A step-by-step guide to writing your organization's AI policy.
Your teammates are looking to you to let them know what is okay to do with AI and what is not okay.
This guide turns Janette Roush's AI Academy session into ten steps any DMO, CVB, or county tourism office can work through, plus the strategy conversation that followed.
AI governance is your plan for using AI ethically, strategically, and safely. Clear limits give people the confidence to explore. Every AI policy answers three questions.
People means your staff, your clients, and your vendors. Protection starts with the tool itself and then moves to what goes into it.
GO TO STEPS 03–04 →"When they leave your organization, every chat, every single thing they've ever uploaded into their personal AI tool leaves with them. That's a data security nightmare."
GO TO STEPS 02–03 →"If you would not happily tell someone how you used AI to create the output, don't use AI to create that output."
GO TO STEP 05 →Open each step, do the work, and mark it done. Your progress is saved in this browser so you can come back to it.
Borrow a structure that already works.
Asked where an organization with no policy should begin, Janette pointed to two models:
Then use the six-section outline in step 09 to shape your own version.
Someone who crosses departments and cares about AI.
"You need to put somebody in charge of this, even if it is only a part-time responsibility."
Ownership can sit with your HR team, the person who runs operations, or your CFO. IT is one option among several. AI works differently from other technology, and the right owner understands what each department does and how AI could help them.
Choose someone who works across departments. A marketing lead fits when that person is the most engaged with AI in your organization and is willing to work with every team.
The tool is your first guardrail.
On a free account, what you type most likely trains the next version of the model. Provide paid accounts, then check each tool for these six things:
"Improve the model for everybody, which is their polite way of saying, let us steal your information. And the answer to that is to turn it off."
Ask whether you have the right to share it, and why.
Assuming a paid tool with training turned off:
| WHAT IT IS | IN A PAID AI TOOL? |
|---|---|
| Your own drafts, notes, ideas | ✓ Yes |
| Public information | ✓ Yes |
| Internal strategy documents | ✓ Usually |
| Contacts from your inbox or CRM | ✓ For the reason they gave it to you |
| Licensed research reports | Check the license first |
| Partner confidential materials | Ask their permission |
| Visitor lists in bulk | Check your privacy notice |
| Employee HR data, salary spreadsheets | ✗ Keep it out |
Personal information follows the purpose. When someone hands you a business card, it's for doing business together. Using it in an AI tool for that business is fine. Using it for promotions unrelated to why they gave it to you is a different purpose. This tracks the purpose-limitation principle in Europe's GDPR; check your own privacy notice and counsel.
Research you license is someone else's product. A report you commissioned and own can go in. A report you license alongside other buyers usually comes with a contract term about sharing it with third parties, so read that term first.
"Ultimately, what the law comes down to, it's less this is allowed and this is not allowed. It is more, what are you doing with the information? Why are you putting this information in?"

Transparency, verification, and a human who owns the result.
Verify, and own it. The City of Boston put it plainly in May 2023: if autocorrect changes a word and changes the meaning of what you wrote, you are still responsible for the text. AI works the same way. Whatever tool produced the draft, the person who shares it is accountable.
Be open about it. Brand USA advocates strongly for transparency, because people hate to feel fooled or tricked.
"If I wouldn't be proud to say in a caption, this was created with AI, then I've crossed the line. And I think that's true for any use of AI."
Sharing when and how you used AI also teaches your colleagues what AI is good for in daily work.
More risk, more people involved.
Pick a level to see how much review it needs:
Drafting emails, brainstorming, research. Treating AI like fancy Google and letting it hunt down information for you.
Normal review, in approved tools.
Working with company data, analyzing spreadsheets, writing code for a public website.
More human in the loop. Check the output, and the business owner signs off.
Sensitive data. Anything touching HR, legal, or financial decisions. AI agents that can take actions on your behalf.
The most people involved: formal approval, testing, and tracked review of what it produces.
Start agents on read-only access. Connecting AI to your CRM so it can read your records is a big win. Letting it write updates into the CRM is a bigger step. "It's always a good idea to start small and then add in more access as you understand better how it works."
