Why Retailers’ Landing Pages Tanked Conversions for AI’s Highest-Intent Shoppers
The Analysis: £8M Revenue, 170,000 Orders, 4 UK Retailers (Nov 25 – Dec 2)
We conducted a deep analysis of AI-driven shopping behaviour across four UK retailers, leveraging £8 million in revenue and over 170,000 genuine customer orders from the Black Friday period. Our data incorporates live traffic from leading LLMs, including ChatGPT, Perplexity, Gemini, and Copilot.
This is the first real-time data set, tracking real customer sessions through the year’s biggest retail moment.
What emerged was an unexpectedly consistent pattern: AI assistants are quietly establishing a new layer in the buying journey, influencing purchasing decisions long before the customer even reaches the retailer’s site.
1. AI Has Become a Layer Above the Funnel
Across all four retailers, customers behaved as if their chosen AI assistant—not the retailer’s website—was the starting point for their decision-making.
They leveraged assistants to narrow choices, compare models, clarify delivery options, interpret specifications, and eliminate unsuitable products. Essentially, the critical thinking work happened entirely upstream.
This resulted in a distinctive and surprising performance pattern: AI-sourced sessions spent 2–3 times longer on site than organic search yet converted 25–40% lower than expected, given the extended dwell time.
At Harton Works, we contend this is not a problem of low intent, but of fundamental misalignment. Retailers were receiving decision-ready visitors on pages designed only for decision-forming visitors.
2. Each AI Assistant Generates a Distinct and Repeatable Behavioural Signature
A key surprising finding is that across all four retailers and hundreds of thousands of sessions during Black Friday weekend, each AI platform generated a stable and repeatable behavioural pattern.
Crucially, this is not based on individual user identity. While we cannot track whether the same person uses multiple tools, we observed that traffic arriving from each assistant behaved consistently, regardless of the retailer, category, or audience.
ChatGPT: Deep, Exploratory Behaviour
● Longest session durations across every retailer
● High engagement and multiple product interactions
● Moderate conversion despite strong on-site depth
● Indicates “talk-it-through” research before committing
Perplexity: Fast, Factual, and Shallow
● Sub-one-minute sessions almost universally
● Minimal engagement and page interaction
● High intent in queries but low on-site behaviour
● Suggests single-detail checks rather than exploration
Gemini: Reflects Ecosystem-Default Usage
● Strong mobile bias across every retailer
● Lower session time and engagement
● Patterns consistent with Android and Chrome integration
● Behaviour suggests casual discovery rather than active shopping
Copilot: Aligns with Weekday, Desktop-Led Research
● Sessions peak Monday to Friday
● Significant drop on weekends
● Strong desktop skew
● Consistent with workplace-integrated AI being used during task-driven research
Across four completely different brands, the same pattern held consistent: each assistant generated the identical behavioural signature wherever it appeared.
This consistency is one of the strongest signals from the dataset. It proves that AI-driven traffic is not random, not chaotic, and not evenly distributed. Instead, it arrives with recognisable shapes, rhythms, and intents, and those shapes are beginning to matter.
3. ChatGPT Has Become the Deepest Research Environment in Retail
ChatGPT consistently produced both the largest volume of AI sessions and the deepest on-site engagement. In specific instances, we recorded session durations of:
● 11 minutes (Average Session Time)
● 14 minutes
● Even 21 minutes
For comparison, organic search sessions averaged only 5–7 minutes.
These extended session lengths reflect a significant shift in where decision-making is executed. Customers leveraged ChatGPT to:
● Compare models
● Clarify specifications (specs)
● Ask follow-up questions
● Test different scenarios
● Explore ‘best-for-me’ options
● Resolve delivery or returns concerns
This used to happen on the website. It now happens before the customer arrives.
When they do arrive, they behave like visitors who are already halfway through the purchasing journey, because they are.
4. Perplexity: The Final Certainty Check Assistant
Perplexity’s on-site behaviour was consistently short, direct, and surgical:
● Brief sessions
● Shallow engagement
● Single-detail intent
Users arrived to confirm one specific detail—a warranty, a version number, a feature, or an availability promise—and they left immediately once that information was confirmed or denied.
Because Perplexity is positioned as an AI search engine, we find that customers using it prioritise fast, conclusive checks on their decision. In contrast to the in-depth dialogue seen with ChatGPT, they seek to get straight to the heart of the matter.
Therefore, the retailer’s site has a critical, instant mandate: it must immediately either confirm or contradict that certainty.
5. Gemini: Behaviour Driven by Device Defaults, Not Active Intent
While Gemini’s traffic footprint was smaller (less traceable?), the behaviour observed was consistent:
● Mobile-led
● Top-funnel exploration
● Lightly engaged sessions
● Driven by AI Overviews
Users did not actively choose Gemini to shop; they encountered it because it is natively integrated into the Google ecosystem they already use.
