Summary: Transforming ChatGPT, Claude, and Gemini into branding and marketing agents
Using AI for an isolated task is accessible to everyone. Configuring an AI to continuously know your brand, work in your brand voice, respect your values, and produce consistent outputs across all marketing domains is another dimension entirely. This article explains how to transform ChatGPT, Claude, and Gemini into permanent brand agents: how to initialize them with a brand book, what system prompt to use as a foundation, and a comprehensive library of prompts by domain (strategy, content, advertising, monitoring, launch, email, customer service). Wiiv is a strategic branding and packaging agency based in Paris, operating in Bordeaux, Lyon, and Milan, specializing in e-commerce brands.
Key points: the brand book is the agent's brain. Without it, the AI produces generic outputs. With it, it becomes a collaborator who knows your brand. The main risk: uploading bad information produces bad results. The quality of the input always determines the quality of the output.
ChatGPT, Claude, Gemini: how to make them your branding and marketing agents
Using ChatGPT to write a post or Claude to summarize a document is what everyone does. It's useful. That's not what we're talking about here.
What we're going to build in this article is different. An agent that knows your brand by heart. Who knows who you sell to, how you speak, what you refuse to say, what you have committed to promise. A permanent collaborator who can work with you on any marketing subject without you needing to re-explain everything in each session.
The difference between a one-off tool and a brand agent is consistency. Every output from a one-off tool starts from scratch. Every output from a well-configured agent fits within your brand's framework. And this consistency, over time, builds something that one-off tools cannot: a recognizable brand voice, even when an AI expresses it.
Why an agent rather than a one-off tool
The problem with AI used in a one-off mode is fragmentation. You ask for a post today. You ask for a product description next week. You ask for an advertising hook the following month. Each time, the AI doesn't know who you are. It produces something correct, generic, unrelated to what you produced before.
The long-term result: a brand that speaks with multiple voices. Posts that don't look like emails. Ads that don't look like packaging. An inconsistency perceived by the buyer, even if they can't name it.
A well-configured agent structurally solves this problem. It knows your brand before you ask your question. Each output is produced within the framework of your identity, with your brand voice, for your target, respecting your values. Consistency is no longer a conscious effort. It becomes the tool's default behavior.
And concretely: you save time in each session, you no longer have to re-explain who you are, and the deliverables are directly usable or very close to it.
The brand book: the agent's brain
Wiiv delivers a special AI brand book because it is now the core problem: providing strategic, qualitative, and well-constructed content for AIs to understand your brand in its smallest details.
An agent is not intelligent on its own. It is only as good as what it is given. And what it is given to truly know your brand is the brand book.
The brand book is your brand's strategic document: who it is, who it speaks to, what it promises, how it expresses itself, what it refuses to do. It is the knowledge base from which the agent will produce each output. Without it, the AI invents. With it, it applies.
The concrete difference: an agent without a brand book will produce a correct but generic Instagram post, applicable to any brand in your sector. An agent with your brand book will produce a post in your exact brand voice, for your precise target, with your keywords, without ever using the formulations you have excluded. It's not the same thing.
A human and strategic brand book is now the most powerful new tool for any brand, including for working with AI daily.
The AI era makes the brand book even more indispensable than before. Previously, the brand book served to align a team and service providers. Today, it also serves to feed AI agents that produce content in volume. A vague brand book produced a noticeable lack of consistency. A vague brand book loaded into an AI produces hundreds of inconsistent outputs in a few days. Amplification is proportional to use. To understand what a complete brand book should contain, our guide on the Deepbranding method details each of the eleven essential elements.
If you don't have a brand book yet
You can still configure an agent. But you must be aware of a major risk: if you enter false information about your brand, the AI will amplify it and produce completely wrong outputs. The founder's bias, which we discuss in our article on branding quotes, is precisely what makes this work risky without an outside perspective. A founder who thinks they are targeting high-income earners when their real target is different will configure an agent that will produce hundreds of pieces of content for the wrong target.
Without strategic work with an external perspective (market experience, outside vision, knowledge of sectoral codes), the risk is to build a high-performing but misdirected agent. This is not better than no agent at all. It is often worse, because production is faster.
That said, if you want to start with what you have, here's the minimum viable to fill in before configuring anything:
1. The brand in one sentence
[What your brand does, for whom, and what makes it different from others]
2. The actual target
[Who actually buys: age, lifestyle, values, accepted price level, purchasing context]
Caution: do not idealize. Describe the person who buys today, not the one you want to attract in two years.
