Similarly, the Publishing sector shows 10.49% LLM conversion vs. 6.57% organic [27]. AI chatbots summarise articles or recommend niche blogs; a user who clicks is more likely to subscribe or read further. However, both sectors still receive most of their volume from organic search. As Productive Shop’s study shows, AI sessions represent less than 0.5% of total traffic [28]. Even a high conversion rate won’t pay your bills if only a handful of people arrive.

How to raise your “score” in an LLM’s chain of thought

At this point you might wonder: can I game ChatGPT like I gamed Google? Thankfully (or unfortunately, if you enjoy keyword‑stuffing), the answer is no. While exact ranking algorithms are proprietary, analysis of AI answer behaviour and guidelines from market research firms provide some clues:

Quality over quantity. LLMs prefer well‑researched, factually correct content with clear context. Articles written solely for SEO – stuffed with synonyms and fluff – often get summarised poorly, reducing the likelihood of being cited in an answer. Focus on originality and real insight.
Structure matters. Use question‑style headings, concise paragraphs and bullet lists so AI models can extract relevant information easily [29]. Incorporate schema markup to identify FAQs, how‑to steps, product attributes, ratings and other structured data. Google’s AI overviews and ChatGPT’s browsing tool both rely heavily on structured signals.
Demonstrate expertise and trust. Google’s E‑E‑A‑T (Experience, Expertise, Authoritativeness, Trustworthiness) guidelines still apply. Cite reputable sources, provide author bios and disclose affiliations. Verified authorship increases the chance of being referenced by AI answer engines.
Write for humans, not robots. Ironically, LLMs favour a natural, engaging writing style -- the same type of prose your LinkedIn followers enjoy. Resist the temptation to produce formulaic “SEO‑friendly” copy. Use humour or storytelling when appropriate; answer engines recognise user‑experience signals such as dwell time and low bounce rates.
Diversify your distribution. Because AI referral volume is still tiny, continue investing in organic search, social media, newsletters and other channels. LLM traffic may grow, but it’s a complement, not a replacement.
Focus on revenue, not vanity metrics. Traditional SEO metrics (impressions, clicks) become less meaningful when AI answers reduce click‑through rates. Track engagement, conversions and customer lifetime value. Even if AI referrals deliver fewer visits, the users they bring may be ready to buy.

Conclusion – don’t abandon SEO yet

The hype around LLMs is real, but the numbers show a more nuanced picture. ChatGPT and friends are attracting billions of visits and processing billions of queries, yet they still account for a fraction of global search traffic [30]. AI answer engines are nibbling away at informational search queries, especially in sectors such as news, careers and real estate, but organic search remains the workhorse for ecommerce, travel and education [31].

For businesses and content creators, the takeaway is simple: don’t jump blindly. Keep your SEO fundamentals strong while experimenting with AEO tactics. Structure your content so that LLMs can understand and summarise it. Write with authenticity (bonus points if you can slip in a dad joke or two). Diversify your traffic sources and measure success by business outcomes rather than page‑view vanity metrics.

The age of large language models isn’t the end of SEO -- it’s the beginning of a more human‑centric, multi‑modal search landscape. Get your structured data and storytelling hats on; the answer engines are listening.