Question 1

What is the CRIT framework for AI prompts?

Accepted Answer

CRIT stands for Context, Role, Interview, Task. It's a framework developed by Janette Roush for creating effective AI prompts that emphasizes providing detailed, spoken context rather than simple typed commands. The framework ensures AI systems have sufficient industry-specific context to produce strategic outputs for destination marketing professionals.

Question 2

What is Model Context Protocol (MCP) and why does it matter for tourism?

Accepted Answer

Model Context Protocol (MCP) is a technical breakthrough designed to solve AI's trustworthiness problem in travel planning by acting as a 'source of truth.' It allows personal AIs to reliably connect with real-time, verified data from destination marketing organizations—such as accessible travel routes, venue capacities, or attraction details—replacing hallucinated information with authoritative data. This is critical for tourism applications where accuracy directly impacts visitor experience and safety.

Question 3

What are the four types of AI agents for destination marketing?

Accepted Answer

The four types are: (1) Operator Agents—automate browser tasks like lead generation or web scraping; (2) Researcher Agents—perform deep analysis and competitive research; (3) Builder Agents—create digital products like websites or applications from text prompts; (4) Automator Agents—connect different apps and orchestrate multi-step workflows. Each type serves different DMO operational needs.

Question 4

How can DMOs prevent AI hallucinations in their tourism applications?

Accepted Answer

The primary solution is connecting AI systems to a 'source of truth'—a verified, authoritative database maintained by the DMO. Technologies like Model Context Protocol (MCP) enable AI to query these verified data sources in real-time, ensuring responses are based on accurate, current information rather than potentially fabricated details. Examples include databases of venue capacities, accessibility features, or attraction operating hours.

Question 5

What is the difference between Custom GPTs and Claude Artifacts?

Accepted Answer

Custom GPTs are customized versions of ChatGPT configured with specific instructions and knowledge for particular use cases—like meeting follow-up or lead research. Claude Artifacts generates reusable, interactive applications from text prompts—like calculators, data visualizations, or tools. Both allow non-technical users to create AI-powered solutions, but Custom GPTs focus on conversational workflows while Artifacts creates standalone applications.

Question 6

Can my competitor see what I upload to ChatGPT?

Accepted Answer

No—and even with model training on, it doesn't work the way people fear. AI tools aren't Wikipedia. Your query doesn't enter a searchable database. Instead, words in a particular order are added to a training corpus of billions of words to calculate the percentage likelihood of word sequences. You can't ask ChatGPT 'What did Janette prompt about her company in October 2024?' That's not how it works. That said, enterprise plans (ChatGPT Team, ChatGPT Enterprise) have model training off by default, and you can disable it in Settings → Data Controls on free accounts.

Question 7

Why did ChatGPT make up a fake LinkedIn profile when I asked for leads?

Accepted Answer

Because LLMs predict plausible-sounding text—they don't verify facts. In a live demo, ChatGPT was asked to find 20 corporate meeting planners at Fortune 500 pharma companies with LinkedIn URLs. It returned convincing results, but when the links were clicked, many led to completely different people or didn't exist. The AI hallucinated names like 'Alicia Grant at Eli Lilly' with fake URLs. This is why operator agents must be connected to verified data sources using Model Context Protocol (MCP), Connectors, or APIs.

Question 8

What's the difference between a Custom GPT and a ChatGPT Project?

Accepted Answer

Projects organize YOUR conversations; Custom GPTs are tools for your TEAM. A Project is a folder where you can group related chats—like all conversations about a specific campaign. A Custom GPT is a configured assistant with specific instructions and knowledge that can be shared with colleagues. If you're building something just for yourself to organize work, use a Project. If you're building something your team should use, create a Custom GPT.

Question 9

Why does Claude sometimes refuse to continue even when I'm paying for it?

Accepted Answer

Claude has aggressive rate limits—even on paid plans. As Janette puts it: 'Claude is a real jerk about rate limits.' The tool will frequently run out of credits mid-conversation and tell you to wait, even with a paid subscription. This is a known limitation. Workarounds include breaking tasks into smaller chunks, switching to ChatGPT for high-volume work, or using Claude's API directly with higher limits.

Question 10

Should I type or talk to AI?

Accepted Answer

Talk. Voice is becoming the primary way to interact with AI because typing takes too long and limits the detail you provide. Using voice-to-text tools like Monologue (from Every.to), Wispr, or Wispr Flow allows you to provide much richer, more natural context than you would ever type. The CRIT framework specifically recommends 'talking' your prompts because the quality of AI output is directly proportional to the context you provide—and people naturally give more detail when speaking.

Question 11

What if ChatGPT writes a bio for me that's completely made up?

Accepted Answer

In a real example, ChatGPT was asked to write a conference bio and returned impressive text claiming awards and accomplishments that never happened. The AI doesn't know what's true—it predicts what sounds plausible. Always fact-check AI-generated biographical content, and consider 'priming the prompt' by uploading your actual resume or LinkedIn profile as a source of truth.

Question 12

What's the difference between 'agentic AI' and 'AI agents'?

Accepted Answer

Nothing. They mean the exact same thing. As Janette explains: 'Agentic sounds fancier than just agent, but it means the same exact thing. Somebody was trying to make a regular word sound fancier.' Both refer to AI systems that can autonomously plan and execute multi-step tasks to achieve a goal.

Question 13

Who owns AI policy at a DMO—IT, Legal, or Marketing?

Accepted Answer

It's a blend. AI governance doesn't fit neatly into one department. Typically it involves: the COO or operations lead (overall accountability), IT (security and systems), Legal (compliance and contracts), and HR (employee guidelines and training). The key is having clear ownership of the three core questions: What are we protecting? What are we providing to employees? What are we expecting from them?

AI for Tourism Glossary

Key Terms

Operator Agent

Researcher Agent

Builder Agent

Automator Agent

CRIT Framework

Model Context Protocol (MCP)

AI Hallucination

Custom GPT

Claude Artifacts

Agentic AI

Source of Truth

Frequently Asked Questions