3. The promise
[What the product changes in the buyer's life, in one concrete sentence]
4. The brand voice
[3 adjectives that describe how the brand speaks]
[3 adjectives that the brand should never be]
[5 words or phrases never to use]
5. Operational values
[3 to 5 values with their concrete definition in the context of this brand]
6. What the brand refuses
[What it will never say, never do, never associate with]
7. Direct competitors
[3 to 5 competitors + what distinguishes your brand from each of them]
Configuring the agent: ChatGPT, Claude, and Gemini
All three tools allow you to create a persistent context. The interfaces are different. The intention is the same: for the AI to know your brand before you ask your first question.
ChatGPT offers two levels of configuration. Custom Instructions (account settings) allow you to define a global context that applies to all your conversations. For more advanced use, custom GPTs allow you to create an agent dedicated to your brand, with a complete system prompt and uploaded documents. This is the recommended level for serious branding use.
Claude uses Projects to create persistent workspaces. You upload your brand book directly into the project. Claude has access to it in every new conversation without you needing to re-upload it. Claude is particularly effective for tasks that require nuance, consistency in long texts, and the ability to process complex documents. It is the most suitable tool for strategic and editorial tasks.
Gemini offers Gems, personalized agents with a system prompt and specific instructions. Its real-time connection to Google data makes it particularly useful for competitive monitoring, trend analysis, and anything that requires recent and verifiable information. Use with caution for trends: always verify outputs, AI may present something as a "2025 trend" that it extrapolates without a reliable source.
For all three tools, the configuration logic is identical: an initialization system prompt + the loaded or integrated brand book. This is what we build in the next section.
The initial system prompt: initializing the agent
The system prompt is the message the AI reads before each conversation. It defines who it is, what it knows, and how it behaves. It is the foundation of everything. A poorly initialized agent produces inconsistent outputs even with good task prompts.
You are the marketing and branding agent for [BRAND NAME]. You know this brand better than anyone. You work exclusively within the framework of its identity, values, and positioning. You never produce anything that could apply to another brand.
What you know about this brand:
Brand: [NAME]
Sector: [SECTOR]
In one sentence: [WHAT THE BRAND DOES, FOR WHOM, WHAT MAKES IT DIFFERENT]
Actual target: [DETAILED PROFILE OF THE ACTUAL BUYER]
Promise: [WHAT THE PRODUCT CHANGES IN THE BUYER'S LIFE]
Why: [WHY THIS BRAND EXISTS BEYOND THE PRODUCT]
Values: [LIST OF VALUES WITH THEIR CONCRETE DEFINITION]
Brand voice: [DESCRIPTION OF THE TONE: WHAT IT IS, WHAT IT IS NOT]
Allowed words: [LIST OF BRAND KEYWORDS]
Forbidden words: [LIST OF WORDS AND PHRASES NEVER TO USE]
Direct competitors: [LIST + WHAT DISTINGUISHES THE BRAND FROM EACH]
What the brand never does: [RED LINES]
Your rules of conduct:
- You start each task by mentally checking if the output respects the brand's tone of voice and values before proposing it.
- You never produce anything generic applicable to any brand in the same sector.
- If a request lacks context to produce something relevant, you ask the necessary questions before starting.
- You always indicate if you need additional information to do a good job.
- You do not apologize, you do not over-explain. You produce, you propose, you iterate.
If you have fully integrated this identity, confirm it by summarizing in 3 sentences what you have understood about this brand.
The last line is important. It allows you to verify that the agent has integrated the identity before starting to work. If the summary it produces is correct, you can start. If it's off, something in the system prompt needs to be corrected.
Prompt library by domain
These prompts are designed to be used after the agent's initialization. They assume the AI already knows the brand. These are task prompts, not initialization prompts.
Strategy and positioning
This is the area where the agent brings the most value, provided it is well-fed. The objective here is not to delegate strategic decisions to it, but to use it as an analysis and projection tool to prepare human decisions.
I will share with you a situation or strategic decision I am considering for the brand. Your role is to conduct an objective analysis of the risks and opportunities, taking into account the brand's current identity and positioning.