Audit file permissions before you connect AI to your files. A connector sees what that person is allowed to see. A folder hidden in an obscure location gives no protection, because AI can see everything that person's permissions allow. See Janine's question in the Q&A below.
Agencies and rep firms follow your policy too.
"It's no good if everybody at your organization is very clear on how transparent you need to be and when humans need to review outputs and your vendors aren't following those same guidelines."
Contract renewal is the natural moment to raise it.
Before you sign up for any new AI tool.
Use AI to check a new tool's privacy policy and terms of use against your own policy. Paste all three into your paid AI tool with this prompt:
I've pasted three documents: the terms of use and the privacy policy for an AI tool we're considering, and our own AI policy. Does anything in these terms clash with our policy? Is there anything that should concern us about using this tool? Check whether it trains on our data, who owns what it produces, single sign-on, and encryption. Quote the clause behind every concern.
Asking for the exact clause lets you check the answer yourself, which is step 05 in action.
A policy outline any DMO can use.

Each section answers one question. Your answers from steps 02 through 08 fill most of it in.
Structure from Roxanne Steinhoff (now general counsel at Miles Partnership) and Kara Franker (CEO, Visit Florida Keys), developed for the AI Opener for Destinations program.
Out of the shadows, all year long.
"Take this out of the shadows, start having conversations about it."
A policy works when people talk about it often, with staff and with vendors alike. Name who keeps your team's understanding current as AI changes, and include partners in your training where you can. Brand USA invites its 15 international offices to its staff AI trainings.
After the policy walkthrough, James Sauter asked how an AI strategy fits inside an organization's larger plan. Janette shared Brand USA's, with the caveat that yours will look different.
To lead global destination marketing into the AI era by using intelligent systems to connect the world to the stories, people, and places of the United States—making discovery personal, inspiration effortless, and travel decisions frictionless.
So Brand USA is as AI-first as it makes sense to be.
Helping U.S. tourism work with AI, because working smart beats trying to outspend competitors for international visitors.
Using AI tools, like the Mindtrip planner on the consumer website, to make the U.S. easy for AI to find and book.
Brand USA's FY27 strategy rests on two foundations: America the Beautiful, which brings the U.S. to life through its people for international travelers, and AI, which the strategy names as Brand USA's competitive advantage. AI is also one of four strategies, and every new initiative has to show how it uses AI.
"AI sits alongside our master campaign, America the Beautiful, for how we actually execute on the work."
Brand USA used AI to organize the company strategy, then built a website to track work against it in place of a PowerPoint.
The policy thinking took six to eight months of reading, listening, and distilling people's fears. The strategy grew from conversations with Brand USA leadership. "We were in lockstep that it needs to prioritize AI, that this is where travel discovery is moving."
The front door is how visitors discover you. Some moves cost nothing: audit how AI sees your website, check robots.txt so you allow AI crawlers, and add schema so AI tools can tell what makes you authoritative.
Fix website hygiene now. Train staff before you put AI tools in front of customers.
The session's audience was New York's DMOs, CVBs, and county tourism offices. Many run on public funds with small teams, and their questions reflected that.
You've got a choice. You can sit on the beach and pretend nothing's happening and hope that you don't get wiped out. You can run for a bunker. You can run for the hills. Or you can grab a surfboard and paddle out and ride the wave.— BOB PROVOST, NYSTIA, DURING THE Q&A
"We are stewards of public funds... I think we have a responsibility to spend them in the most responsible way possible. And I think that includes using AI if a task without it takes 7 hours and that same task with AI takes 7 minutes."
Budgets stay flat while ambitions grow. AI is how a small team does more with less.
When AI adoption works at a small organization, the CEO jumps in. Janette pointed to Kara Franker at Visit Estes Park and now Visit Florida Keys, and Tony Cordo at Visit Greenville.
"If you don't have a giant team, then people aren't going to believe that it's important to you unless you're the one doing it."
New technologies take years to show up in the bottom line, because people have to learn them and change their processes first. "If you expect to see those gains in 12 months, that is the wrong timescale to look at."
