Overview
Zappi Marketing AgentsTM are conversational. You open a session, describe what you're working on, and the agents take it from there. There's no syntax to learn.
Learn more about Zappi Marketing AgentsTM.
This article covers practices that help you move from a brief to concepts ready for testing, and how to draw more from the agents in your setup.
How Zappi Marketing AgentsTM work
In each session, you interact with a facilitator agent. It reads your prompt, brings in the relevant agents with the right context, and returns anything it can't resolve. You don't need to choose an agent yourself.
Knowing which agents are in your setup helps you understand the responses and concepts you receive. During a session, you can ask for a specific agent's perspective. Between sessions, it helps you identify context you'd like to add.
| Agent | What it provides | Ask for it by saying |
|---|---|---|
| Creative | Concept language based on insights | Used by default when you ask for ideas |
| Best Practice | Alignment with concept writing best practice | "Pressure test this against best practice" |
| Respondent | Role-play of your consumer audience | "How would our target react to this?" |
| Visual Design | Visuals for the concept | "Bring this one to life visually" |
| Duplication | Uniqueness checks against existing concepts | "Have we done this before?" |
| Customer Data | Your past concept test results | "How did our past concepts here score?" |
| Normative Data | Zappi norms for your market and category | "How does this compare to the category norm?" |
Your setup may also include custom agents for areas such as brand and tone, audience, category and consumer trends, qualitative insight, competitor intelligence, compliance, and feasibility. The agents available to you depend on your implementation.
To add context or perspectives to your agents, contact your account team.
Ten practices for better sessions
How you prompt
- Share full context up front and add guardrails later. Start with everything you know: category, market, audience, innovation space, and your current range. Hold back guardrails such as feasibility, claims, and brand rules until you have territories to narrow. The agents build on what you share at the start, so a complete brief improves output for the rest of the session.
- Make one change at a time. After your opening prompt, steer with single adjustments. This keeps each version traceable to the change that produced it.
- Recap before a big request. Before you ask for a full concept, restate what you've agreed so far. A one-line recap carries earlier decisions into the build, so the concept comes back complete.
-
Branch instead of steering. If a direction stops working, use the Fork option to start a new thread from the last output you liked. Two or three branches from a strong territory usually produce better results than one long thread.
- Ask what an answer is based on. Agents draw on the documents and data loaded for your hub. Asking for the source shows you how much to rely on an answer and where to dig deeper.
How you run the session
- Work with a colleague. The agents generate ideas faster than one person can evaluate them. A second reviewer helps you keep pace and spot territories that are the same idea in different words.
- Set aside a dedicated block of time. Treat a session like an ideation workshop with a clear goal.
- Treat the first output as a starting point. Output improves with each iteration, so the strongest work usually comes from the second version onward. Setting this expectation with your team helps you use sessions effectively.
- Develop language and visuals together. Consumers judge a concept as a whole. Iterating on the words and the pack visual side by side gets you to a testable version faster than refining either one alone.
- End with a summary prompt. A summary makes your session output easy to move into a deck or a test.
Good to know
- When you ask an agent to change output, it regenerates the output rather than editing it. A request to change one word can also change surrounding detail. For small changes, edit the concept text directly in the output and save.
- Image generators treat text as a visual pattern, so words on a generated pack are directional rather than final.
Example prompts
Use these as starting points, not scripts.
| Purpose | Example prompt | Why it works |
|---|---|---|
| Set the brief | We are working in {category} in {market}, targeting {audience}. The innovation space is {space}. What we already have here is {X}, so I am looking for directions beyond that. Before generating anything, tell me what you already know about this and where you would want more. | Gives the full brief in one prompt and points the session toward new space rather than your current range. Asking what the agents need shows you early where to add context. |
| Narrow | Of those territories, which is weakest and why? Judge them on how distinctive they are and how squarely each answers a real unmet need. | Asking for the weakest option surfaces the reasoning behind the ranking. Stating your criteria means the answer uses the terms you care about. |
| Build one out | Recapping where we are: we are building territory {number}, for {audience}, within {constraint}. Build it into a full concept: name, the consumer insight in one sentence, the key benefit, a reason to believe, and how it is used. Keep it readable in under 20 seconds. | Carries the agreed position into the build and names every component you want back. The read-time limit reflects how respondents read concepts. |
| Close the session | Summarize this session: the concepts we landed on, the evidence behind each, and the open questions. Format it as a handover I can paste into a deck. | Naming the audience for the summary shapes its format. Asking for open questions gives you a starting point for your next session. |
Prompts to use at any point
- Test an answer: "You've told me {claim}. What is that based on, and what would have to be true for it to be wrong?" The second question shows the conditions the answer depends on, so you can build on it with confidence.
- Go deeper on a dimension: "Give me more {audience / category / trend / qualitative} context on this specifically." Naming a dimension brings forward a different agent. Asking for more detail without naming one only makes the response longer.
- Say what to leave out: "Don't include {thing}." This is often quicker than describing what you want, and it narrows the output without limiting the thinking.
Ways to go further
- For broader thinking: Add what you know about the space and audience to your opening context, then start a new session. More context up front produces more distinctive output.
- For more depth on one dimension: Ask for the dimension by name. If you need more than your hub currently holds, let your account team know which dimension you'd like to add.
- To explore alternatives in parallel: Use Fork on a strong output to take two directions from the same starting point. You keep both options and avoid moving one thread back and forth.
Share feedback
Use the thumbs up and thumbs down on any response to share feedback with the Zappi team. Your feedback helps shape how the agents develop.