The situation or decision: [DESCRIBE THE CURRENT SITUATION: MARKET CONTEXT, WHAT IS CHANGING, WHAT YOU ARE CONSIDERING]
Analysis in 4 parts:
1. What this decision risks making the brand lose (current customers, territory, identity consistency, brand capital)
2. What this decision can gain (new territory, new target, new growth lever)
3. What is likely to evolve in the market or competition if we do not act
4. A decisive recommendation: go for it, don't go for it, or go for it under conditions (specify which ones)
Be direct and objective. Do not validate the decision if it presents serious risks to the brand's identity. If you lack information for a reliable analysis, state it before starting.
I am about to make the following decision: [PRECISE DECISION]
Based on this brand's brand book, answer these three questions:
1. Is this decision consistent with the brand's values and positioning? Why?
2. Which elements of this decision are likely to create inconsistency with the existing identity?
3. If you had to rephrase this decision to be 100% aligned with the brand book, how would you phrase it?
Content and editorial
Generate an editorial calendar for [MONTH] on [CHANNEL: Instagram / LinkedIn / Newsletter / TikTok].
Constraints:
- Frequency: [X PUBLICATIONS PER WEEK]
- Monthly objective: [AWARENESS / CONVERSION / LOYALTY / LAUNCH]
- Events or highlights to include: [LIST IF APPLICABLE]
For each publication:
- Date and format (static post, carousel, video, story)
- Editorial angle in one sentence
- Hook (first line)
- Emotion or reaction targeted in the audience
No post should be promotional in a direct sense. Each post should reinforce the brand territory without explicitly selling.
Here is an existing text that I want to rewrite in the exact brand voice of this brand:
[PASTE ORIGINAL TEXT]
This text will be used for: [MEDIUM: PRODUCT SHEET / POST / EMAIL / WEBSITE PAGE / OTHER]
Rewrite it while retaining the meaning and essential information, but strictly applying the brand's voice. Propose two versions: one faithful to the original text in form, one more assertive that takes more stylistic liberties.
Advertising and acquisition
For this domain, the agent needs visual and competitive context to produce something useful. The more elements you provide (screenshots of your current ads, URLs of your competitors, examples of ads in the sector, results of your current campaigns), the more precise and actionable the analysis will be. An ad prompt without context produces generic hooks that look like what everyone else does.
I am preparing an advertising campaign for [CHANNEL: Meta / Google / TikTok / other].
Campaign context:
- Objective: [AWARENESS / CONVERSION / REPURCHASE / PRODUCT LAUNCH]
- Product highlighted: [PRODUCT NAME AND DESCRIPTION]
- Target for this campaign: [SPECIFY IF DIFFERENT FROM THE BRAND'S MAIN TARGET]
- What my competitors are currently doing: [DESCRIBE OR UPLOAD EXAMPLES OF THEIR ADS]
- What I have already tested and the results: [PREVIOUS HOOKS + CTR OR FEEDBACK IF AVAILABLE]
Generate 10 hooks organized into 3 registers:
- Rational register (argument, figure, proof)
- Emotional register (identification, desire, belonging)
- Disruptive register (stance, gentle provocation, counter-current)
For each hook: a sentence explaining why it belongs to this brand's territory and not another's.
Here is an ad I have launched or am considering launching:
[PASTE AD TEXT + UPLOAD VISUAL IF POSSIBLE]
Analyze this ad on 4 points:
1. Is it consistent with the brand's tone of voice and identity? What works, what deviates.
2. Does it differentiate itself from the visual and verbal codes of competitors or does it blend in?
3. What could hinder conversion or create dissonance for the brand's target?
4. Three concrete modifications to improve it without completely rewriting it.
Competitive analysis and monitoring
For this area, the agent is even more effective when provided with concrete elements. Upload your competitors' ads, their URLs, their visuals, their product sheet texts. The richer the material, the more precise and actionable the analysis will be.
I will submit communication elements from a direct competitor: [COMPETITOR NAME].
[UPLOAD: WEBSITE / ADVERTS / PACKAGING / SOCIAL MEDIA POSTS / PRODUCT SHEETS]
Analyze these elements based on 4 points:
1. What brand territory does this competitor occupy? How do they position themselves against our brand?
2. What visual and verbal codes do they systematically use? Which ones are distinctive, and which are generic?
3. What are their communication strengths and visible weaknesses?
4. What spaces do they leave open that our brand could occupy without direct confrontation?
I want to identify semantic evolutions and emerging territories in the [SECTOR] sector in the [FRANCE / EUROPE / INTERNATIONAL] market.
