A practical workflow for designing a website content map through human-AI collaboration, keeping your audience’s needs at the center.

In recent articles, I introduced the One-Page Site Plan to guide the process of building a website. The third step of the process is creating a sitemap, which describes all the high-level pages your website will contain. This sounds easy in theory, but it can be surprisingly difficult to nail down the right architecture for your site content.

The sitemap needs to dance on the fine line between what matters to your organization and to your audience. 

Frequently, we want our audience to be interested in everything we do or have accomplished. But our site visitors have an agenda of their own. They are typically looking for ways to solve their problems. You’ve got mere seconds to show site visitors how your work addresses their needs, or they will disappear in a flash.

The problem compounds when you’re working in a larger organization with many departments and lots of diverse audience segments. Even smaller organizations with only a few target audience groups can find it challenging to organize their websites in a way that makes it easy for visitors to “see themselves” in the website.

Enter the Centaur — it’s easier than ever before to brainstorm, design, and pressure test a sitemap by working with AI as a thought partner. Here’s the Centaur Mode approach to sitemap creation.

Activating Centaur Mode for sitemap design

Rather than handing things off to AI and expecting magic, we get better results through a back-and-forth collaboration. I generate ideas, pass them to a GenAi (I personally prefer Claude, but whatever works for you), and then we iteratively pressure test and refine the ideas until they’re ready for primetime.

For a sitemap, we can use this Centaur approach in three stages: drafting initial ideas for key pages, refining and streamlining through multiple passes, and finalizing the map.

Draft

When considering which pages achieve the coveted status of appearing in your navigation menu, I like to apply the “four quarters” principle from the 4 ACES messaging framework. No one wants a pocketful of pennies, they’d much rather have four quarters. In comms, this principle works wonders, helping us avoid overwhelming our audience with minor details.

What are the “four quarters” your website needs to convey? 

Answering this question gets a bit trickier when a key audience is the boss, your board, or other internal stakeholders. But ideally, we will consider what matters most to the audience.

Looking back on your One-Page Site Plan, the sitemap should help advance your goals. That typically involves getting specific groups of people to take specific actions.

If you’ve created audience personas, you can use these to help narrow down some clear ideas for what matters to each audience segment. From there, you can begin to sketch out some ideas for crucial high-level pages these visitors would be interested in. Then brainstorm the most important subpages of those landing pages — guiding them deeper and deeper into the site and towards various calls to action.

Example:

Once you have draft ideas, it’s time to fire up your AI partner.

Refine and streamline

Begin by laying a foundation for how you’d like to work with your AI. If you’re working in Claude or ChatGPT, the Projects feature will be very helpful for this work. Your initial prompt should describe the goal (creating a sitemap) and reference any files that will be helpful to review (such as your audience personas, organizational program descriptions, annual reports, etc).

Paste into the prompt your draft sitemap and your One-Page Site Plan, which includes your website goals, audience, design ideas, and key functionality. It also helps to provide any additional criteria or guidelines for your AI. For example, I like to ask Claude to take the perspective of one of my personas to try to simulate their experience.

Example prompt:

I’m working on a sitemap for my nonprofit, which focuses on conservation of the Chesapeake Bay watershed. I’ve created an initial draft of the sitemap, which I’ve pasted below. We want the new website to be easier to navigate for our key target audience segments, which include: * Local community members who use the bay for recreation  * Regional foundations who provide grants for environmental conservation * Local government officials * Other connected nonprofit organizations interested in bay conservation The Audience personas 2025 document provides greater detail on the values, pain points, and desired gains for each of the segments above. The Site Plan referenced below includes some of the goals we have for the website, which include increasing our recurring small-dollar donations by x% and attracting institutional donors. Please refer to the following Project files: * One-page site plan * Audience personas 2025 * Annual report 2024 * Annual report 2023 * News articles It will also be helpful to review the following chats for helpful context from previous conversations: * [YOUR CHAT 1] * [YOUR CHAT 2] * [YOUR CHAT 3] Here is the current draft sitemap: * What we do – Watershed restoration – Education programs – Policy advocacy * Impact – Conservation results – Success stories – Annual reports * Resources – Educational materials – Research & data – News & blog * Get involved – Volunteer – Donate – Advocacy & action * Contact With all of this data, provide some initial feedback on what is working well and what can be improved.

Next, kick off some back-and-forth discussion with your AI. The AI’s initial suggestions won’t all be gold, but you will get lots of ideas to work with. The gold comes from the iteration — continue to tweak, add, remove, consolidate, and otherwise refine the sitemap. 

Finalize

It will probably take several passes and iterations to get something you’re close to satisfied with. As you work towards a final version, it’s time to pressure test. Ask the AI to ask you questions that will help you think about what’s really important vs what’s pretending to be. 

One common concern with AI chats is the tendency towards flattery — every question or response we provide often elicits gushing praise from our AI partners (“Wow, that is such a brilliant idea, Ryan!”). To minimize this, I like to encourage the AI to push back and test my ideas. I add to my Claude preferences an instruction to challenge my ideas and offer differing perspectives so that I don’t fall into the trap of believing all the AI hyper-flattery. 

One of the most challenging parts of the sitemap is coming up with clear and concise labels for each page. Typically we want one-word labels for the navigation menu, but finding the right word isn’t always easy. Ask your AI partner to come up with some alternatives, and then compare and contrast them.

Example prompt:

I’m going to share another draft that incorporates the insights from our various conversations. I’d like you to help me pressure test this near-final draft. Since I’m very close to our organization’s work, ask me questions that will help uncover some blind spots. What am I missing?   Put yourselves in the shoes of each target audience group. How hard is it to find specific kinds of information for each group? What impressions will they have? I’d also like you to challenge me — push back on my ideas so that I can see them from multiple angles and consider other perspectives. I’d rather find problems with the architecture now than have our audience members find those problems. And I’d like to make sure that each navigation menu label is concise, clear, and easy to understand. Offer any suggestions on what we can do to improve, what might not be clear, etc. Here is the latest draft: [PASTE IN NEXT DRAFT OF YOUR SITEMAP]

Soon enough, you’ll be done — with a version 1.0 sitemap in hand.

Get started

Note that I still recommend conducting user experience tests like tree testing or direct facilitated tests where possible. However, this Centaur approach is a great way to get something workable much faster. You can move beyond the generic and largely useless responses from merely handing the sitemap generation to AI. Treating AI as a thought partner means you have a tireless collaborator that can help you see your assumptions and biases. You can bounce ideas back and forth, building from your own experience while leveraging the creative outputs from your AI.

Most importantly, you keep human judgment at the center while using AI to accelerate the exploration and refinement.

Ready to try it out? Download the resource below, draft your ideas, and fire up a conversation with your AI thought partner of choice.

Resources