AI is changing the internet

The shopfront of the internet is changing.

In aggregate, how we use the internet looks the same as it has for the last decade or so. Google and traditional search remain the dominant landing point for online traffic, along with social feeds. But at the margins, there is a significant change in behaviour as traffic is shifting toward AI products that summarise, decide and, increasingly, transact in a different way.

This dramatic shift at the margin will soon have a material impact on the overall market, with significant implications for every industry. Three areas where we are already seeing a significant impact are advertising, product and service discovery, and payments & checkout experiences.

This generational shift across advertising, discovery, and payments is exactly why Square Peg invests across the full spectrum - from early-stage companies defining these new experiences to listed companies navigating the transition.

From here, we will talk about a few related topics:

• First, we’ll cover how internet traffic and user behaviour is changing now
• Second, we’ll dive into the early impacts of AI on the traditional marketing funnel
• Lastly, we’ll look at how discovery of products and services with AI creates a host of new problems for merchants to solve

While all of this change is manageable, only the strongest teams with enough talent density will be able to execute consistently in the years ahead.

Setting the scene: The internet we’re used to is not built for AI

The introduction of large language models (LLMs) and the applications they power, such as OpenAI’s ChatGPT, Anthropic’s Claude and Google’s Gemini and AI Overviews, have provided consumers and businesses with an entirely new starting point for engaging with the internet. Why is this such a different experience?

• Queries to LLMs can and tend to be longer and more detailed. They can also evolve into multi-turn conversations, providing the models with a great deal of context.
• This allows user intent to be better understood and addressed directly, rather than with a list of options as we’re used to with Google search.
• This means that it’s more likely a query will be solved within the LLM product rather than requiring a click-out, so referrals are fewer but potentially of much higher value.

Matthew Prince, the CEO and co-founder of Cloudflare (which facilitates ~20% of global internet traffic through its networks), recently shared some fascinating data about how the summarising effect of AI answers is meaningfully changing the nature and volume of traffic to Google. He also shared data showing that LLMs more broadly are limiting the need for users to visit the websites where information is sourced.

Over the past decade or so, Google has consistently delivered one website visit for every two pages it crawls or indexes for search purposes. This was a very stable dynamic as the internet scaled up, allowing businesses to optimise their internet presence for search, pay-for-performance marketing to convert traffic, and advertise across the open web seamlessly.

By the end of 2024, this had shifted materially as Google began to disrupt itself with AI.

Practically, this means that Google is no longer the straightforward traffic funnel it has been for new or established businesses. The culprit is Google’s AI Summaries, which answer the user’s query immediately rather than presenting a series of paid and unpaid links to choose from. We can also see this in Google’s paid clicks growth, which has slowed to just 3% in the first half of 2025.

Looking beyond Google at the newer players where the highest growth in user queries is taking place, the dynamic is far more pronounced. This is because LLMs provide value to users in a totally different way to how search directs traffic.

As more consumers and businesses begin their internet journey on LLMs or even just use AI Summaries within Google, the underlying websites contributing to the answers are receiving very little or no human traffic.

This rapidly scaling channel has material implications for how businesses market to their customers, how products and services get discovered, and also how buying itself occurs in terms of payments and checkout behaviour.

Marketing on the internet is likely to change materially, and soon

The structure of online advertising needs to change for AI. If an AI Overview or LLM answer satisfies a user question, fewer or no clicks reach the open web. As internet behaviour heads in this direction, it completely upends the traditional digital customer journey.

Classic scaling tactics and standard team structures and disciplines will need to change continually - the optimal approach to building a “marketing machine” has yet to emerge in the AI era. Performance marketing budgets will need to be rethought and without the clarity of “pay per click” direct response that Google has delivered for so many years. Intermediation from LLMs could also reduce the strength of more organic, word-of-mouth-driven growth, making it harder to create separate communities or loops. Disciplines and career pathways have been built around all of these approaches to scaling growth and they will all change in some way in the coming years.

Businesses are scrambling to figure this channel out as the LLMs themselves begin to explore different models and how they optimise what appears in answers to queries.

Google is already experimenting with monetisation of its new formats, and we are likely to see similar monetisation efforts from other LLMs. This is an enormous challenge for a scaled incumbent like Google to navigate.