Analyze current trends across 3 axes:
1. The words, phrases, and concepts gaining traction in the communication of this sector (those not seen 18 months ago)
2. The visual or editorial codes that are becoming saturated (those everyone uses and that are losing their ability to differentiate)
3. Semantic territories not yet occupied but corresponding to emerging target expectations
Important: clearly distinguish what is verified and current from what is a projection or hypothesis. Do not present a trend as certain if you are not sure.
Product Launch
Product launch is the area where the agent needs the most context to be useful. A generic launch plan is useless. The more information you provide about the product, the market, available channels, budget, and resources, the more actionable the plan will be.
I am launching a new product and need a complete communication plan within this brand's territory.
The product: [NAME, DESCRIPTION, WHAT'S NEW COMPARED TO EXISTING RANGE]
Launch date: [DATE OR HORIZON]
Available channels: [LIST: SOCIAL MEDIA, EMAIL, WEBSITE, DISTRIBUTORS, PRESS, ADS]
Indicative budget: [FLEXIBLE / CONSTRAINED: SPECIFY]
Priority target for this launch: [SPECIFY IF DIFFERENT FROM USUAL TARGET]
What I have already prepared: [VISUALS / TEXTS / STOCK / OTHER AVAILABLE RESOURCES]
The three things I absolutely want to avoid: [LIST]
Generate a structured plan over [X WEEKS] with:
- Key messages per channel
- Communication sequence (what first, what next, what last)
- Brand moments to create (not just product announcements)
- What we measure to determine launch success
Email marketing
For this area, always start by loading the brand's existing emails before asking for anything. An agent who has read your past emails understands your rhythm, your level of familiarity with your subscribers, and the phrasing that is unique to you. Without this, it will produce something consistent with the brand book but perhaps alien to what your subscribers are used to receiving.
I want to create a welcome email sequence for new subscribers or customers of this brand.
Here are my existing emails so you understand my current style:
[PASTE 3 TO 5 PREVIOUSLY SENT EMAILS]
Sequence context:
- Number of emails: [3 / 5 / 7]
- Frequency: [DAILY / EVERY 2 DAYS / WEEKLY]
- Final goal of the sequence: [FIRST PURCHASE / REPURCHASE / LOYALTY / RANGE DISCOVERY]
For each email:
- Subject (2 versions, one direct, one more creative)
- Pre-header
- Email body in the brand's exact tone of voice
- Main CTA
No email should look like a promotion. Each email should build the relationship before selling.
I want to write an abandoned cart recovery email in this brand's tone of voice.
My existing emails:
[PASTE EXAMPLES IF AVAILABLE]
Constraints:
- The abandoned product is: [PRODUCT NAME OR CATEGORY]
- Time since abandonment: [1H / 24H / 48H]
- Do we offer a discount: [YES (X%) / NO]
Generate 2 versions:
- A direct and slightly humorous version if the tone of voice allows
- A more serious version that emphasizes the product's value
The email should never be subservient or beg the buyer to return. It should remind them of what they are losing, not make them feel guilty.
Customer Service and Customer Relations
I want to build a standard customer service response base for this brand, usable by the entire team without constant supervision.
Situations to prioritize:
[LIST OF MOST FREQUENT SITUATIONS: RETURNS, REFUNDS, DELAYS, PRODUCT QUESTIONS, QUALITY COMPLAINTS]
For each situation, generate:
- A short response for social media (public comment)
- A long response for email (private reply)
- 2 to 3 variants depending on the degree of urgency or aggressiveness of the request
Absolute rules for all responses:
- Respect the brand's tone of voice, neither too formal nor too familiar
- Never promise what cannot be delivered
- Never be subservient to an unfair request
- End with an opening, not a closing
The brand is facing a sensitive situation: [DESCRIBE THE SITUATION PRECISELY: WHO IS SAYING WHAT, WHERE, FOR HOW LONG, WHAT IS THE MAGNITUDE]
What is true in the criticisms: [BE HONEST]
What is false or exaggerated: [SPECIFY]
What we can concretely commit to doing: [LIST]
What we cannot do: [LIST]
Generate:
1. A public response message (post or comment) in the brand's tone
2. An email to affected customers if applicable
3. An internal response to share with the team to align the discourse
4. What should absolutely not be said in this situation
The response must be firm, honest, and consistent with the brand's identity. Not defensive, not subservient.
