AI Traffic to Recipe Sites

Over the past few years, large language model (LLM) systems have been widely discussed as a potential disruption to traditional search traffic. Conversational AI tools can now answer cooking questions, suggest recipes, and summarise instructions directly within their interfaces – prompting concern among food bloggers that users may no longer need to visit recipe websites at all.

However these LLMs are predictive text machines, they sometimes recommend you use glue as an ingredient in your pizza. The general public is becoming aware of this.

Some questionable service providers are now promoting AI optimization services specifically to food bloggers.

In response, many site owners are questioning whether they should invest time and resources into “optimising for AI” in the same way they optimise for Google search.

Despite the discussion, there is very little publicly available data showing how much traffic LLM systems actually send to publishers in practice. This is particularly true within the recipe niche, where seasonal demand, unique expert content creation skills, and highly structured pages make food blogs a common reference point for AI systems – but not necessarily a destination.

Study Parameters

This study examines observed referral traffic from AI systems to a sample of independent recipe websites using Google Analytics 4 data. The analysis focuses on a 45-day period encompassing both Thanksgiving and Christmas 2025 – two of the highest-traffic events in the food blogging calendar. Providing a stress test of AI-driven referrals during peak consumer demand.

The dataset currently includes 10 recipe websites, representing approximately 11.5 million pageviews over the study window.

Only explicitly identifiable AI referrals (such as those originating from known AI system domains) are included. Direct or “unknown” traffic has been deliberately excluded to avoid speculative attribution, ensuring that all figures represent conservative, verifiable measurements rather than inferred behaviour.

A key requirement for inclusion was that none of the sites were blocking AI crawler bots, so these platforms had full access.

Study Intent

The intent of this study is to describe the current state of play. It does not attempt to model future traffic shifts, make assumptions about how AI systems might behave, or importantly make assumptions about how the general public will use AI systems.

Instead, it asks a simpler and more actionable question for publishers: what is actually happening right now? Specifically, how much traffic do AI systems send to recipe websites today, and how meaningful is that traffic relative to overall site performance?

Early results show that, even during peak seasonal periods, traffic from AI systems accounts for 0.05% or less of total sessions across the sampled sites. This suggests that, at present, LLM platforms are not a material source of traffic and more importantly revenue for recipe website owners – particularly when compared to websites in other verticals or niches where AI referrals have begun to play a measurable role in business performance. ie. tech and local businesses.

There are businesses that absolutely do generate significant revenue from AI systems, but the data shows that recipe websites are not among them.

Furthermore, SEOs have been prodding and testing LLMs and have found that they are relatively easy to manipulate. Think back to Google’s search algorithm from 15 years ago. Which makes sense when you realise how long Google has had to work on search, and how little time Open AI, Perplexity Labs and even Google’s Gemini teams have had to work on their AI systems.

By grounding the discussion in real analytics data rather than speculation, this study aims to provide recipe publishers with a clearer baseline for decision-making: whether AI optimisation is currently a business priority, or whether traditional traffic channels remain overwhelmingly dominant.

Study Comments

• For site 1, even the smallest traditional search engine sends 2–60× more traffic than all AI systems combined.
• For site 2, this site has the highest number of AI pageviews in the dataset, yet Bing alone sends ~18× more traffic.
• For site 3, AI traffic is statistically negligible relative to even secondary search engines.
• For site 4, Even at the upper bound, AI traffic remains under one tenth of one percent.

Study Results