Track savings as they appear. Brand USA keeps a list of subscriptions it cancels because AI now does the job; retiring a PDF software subscription saved more than the company's Claude subscription costs.
Forcing someone with real concerns to use AI fixes nothing. The floor: your visitors use AI, so anyone touching the website needs to understand how it works.
For ethical concerns, Janette recommends Karen Hao's Empire of AI (2025) and the AI Resist List. "Don't tell yourself that refusing to use it at work is taking a stand against AI."
James Sauter's conversation questions and the audience's questions, with Janette's answers. Filter by topic.
Look at the City of Boston's AI guidelines, and Brand USA's current AI policy. Brand USA is moving its AI policy into the same format as about 20 other company policies it's putting in place. "The principles of it are not changing."
Make it a regular conversation, something more frequent than a once-a-year topic. Contract renewal is a natural moment. Brand USA includes its international offices in staff AI trainings.
Janette's bigger concern runs the other way, and it's about hours: "Don't charge me an hourly rate and then spend 3 hours writing an update that should be automated by now."
Audit who has access to which files before you connect AI to them. At Brand USA, files live in Box, and the Box connector shows Janette only the files she has permission to view. HR files sit somewhere she can't reach.
Shared Claude projects connected to Box work the same way: someone with access to the project still sees only the files their own permissions allow. Hiding files in an obscure folder offers no protection, "because AI can see everything." Once the audit is clean, "those connectors are what make AI turn from a little play toy into something very powerful."
It sits inside. Brand USA used AI to organize the company strategy and built a website to track progress on it. The strategy centers on the America the Beautiful campaign, and "AI sits alongside our master campaign... for how we actually execute on the work."
The policy work came first: six to eight months of study, driven by being asked to speak on it before feeling ready. By the time Janette joined Brand USA, that thinking was done, and the strategy grew out of conversations with Brand USA leadership about what the new Brand USA would look like. Board members work for commercial companies and will learn what these tools can do, so strong AI education and tool use is part of showing fiscal responsibility.
A little of both. On the front door, a website chatbot is the expensive option. The free moves: audit how AI sees your website, check robots.txt, add schema. "I wouldn't wait until everybody at your company is trained on using AI before you're addressing website hygiene. But at the same time, I would address staff training before I'm implementing AI tools on my customer-facing sources."
When it works, the CEO jumps in, as Kara Franker did at Visit Estes Park and Tony Cordo is doing at Visit Greenville. Teams can tell how deeply their leader uses AI. "I can tell if you're only using it to rewrite emails... or if you're only using it for a replacement for Google versus really in-depth use cases, and your team will be able to tell."
Carrot works better than stick. At a minimum, everyone needs to understand that customers and visitors are using AI, and anyone working on the website needs training. Brand USA has built a culture of curiosity and exploration; where a role truly has no need for AI, that's okay.
For environmental and ethical concerns, education and empowerment help. Read Karen Hao's Empire of AI, and use her AI Resist List if you want to take action. "You have to actually do the work."
Companies that tried to promise cost savings up front hurt their own adoption, because people need time to learn a tool and change their processes. Expect a longer timescale than 12 months. Budget about $30 a person a month for ChatGPT or Claude (Gemini comes with Google Workspace), and turn on connectors so it can help with email and documents.
Then track savings as you go. Brand USA keeps a spreadsheet of canceled subscriptions; retiring a PDF software subscription saved more than the company's Claude subscription costs.
On a flight home from Seattle, Janette asked Claude to splice new conference footage into her speaker reel. It found the footage in Google Drive, picked the most useful segments from the transcript, and cut them in. After one note about keeping the background music under the new clips, the edit was done. "There's no fancy prompting."
About $30 a person a month for ChatGPT or Claude; Gemini if you're on Google Workspace. Find "improve the model for everyone" and switch it off.
Someone who crosses departments and is engaged with AI. A part-time role works.
Start with the City of Boston's guidelines and fill in the six sections from step 09.
Connectors make AI genuinely useful at work. Clean permissions keep them safe.
Every retired subscription goes on the list. It becomes your answer when someone asks what AI is worth.