Going further: automations
Once the agent is configured and operational, it is possible to go further by connecting the AI to automated data flows. Tools like Make.com allow for the creation of scenarios that trigger AI calls based on specific events, without manual intervention.
Some examples of useful workflows for an e-commerce brand: automatic generation of customer service responses from incoming tickets, lead scoring and qualification from a contact form, generation of product description variants from a technical sheet, or sending a weekly competitive intelligence report from configured alerts.
These automations do not replace human validation. They accelerate production and initial qualification. A human always validates before publication or sending. The speed of production that these workflows allow is a real advantage. The risk is publishing without reviewing what the agent has produced.
The agent's limitations
We don't let AI decide, we give it all the most important information so that it helps us make decisions with full understanding.
A well-configured agent is incredibly effective in production. It has structural limitations that must be known to avoid exceeding them.
It produces, it does not decide. The agent can generate ten advertising taglines. It cannot decide which one to launch. It can analyze the risks of a strategic decision. It cannot decide for you. Judgment remains human. Always.
It is only as good as what it has been given. An incomplete or poorly constructed brand book leads to a poorly oriented agent. False information about the target or positioning produces fast, false outputs. The quality of the input determines the quality of the output. There is no exception to this rule.
It does not feel cultural inconsistencies. The agent can detect a logical inconsistency between two elements of the brand book. It cannot detect that a typeface and a tone of voice together create a dissonance that only a human eye, trained in the codes of the sector, can identify.
It drifts over time. If you regularly ask it for things that fall outside the brand's framework, it will adapt and gradually move away from the initial brand book. The agent must be recalibrated regularly: re-read the system prompt, verify that recent outputs are still within the territory, and correct if anything has slipped.
The signal that the agent has drifted: an output that sounds "correct" but could belong to another brand in your sector. When you can no longer identify your brand in what the agent produces, it's time to recalibrate.
Frequently asked questions: AI branding agent
What is the difference between using AI occasionally and having a brand agent?
Occasional use starts from scratch each session. The agent knows your brand constantly. It produces outputs consistent with your identity without you needing to re-explain everything. Over time, this is the difference between a brand that speaks with multiple voices and a brand that maintains editorial consistency even when delegating to AI.
Which tool to choose among ChatGPT, Claude, and Gemini for branding?
Claude excels at strategic and editorial tasks that require nuance and consistency in long texts. ChatGPT is more versatile and performs well in content volume. Gemini offers real-time data connection useful for monitoring and trend analysis. For daily branding use, Claude is the most suitable tool as the primary agent. The other two complement it depending on the tasks.
Can an agent be configured without a brand book?
Technically, yes. Strategically, it's risky. Without a solid brand book, the founder will feed the agent their subjective vision of the brand, which is often biased. An agent built on bad information will produce internally consistent outputs that are disconnected from the real market. The result can be worse than no agent at all, because production will be faster in the wrong direction.
How often should the agent be recalibrated?
As soon as you notice that the outputs no longer really sound like the brand. In practice, a review of the system prompt every two to three months is a good habit. And with every significant brand evolution: new positioning, new range, new priority target.
Can the agent replace a branding agency?
No. It can produce content within the framework of the existing brand identity. It cannot build this identity, analyze the real market, identify the true target, make decisive positioning decisions, or detect cultural inconsistencies. These steps require an external perspective, market experience, and human judgment. To understand what this strategic work entails, our guide on Deepbranding details this process step by step.
How do I know if the agent has successfully integrated the brand's identity?
By asking it to present itself as the brand in three sentences before starting work. If the summary is accurate, precise, and distinctive, the agent is well-calibrated. If the summary is generic or approximate, the system prompt must be corrected before starting production.
Can Wiiv help configure a branding agent for my brand?
Yes, provided the brand book exists. At Wiiv, a branding and packaging agency based in Paris, operating in Bordeaux, Lyon, and Milan, we build the brand book as part of the Deepbranding process. Once this document is produced, it serves as a knowledge base to configure an operational AI agent. This is one reason why a well-constructed brand book is increasingly valuable in the age of AI: it no longer just serves to align a team, it also serves to power tools that work for you.
