Most people who use AI regularly are still dabbling. They open the chat window, type a question, read the answer, and move on. Sometimes it’s useful. Often it’s fine. Rarely does it produce the time savings and sharper output these tools are actually capable of delivering.
The leap between dabbling and real results is smaller than most people think. It doesn’t require a technical background. It mostly requires a few habits that some users never develop because nobody told them to.
Here’s what actually moves the needle. If you don’t have a platform yet, start with ChatGPT or Claude. Both have free tiers and require nothing beyond an email address to sign up.
If you’ve already experimented with AI for basic advocacy tasks (like drafting press releases, social posts, or summarizing policy papers, as I covered in my earlier post on how ChatGPT can improve your advocacy campaigns), this guide shows you how to level up those results with better habits and smarter workflows.
Write a Real Prompt
Prompting is a skill, and it’s the most important one to develop. The quality of what you get out is almost entirely determined by what you put in. A vague request produces a generic response. A specific one produces something usable.
Every prompt worth writing has four parts: objective, context, source, and format. Most people include the first and skip the other three entirely.
Here’s the difference in practice. A weak prompt looks like this: “Summarize the arguments against expanding Obamacare subsidies.” A strong one looks like this: “I’m a policy staffer preparing a two-paragraph briefing for a conservative congressman who already understands the basics. Summarize the fiscal and political arguments against extending the expanded Obamacare subsidies, drawing from the attached coalition letter. Keep it under 150 words and format it as a briefing summary.”
Same underlying request. Completely different output. The second prompt gives the model a reader, a source, a format, and a length. Each one narrows the frame and gets you closer to something you can actually use on the first try.
One more addition that costs ten seconds and consistently improves output: open with a persona. A persona is a role you assign the model before making your request. For example: “Act as a senior policy analyst drafting for a Hill audience.”
Don’t Stop at the First Draft
This is the most common mistake people make. They treat AI like a vending machine. Input goes in, output comes out, transaction complete.
The better mental model is iteration. The best results almost always come from the second or third exchange. If the tone is off, say so. If it missed your actual point, redirect it. If one section is close but not quite right, push on that piece specifically. Each follow-up builds on the same context window, the model’s running memory of everything exchanged so far. Think of it as a conversation the model hasn’t forgotten yet. The longer the exchange, the more it has to work with. Most people just never bother to follow up.
Two more habits that consistently improve results. First, before the model generates anything substantial, ask it to surface its assumptions. Add a line like “Before you answer, ask me any clarifying questions you have.” It forces gaps into the open early and frequently cuts the number of follow-up rounds in half.
Second, for anything complex, tell the model to think step by step. This is called chain-of-thought reasoning, and it means asking the model to show its logic before delivering a final output. You’ll get fewer leaps, a tighter structure, and something that’s considerably easier to review. Both habits take about ten seconds to build in and consistently move you from a decent first draft to something actually usable.
Use the Features Beyond the Chat Window
The chat window is the entry point, not the whole tool. Most major platforms now offer a project or notebook feature that lets you upload documents and query across all of them. Instead of opening four policy backgrounders and tabbing between them, you upload them once and ask questions across the full set. If you’re regularly working a complicated issue area, that’s the difference between a research session that takes an hour and one that takes fifteen minutes.
If you’re using Microsoft Copilot through a work account, it connects directly to Outlook and Teams. Ask it to summarize your overnight emails and flag priorities. A prompt like that takes about thirty seconds and can replace twenty minutes of inbox triage. Most of these features go unused because they require a few extra steps upfront. Set them up once and the payoff is immediate.
Always Review What Comes Out
AI tools are confident even when they’re wrong. The term for this is hallucination: when a model generates something that sounds plausible but is factually incorrect. It will cite nonexistent sources, attribute quotes that were never said, and produce statistics that sound right but aren’t. In policy or advocacy work, this is where mistakes become reputational risks.
Read the output. Verify anything that matters. Apply your own judgment before it goes anywhere. AI accelerates the drafting process. It does not replace the review process. Keep those as two distinct steps and the tools will make you faster and more accurate. Blur them together and you’ll eventually send something you regret.
Save What Works
A well-crafted prompt for a recurring task is worth keeping. Most platforms let you save prompts. If you’ve built something that reliably produces good results, save it and use it again next time. If a colleague has already solved the same problem, use their version instead of starting from scratch and share yours in return.
The tools are genuinely capable. Most people are just capturing a fraction of what they can actually do. If you’re using AI in policy, advocacy, or communications and want to compare notes, reach out. I’m actively refining workflows that save hours each week. And if this helped you move past the dabbling stage, share it with someone who hasn’t yet figured it out.