Google’s history with ad products may provide clues as to how the new world will evolve. When the PageRank algorithm became the dominant way to search the internet a tension emerged between those looking to optimise their results (through search engine optimisation or SEO) and Google trying to protect the user experience. When AdWords was introduced in 2000 it became necessary to balance the needs of paid marketers, too.

We are in the very early days of LLMs and AI. Let’s call it the pre-AdWords phase, where optimisation and gaming the system to maximise your chances of being cited in an answer is the priority. It is unclear how long this period will persist. Creativity is being rewarded in these early days, and we've seen the emergence of a variety of GEO (Generative Engine Optimisation) and AIO (AI Optimisation) practices and tools. Practically, GEO/AIO today means:

• Maximising machine readability: Optimising the structure of publicly accessible data so that it can be effectively crawled by LLMs.
• Moving from keywords to questions: Finding the questions for which your product can be an answer, and avoiding purely informational areas, which will be answered by the LLMs themselves without the need for links and citations.
• Increasing breadth and depth of content: Demonstrating expertise across multiple facets of a topic to maximise the chances that LLMs will find a site relevant, especially for more specific "long tail" searches.
• Building reputation on key sites and communities: Maximising appearances on reputable third-party sites that the models “trust” to drive mentions in popular queries.

Here is a practical example comparing Google search results for Square - a point of sale provider that forms part of Block, Inc. (a Square Peg investment) - in the US market compared with frequency of ranked order of appearance in ChatGPT searches when asking for the best options available. Square appears to have been successful in optimising for appearing in LLM queries in the short-term.

Over the next couple of years, these key areas of focus will change as the AI ecosystem becomes more sophisticated and different advertising models emerge. It will demand different capabilities and constant evolution from businesses and teams seeking to use it as a path to market. Talent density and teams with a mindset of experimentation will be critical to navigating this period successfully.

The importance of owning the user traffic

This shift is reinforcing the value of first-party relationships with consumers and businesses. In a world where AI models can reduce a brand to a single line in response to a query, unique content and engaging experiences become significantly more valuable.

Today, these experiences are primarily offered by social networks like Instagram, Facebook, TikTok and Snapchat, along with high-quality subscription or paywalled media content that drives direct traffic. The social networks function as "walled gardens" - ad platforms with their own communities and unique inventory tied to differentiated content, maintaining direct relationships with advertisers and preserving measurable human-led browsing behaviour.

Beyond the major social networks, emerging players with deep engagement can offer advertisers compelling solutions for both awareness and conversion. Reddit - a social media platform structured around 'subreddit' communities for both niche and broad topics - exemplifies how niche platforms can deliver outsized value through concentrated engagement.

The company is in the midst of a journey from higher-level display advertising to more engaging formats, such as its Conversations product, which leverages deep context and community engagement in threads. As it seeks to become an Answer Engine itself, Reddit will need to balance the same dynamic discussed above between encouraging the depth of community discussions and providing immediate utility in answers for users. Highlighting this tension (a good problem to have!), over 70 million WAUs searched on Reddit in the three months to 30 June 2025, with Reddit Answers reaching six million WAU (up 5x compared to the previous three months).

Discovery of new products through AI creates a host of new problems for merchants to solve

LLMs are creating new channels for e-commerce merchants and anyone else selling in competitive online categories. This is an area where Google search and Meta social media funnels have reliably provided predictable and scalable approaches to building awareness, establishing intent, and driving conversions to sales. Most merchants have optimised their selling channels for these approaches, and that won’t work in the same way with AI.

Shoppers are increasingly starting their journey by asking an AI to solve a product discovery problem for them (“I need a waterproof jacket under $300 that can ship to Singapore this week”) rather than visiting a site.

For merchants, this is a new, interesting and challenging channel to manage relative to the many channels that have emerged over time. This is not just from an advertising perspective in terms of reach, but also in terms of brand and how products are described and represented to the end user.

In many ways, this is how Amazon has been operating for more than two decades and has been successful as an “everything store” on the internet. Its catalogue is organised, accessible and executable with minimal friction. How do independent merchants compete and solve this problem without surrendering to a dominant channel and unique ecosystem like Amazon? How do they attract the right talent to figure out what works in an entirely new selling paradigm?

