AI Search Strategy
Defining and shipping the first AI-native product
Background
The challenge
Search was Trainline's core experience, but it had not yet leveraged the power of AI. Meanwhile, travellers were increasingly turning to ChatGPT before ever opening a booking app, asking in plain language for times, prices, and the best option. That shift put Trainline's acquisition funnel at risk, and created an opening: meet travellers where planning already happens, without breaking the trust that made Trainline the place people chose to actually book.
Business context
Trainline had incorporated AI only focused on post-purchase support, but nothing yet inside Search. I set out to define that strategy: where AI could meaningfully change how people plan and book train travel, and what it would take to earn enough trust to act on Trainline's behalf. The brief was to capture intent earlier, reduce planning effort for complex trips, and protect Trainline's position as AI tools increasingly entered the travel-planning journey.
WHAT USERS TRUSTED
Trainline's data. Travellers assumed prices, times, and availability would be accurate: brand trust extended automatically to the data layer.
WHAT USERS DOUBTED
Whether an AI assistant could interpret ambiguous requests, weigh trade-offs, or pick the genuinely best option, not just a correct one.
THE IMPLICATION
Perceived intelligence, not accuracy, was the real adoption barrier. The product had to visibly reason, not just respond.
USER REQUIREMENTS
3–5 options with a clear rationale. Light, scannable layouts over chatty text. Near-instant responses with visible progress.
My role
I spearheaded and led Trainline's AI strategy for Search. I co-authored the strategy with the VP of Product, presenting it alongside the Head of Design to the CPO and CTO. It was signed off and pitched to the CEO, becoming one of the company's top two initiatives. From there, I directed concept testing with UX Research and led design across both resulting workstreams: a shipped ChatGPT integration and the native Search Assistant vision.
Sep–Oct 2025
Concept designs & testing
Nov 2025
Research synthesis, strategy pitched & signed off – 2 tracks greenlit
Dec–Feb 2026
ChatGPT integration designed & shipped
After Feb 2026
Native Search Assistant entered design
Architecture
One Search & AI Strategy
powered by the same search agent, tools, and data, delivered across 2 surfaces:
Market signal
The search landscape was shifting fast, and trust was a real barrier.
Figures below are from the strategy pitch deck used to build the business case.
62%
Global travellers who've used AI to plan or book a trip
Industry data, Jul 2025
~91%
Of all LLM-sourced traffic to Trainline came via ChatGPT
Trainline analytics
£12M
NTS/year modelled value of shifting 2% of high-value customers to the assistant
Strategy deck forecast
12
UK & Spanish high-value travellers interviewed for the AI Search research
Conversational Search study
Brand trust transferred
Distrust of general AI tools didn't transfer to Trainline. The brand earned credibility by default: travellers assumed Trainline's Assistant pulled from verified, live rail data.
Judgement was the gap
What travellers doubted wasn't accuracy, it was whether the Assistant could interpret vague requests and make the trade-offs a person would.
3–5 options felt safe
Two options felt steered. Five-plus felt overwhelming. Three to five gave a reference frame for judging value without feeling manipulated.
Commit stayed with Trainline
Even AI-enthusiastic travellers drew a clear line: general AI as advisor, Trainline as the place to actually book, pay, and get support if something goes wrong.
Concept mocks
Design tension
How much should live inside the ChatGPT conversation, versus handing off to Trainline's own app? A three-way balance between what OpenAI's surface could render, what built trust in Trainline's data, and what converted into the richer native experience.
North star
Unified cross-surface experience
User research showed that travel planning doesn't start in one place anymore. Some journeys begin inside ChatGPT, others inside Trainline, and many move between both. Instead of designing a single assistant, we created a strategy that met travellers across multiple surfaces while converging toward one connected experience over time.
Third-party apps
Full chat interface for ambiguous, multi-step requests – “I need to be in Paris by 10am Thursday” – with visible reasoning and trade-off explanation.
Trainline Search Assistant
Search would transform multi-step trip planning through natural conversation, with transparent reasoning and tradeoff exploration. The Booking Assistant would provide context-aware guidance embedded directly in search, surfacing relevant suggestions exactly when they're needed.
I defined an ecosystem strategy for how AI should exist across third-party assistants and Trainline's own products.
Shipped: Third-party Apps
(Surface 1)
A conversational interface inside ChatGPT that surfaces live train options, recommends options, and hands off to Trainline to book. Designed end-to-end for Mobile, with Web design directed and mentored.
Desktop Web Experience
Mobile Experience
Up Next: Trainline Search Assistant
(Surface 2)
The strategy that was proposed and signed off by exec leadership to build out the Booking Assistant and to enact on the future of Search leveraging AI in the roadmap.
Impact & Results
Top 2
Company initiative after Search & AI strategy sign-off
18M+
Users reached by the shipped ChatGPT integration
3 mo
From concept testing to shipped MVP: Sep 2025 to Feb 2026
0→1
First AI-native product surface inside Trainline search
Strategy, research, and design leadership
Co-authored the Search & AI strategy with the VP of Product; presented to CPO and CTO. Directed concept testing with UX Research across 12 UK and Spanish travellers, shaping the two-surface decision.
Shipped: ChatGPT Apps
Led mobile design end-to-end across iOS, Android and Web specs; mentored and directed the designer leading Web. Delivered final specs to engineering ahead of the Feb 2026 launch.
Vision: Search Assistant
Designed the Booking Assistant concept and the two-mode architecture in full. Secured exec buy-in and laid the roadmap groundwork before handing off.
The handoff was the product
The deep-link into Trainline's app was the design goal. Figuring out exactly when complexity exceeded what chat could render, and earning enough trust before asking for the jump, shaped every other decision.