The resulting on-site behaviour reflects exploration, not evaluation. However, as AI Overviews expand, Gemini is uniquely positioned to shape initial awareness long before a customer reaches a stage of high intent.
6. Copilot Appears in Weekday, Work-Context Research
Copilot traffic consistently exhibited these signatures:
● Weekday Spikes: Sessions peak Monday to Friday
● Desktop Dominance: A strong desktop skew, with a significant drop on weekends
● Mid-Intent Research: Focused on functional tasks and information validation
● Low Conversion Rates
This is not emotional retail behaviour.
Instead, it is highly task-driven: comparing specifications (specs), checking compatibility, and validating information during a typical workstation session.
As Microsoft deepens Copilot’s integration into Windows, this task-driven cognitive pattern is poised to become highly significant in categories such as electronics, appliances, DIY, and office products.
7. The Landing-Page Mismatch: AI-Informed Users vs SEO-Era Pages
This was the clearest and most costly pattern observed.
AI-informed visitors arrived with a distinct profile, including:
● Detailed expectations
● Specific decision criteria
● An assistant-generated mental model
● Pre-filtered preferences
● Low tolerance for redundant information
However, the landing pages they hit were structurally not built for this type of visitor.
What AI-Informed Visitors Needed:
Fast confirmation of assistant-provided data
Attribute-level, comparative detail
Simple, direct model comparisons
Compatibility and ‘best-for’/’not-for’ guidance
Clear, front-and-centre delivery and returns policies
What They Actually Encountered (The SEO-Era Page):
Lengthy introductions and generic copy
Vague benefit statements and marketing language
Specifications hidden behind tabs or incomplete data
One-size-fits-all product descriptions
Unclear, inconsistent, or buried answers
The outcome was predictable. High scroll depth and long dwell time were recorded, but this resulted in conversion underperformance. This did not occur because the traffic was unqualified, but because the page format no longer matched the user’s journey or intent.
8. The Mandate for 2025: Operational Shifts for SEO and Content Teams
These fundamental behavioural shifts demand immediate, operational changes.
1. Structured Product Content is Now Foundational
LLMs retrieve information; they do not rank it. Clear attributes, schema, comparison blocks, and unambiguous metadata are the primary factors influencing what assistants tell users about your products.
These fundamental behavioural shifts demand immediate, operational changes.
2. Content Briefs Must be Designed for the AI-Informed Buyer
Briefs must now explicitly mandate content that covers:
● Explicit model differences
● Compatibility summaries
● Scenario-based guidance (‘best for’ or ‘avoid if…’)
● Plain-language delivery and returns caveats
● Corrective notes on common assistant errors
Example: ChatGPT repeatedly provided misinformation, asserting a specific model included a feature it did not possess. Shoppers, arriving with this expectation, could not find the feature and immediately bounced.
3. FAQs Must Exclusively Target Assistant Misunderstandings
Five precise answers are more effective than twenty generic ones.
4. Landing Pages Must Implement a Confirmation-First UX
These decision-ready users do not require education. Instead, they require accuracy, clarity, and reassurance, delivered instantly.
5. SEO Strategy Must Acknowledge Upstream Intent Shaping
The click is no longer the beginning of the user journey. It is the precise moment the assistant hands the user to you, and the user arrives mid-thought.
The Inevitable Divide
Across four retailers, thousands of AI-influenced sessions, and over £8 million in revenue, the message is singular:
AI is not replacing search; it is permanently replacing the shopper’s reasoning phase.
The consequence is a fundamental shift in user psychology at the moment of the click. Customers now arrive with decisions already formed, expectations set, and core questions already answered by an external agent.
The Retailer’s Choice
Retailers who adapt to this upstream shift will drastically outperform in 2025 by converting decision-ready demand into revenue.
Those who fail to adjust will experience a different reality: they will continue to see high-intent traffic arrive, perform deep research on their pages, and yet behave like window-shoppers—leaving not because of a lack of interest, but because of a final, frustrating misalignment.
The customer’s reasoning has moved upstream. Our optimisation work must follow it immediately.
The Harton Works Strategic Imperative
Retailers now have a choice: become a destination for confirmation and conversion or remain an irrelevant stage for re-education and abandonment.
Harton Works can provide the definitive strategy and operational roadmap to align your digital storefront with the new AI-informed user intent.
Do not wait to wonder. Contact Harton Works today for a non-obligation 30-minute consultation
Frequently Asked Questions About AI Assisted Shopping Behaviour
Related Posts
Retrieval-First™ Framework
Most businesses are rushing into AI—but only 1% have actually integrated it…
AI Search Trends 2025
As we reach Q4, three critical reports have emerged that collectively challenge…
Learning from Google Ai Overviews Usability Study
The first real usability study of Google’s AI Overviews reveals a dramatic…