In the same way it has with the emergence of previous selling channels, we believe Shopify (a Square Peg investment) is well-positioned to abstract this complexity for millions of merchants. Shopify is a global e-commerce platform that enables businesses to build and scale online stores, monetising primarily through subscription fees and payment processing.

In May 2025, Shopify introduced Catalog, a global, real-time product catalogue that select partners (such as ChatGPT) and AI agents can query for pricing, options, and availability. This is a structured, tightly managed interface between merchant inventory and AI discovery surfaces.

As it has in the past, Shopify is seeking to standardise the merchant catalogue, then make it easy to publish everywhere—search, social, marketplaces, and now AI products. Similar to GEO/AIO, Shopify helps merchants manage the infrastructure for this new channel, allowing them to stay focused on areas where they add value. Additionally, they’ve released Universal Cart, which enables buyers to transact across multiple merchants simultaneously. Watch this video to learn more about the power of these capabilities, which LLMs and agents can seamlessly interact with.

Whether merchants solve this problem independently or work with an infrastructure provider like Shopify, these tactics will need to continuously evolve as the LLM channel does. This is especially the case as buyers begin to entrust agents with executing their preferences or needs, rather than driving the whole process themselves. This leads us to the change happening at the checkout and payment processing.

Checkout and payment processing are already changing to enable AI-led purchasing

Today's payment processing winners are focused on human-led experiences, leveraging pixels, forms, and fraud signals tied to devices and behaviour, seeking to maximise conversion with minimal friction at the checkout.

In the example above (“I need a waterproof jacket under $300 that can ship to Singapore this week”), you have to click through to the merchant to complete the purchase today. However, we should start to see a more seamless experience emerge soon.

LLM providers OpenAI, Perplexity and others are working on these experiences (including in partnerships with Shopify), where payments could be executed natively.

An even more acute shift will be using an agent to shop for you on the websites of different merchants within certain parameters set by the user (a process referred to as “agentic commerce”).

LLM and agentic commerce create new requirements for payment service providers, requiring a rethinking of fraud patterns and prevention. Some challenges include:

• Managing credentials: Reliably secure exchange of payment credentials at checkout, whether authorised by a human or an agent. This has significant implications for payment methods. Wallets like Shop Pay theoretically bind payment methods more closely to individual identities, allowing for greater freedom in their use. KYC/KYB is one of the most important concepts in cross-border transactions and must meet regulatory expectations.
• Guardrails and rules for delegated purchases to agents: Agents require clear parameters and rules (what an agent can spend, where, and when) to enable dynamic authorisation at checkout, with the ability to trigger a step-up in authentication when needed.
• Real-time context awareness: Traditional human signals are absent when agents transact on behalf of users, requiring scalable infrastructure and new vectors within risk models that operate without human input.

Adyen, Stripe, and Airwallex (all Square Peg investments), along with a new generation of start-ups, should be well-positioned to solve these problems for merchants. While network tokenisation, users’ wallets and the large-scale fraud risk engines of modern players are a great starting point, we will need more innovation as use cases become sophisticated to ensure risk remains manageable and friction remains low in new experiences.

The opportunity for these payment leaders is an exciting one, particularly when considering their talent density relative to incumbents, who still dominate the global payments market share and have been perennial share donors to these newer entrants due to their much higher-quality infrastructure and products.

How will payments incumbents and their teams compete in a world where this level of dynamism is required?

AI is amplifying the need for the best talent

All the changes that are occurring are manageable, but only by the best teams with the right skill sets and mindset.

In some cases, start-ups will have the advantage through being entirely AI-native in their approach to not only building new products, but also enhancing internal productivity. They have no existing audience, channels, or processes to protect. They bring a mindset of doing things in an entirely new way for a new era, and this is too big a shift to ignore.

On the other hand, where incumbents have the right talent and ability to continue attracting, they bring a compelling distribution advantage and existing customer relationships to deploy AI capabilities into.

The need for talent density and a very high bar on quality is common across both. Only then can teams navigate the explosive growth of AI and the changing structure of the internet.