My background is political campaigning. For the first seven years of my career, I focused on best practice, noticing patterns, and working out what made campaigns win or lose. I got obsessively good at it. Thousands of data points, cross-referenced in spreadsheets. I could tell you shifts in tone, which tactics were spreading, and which seats were being targeted based on output patterns, which lines had been quietly dropped.
In my current role at NationBuilder, I support many of our customers who are doing this exact type of work. In an attempt to try and better help them, I’ve spent the last few months diving into AI, specifically Claude, as a working tool. It has made me nostalgic in the most productive way possible.
I keep finding myself thinking: What would this have changed if I’d had it in 2015?
AI would have transformed the manual labor of that work. But that brings me to the actual question: Would it have made us better? Or just busier?
Don't do more. Do better.
It has never been easier to produce twenty variations of a newsletter, a strategy paper, or a full suite of campaign materials using AI.
But there is a danger in producing so much that we forget the cardinal rule of campaign communication: Be brief. Be clear. Be consistent. Not as a style preference, but as a strategy.
Here’s how to do it:
-
Build a Claude skill: Create a skill with your campaign’s messaging framework and voice, including its core mission, vision, and approved and off-limit phrases.
-
Test ruthlessly: Run every piece of copy through your skill before it goes out, not to generate more copy, but to test what you have already written.
-
Bring in a human to do a final copy edit: Never treat AI output as a finished draft. All content must pass through a human editorial review to ensure it aligns with your campaign strategy and voice.
Personalize the delivery, not the message
Personalization done well means a consistent message, with adapted context: The same core argument, delivered with a framing that fits with the relationship you have built with each individual person. Because NationBuilder is an all-in-one system, it means you can show dynamic content easily based on the information they have shared with you. Liquid tags in NationBuilder make this practical at scale, taking one piece of content, and surfacing it differently to different segments.
How to do it:
-
Start with your people: Before you write your tailored content, put yourself in your supporters' shoes. What do they already know, what do they care about, and what do you want them to do? Map them into different segments accordingly.
-
Maintain consistency: Use personalization to adapt the framing rather than changing your message (e.g. what you send to a first time donor or new supporters is different from what you would share with someone who has a long relationship with you).
-
Test: Before you send your email or set your automations live, run a sample through your preview tools to ensure your personalization is populating correctly and to do a last check that the framing is appropriate for the segment.
Ask your data better questions
Most organizations are sitting on a goldmine of intelligence without a reliable way to surface it. AI makes correlation analysis and data segmentation accessible to people without a data science background, and when it is done well, it results in deeper relationships with supporters.
How to do it:
-
Protect and secure your data: Prior to uploading information to any AI platform, it is essential to scrub datasets of Personally Identifiable Information (PII), such as sensitive records, personal addresses, or full names. Ensure that appropriate policies and consents are active to permit this work, confirming that your data will remain confidential and won't be shared externally or utilized for training AI models.
-
Share context: The more context your preferred AI platform has, the more likely you are to get good responses back. This could include past campaign copy, data, strategy documents, mission and values of your organization, press releases, etc.
-
Make better decisions: Identify three decisions your team currently makes based on instincts (or assumptions) that you would like to validate with data.
-
Ask specific questions: Write prompts and questions for AI that are as specific and clear as possible to explore the data you have whether it is supporter segments, email engagement stats, or other campaign data.
Below is an example of a clear prompt:
I am an experienced email campaigns analyst for an advocacy organization. I'm sharing our email campaign copy and performance data from the last three months including open rate, click rate, and conversion rate for each send. Review this data and tell me:
- Patterns: What's driving the highest and lowest performing sends? Look at subject lines, tone, length, send timing, and calls to action.
- What's not working: Any signs of fatigue, drop-off, or diminishing returns I should be aware of.
- Recommendation: Based on these patterns, what should our next campaign look like? Be specific about subject line approach, structure, and CTA.
Keep your answer grounded in the data I've given you rather than general best practice, and flag if there's anything you'd want to test to confirm a pattern rather than assume it.
Voice consistency is an asset. Protect it.
Many organizations struggle with fragmented messaging across their website, communications, social media, and beyond. When various people are writing your copy, your brand voice can weaken. Inconsistency not only costs you credibility, it diminishes the effectiveness of your message.
What to do:
-
Build a brand voice skill: Create a skill (or in Claude language, a project) based on a spokesperson’s past speeches, interviews, marketing materials, or press releases, and strategic messaging. Include real examples of "on-brand" copy.
-
Define constraints: When instructing the AI, be explicit about what content must be copied word-for-word (e.g. your campaign slogan) versus what can be personalized to the platform's style (emails or social media posts).
-
Define the scope of AI autonomy: Never authorize an AI to post directly to your social media or email systems without a final review from a human; avoid becoming an AI cautionary tale.
Let AI handle the busywork
AI can meaningfully reduce the administrative weight on your team. This isn't about replacing people; it's about offloading routine tasks so you can focus on the work only humans can do.
Two examples currently possible with NationBuilder + Claude:
-
Monthly supporter reports: Let AI draft the narrative analysis from your data exports, turning a two-hour task into a twenty-minute one.
-
Fast-turnaround campaign pages: Claude can help you generate and iterate on website code, allowing a team without a developer to respond at the pace you need (and the news cycle demands).
How to set guardrails:
-
Limit permissions: When connecting AI agents to your CRM, keep access to "read-only" or "draft-only" roles. Never give an AI agent unlimited rights to edit or delete records.
-
Prioritize backups: Before putting any AI-generated work into production, take a full snapshot or backup of your data so you have a clear way to revert if an error occurs.
-
Audit regularly: Review the AI's activity logs on a regular basis to make sure it's operating within your organization's security policies.
What limits most organizations hasn’t changed: the discipline to be clear about what you are trying to do, who you are trying to reach, and what you want your supporters to do next. AI doesn’t change that.
A great advocacy campaign is defined by focus, not just volume. You succeed when you mobilize people toward a clear, specific outcome, when you can look at your data and see the real human stories behind the numbers. AI can help you see patterns you'd have missed, build a campaign page without a developer, or turn a two-hour report into a twenty-minute one. But it can't replace your judgment, your empathy, or your drive.
So approach AI the same way you approach any new tool. Lead with strategy; let the tool follow in service of it, not the other way around. And don't forget to keep your eyes on the mission, not just the technology.
