The scariest thing about using AI for customer service is the confident wrong answer, a refund policy it invented, a deadline it hallucinated. The fix isn't better prompting; it's changing what the AI is allowed to know. Load your actual documents, terms and conditions, policies, trip documents, into a grounded AI notebook (NotebookLM is the one I teach), and it answers ONLY from those sources. You've built a customer-service brain based on your policies, and it can't make things up, because it cites where every answer came from.
“You can actually build a whole customer service AI based on your policies. That's not making things up.”
What to load
- Your terms and conditions, the document your team paraphrases from memory all day, now answered verbatim.
- Every live trip's documents: itineraries, payment schedules, packing lists, FAQs.
- Your process docs, how transfers work, how changes are requested, what counts as an emergency.
- Update it when documents change; the brain is only as current as its library.
Claire's own numbers
The framework, step by step
- 1
Your terms and conditions — the document your team
paraphrases from memory all day, now answered verbatim
- 2
Every live trip's documents
itineraries, payment schedules, packing lists, FAQs.
- 3
Your process docs — how transfers work
how changes are requested, what counts as an emergency.
- 4
Update it when documents change; the brain is only as
current as its library
Claire's own framework from this essay, in the order she teaches it.
Source: Claire B. Soares, from her own travel business and live trainings.
How it changes the workday
Your team stops interrupting each other with "what's our policy on…", they ask the notebook and get the answer with the source attached. Draft client replies start from your actual rules instead of an approximation. And when you eventually train a full AI Employee for customer service, this grounded library is the textbook it studies. Ungrounded AI is a clever intern with imagination; grounded AI is your operations manual that learned to talk. For customer service, you want the manual.
Common Questions
How do you stop AI from inventing policy answers?
Ground it. Load your actual terms, policies, and trip documents into an AI notebook so it answers only from those sources and cites where each answer came from.
What should go into a grounded customer-service AI?
Terms and conditions, every live trip's documents, and your process documents covering transfers, changes, and what counts as an emergency.
How does this change a travel team's workday?
Staff stop interrupting each other with policy questions, draft replies start from real rules rather than an approximation, and the library becomes the training material for future AI employees.
Written By
Claire B. Soares
6× Condé Nast Top Travel Specialist, Certified AI Consultant, and founder of Travel AI University. $3.4B+ in enterprise travel software sold to Google, IBM, Boeing, the federal government, and the DoD; $16M+ in luxury travel sales with Up in the Air Life.
About Claire
