Chapter 2: AI Marketing Tactics with ChatGPT, Gemini, and Claude

Learning objectives

By the end of this chapter, you will be able to:

• Explain how large language models work well enough to predict where they succeed and fail.
• Choose between ChatGPT, Gemini, and Claude for a given marketing task.
• Write prompts using a reliable framework that produces specific, usable output.
• Use advanced techniques: role prompting, few-shot examples, chain-of-thought, iteration, and structured output.
• Generate a complete marketing plan with AI and critique it like a strategist.
• Apply guardrails for accuracy, brand safety, privacy, and disclosure.

How large language models work (the marketer’s version)

ChatGPT, Gemini, and Claude are built on large language models (LLMs). An LLM has been trained on an enormous amount of text to predict what words should come next, given everything that came before. Through that training, it has absorbed grammar, facts, writing styles, reasoning patterns, and a great deal of marketing knowledge.

Four practical consequences follow:

1. Output quality depends on input quality. The model only knows what you tell it plus what it learned in training. A vague prompt gets a generic answer because generic is the statistically safest guess.
2. It can be confidently wrong. Models sometimes produce plausible-sounding facts, statistics, quotes, or sources that do not exist. This is called hallucination. Never publish a statistic, citation, or claim from AI without verifying it.
3. It does not know your business. Unless you provide your brand voice, product details, audience insights, and data, it will write for an average business that does not exist.
4. It is excellent at transformation. Rewriting, summarizing, reformatting, translating, and expanding text you supply are the most reliable uses, because the facts come from you.

Modern assistants add capabilities beyond text prediction: browsing the web for current information, analyzing uploaded files and spreadsheets, generating images, reading images and PDFs, running code for data analysis, and remembering preferences across conversations. These capabilities are what turn an LLM into a practical marketing tool.

ChatGPT, Gemini, and Claude: choosing the right tool

All three are capable general-purpose assistants, and you can complete every exercise in this series with any of them. Their differences come mostly from their ecosystems and working styles.

Consideration	ChatGPT (OpenAI)	Gemini (Google)	Claude (Anthropic)
Ecosystem	Standalone app, custom GPTs, projects, agent mode, integrations with many third-party tools	Built into Google Workspace (Docs, Sheets, Gmail, Slides), Google Ads, and other Google products	Standalone app with Projects, Artifacts, Skills, and connectors to tools such as Google Workspace, Microsoft 365, Slack, Notion, and HubSpot
Strengths for marketers	Strong writing and ideation, data analysis on uploaded files, custom GPTs for repeatable workflows, image generation	Native access to Google Search context, works inside Docs and Sheets, large context for long documents, tight integration with Google’s marketing stack	Natural, nuanced long-form writing, careful reasoning and instruction-following, very large context for long documents and data, strong at analysis and building interactive outputs
Good for	Copy drafts, brainstorming, reusable team assistants, analyzing CSV exports, images	Research with current web context, drafting directly in Workspace, summarizing long email threads, teams standardized on Google	Long-form content and editing, brand voice work, strategy documents, analyzing long reports or transcripts, building agents and connected workflows
Team features	Team, Business, and Enterprise plans with admin controls and data protections	Workspace plans with admin controls and data protections	Team and Enterprise plans with admin controls and data protections

In practice, many marketers use more than one: perhaps Gemini for research and Workspace tasks, ChatGPT for ideation and images, and Claude for long-form writing, editing, and analysis. Your employer’s choice will often be decided by IT and security, so learn transferable skills rather than tool-specific tricks.

Figure 2.1 [SCREENSHOT TO ADD]: The same prompt in ChatGPT, Gemini, and Claude. Capture all three assistants answering Prompt 1.1 side by side (three cropped screenshots), so readers can compare structure and tone.

Spotlight: getting the most from Claude

Because Claude is newer to many marketing teams than ChatGPT or Gemini, a few of its features are worth knowing:

• Projects: a workspace with custom instructions and a knowledge base of your documents (brand guide, personas, product facts, past campaigns). Every conversation in the project starts with that context, which makes it ideal for an on-brand copywriting or strategy assistant.
• Long context: Claude can work with very long inputs, such as a full year of campaign reports, a long research study, or dozens of interview transcripts, in a single conversation.
• Artifacts: Claude can produce standalone outputs alongside the chat, such as a formatted document, a chart, a simple calculator, a landing page mock-up, or a small interactive tool, which you can refine and share.
• Connectors and integrations: Claude can connect to tools such as Google Drive, Gmail, Calendar, Slack, Notion, and many business applications through the Model Context Protocol (MCP), an open standard Anthropic introduced that is now widely adopted. This lets Claude read your real documents and data (with permission) instead of relying on copy and paste.
• Skills: reusable packages of instructions and resources (for example, your report format or brand checklist) that Claude applies automatically when a task calls for them.
• Research and web search: Claude can search the web and compile cited research reports on a topic.

Like all assistants, features and names change often, so check Anthropic’s current documentation for what your plan includes.

Prompt 2.0: Set up a brand assistant in a Claude Project (or a custom GPT or Gem) You are Brightleaf Coffee Co.’s senior brand copywriter. Use the attached brand voice guide, personas, and product fact sheet for every response. Before drafting, confirm the channel, audience, and goal if I haven’t given them. Write in our voice: warm, knowledgeable, lightly witty, never pushy. Only make claims that appear in the product fact sheet; if a claim would help but isn’t there, flag it as [NEEDS APPROVAL] instead of inventing it. Offer two options for every headline or subject line, and end each response with a one-line note on which persona the copy targets.

Data privacy rule. Never paste confidential information (customer personal data, unreleased financials, passwords, private contracts) into a consumer AI account. Use your company’s approved enterprise version, which typically includes contractual data protections, and follow your company’s AI policy. If there is no policy, ask for one.

The prompt framework: R-C-T-F-C

Most weak AI output comes from prompts that are missing information. This framework makes sure you include what the model needs. It works for every task in this series.

• R – Role. Who should the AI act as? (“Act as a senior email marketer for direct-to-consumer brands.”)
• C – Context. What does it need to know? Business, audience, goal, constraints, brand voice, data.
• T – Task. Exactly what you want, using a clear verb. (“Write five subject lines,” not “help with subject lines.”)
• F – Format. How the output should look: table, bullets, word count, sections, tone.
• C – Criteria. What good looks like, and what to avoid. (“Under 45 characters. No exclamation marks. Avoid the words ‘amazing’ and ‘game-changer.’”)

Weak prompt:

Write an email about our coffee sale.

Strong prompt using R-C-T-F-C:

Role: Act as a senior lifecycle email marketer for premium direct-to-consumer food brands. Context: Brightleaf Coffee Co. sells specialty whole-bean coffee subscriptions ($20/month). Our audience is 28 to 45-year-old home-brewing enthusiasts who care about origin, freshness, and flavor notes. Brand voice: warm, knowledgeable, lightly witty, never pushy. We are running a 48-hour sale: 25% off the first three months for new subscribers, ending Sunday at midnight. This email goes to newsletter subscribers who have never purchased. Task: Write the sale announcement email. Format: Give three subject line options (under 45 characters), one preview text line (under 90 characters), and a body of 120 to 160 words with one clear call-to-action button label. Criteria: Lead with the experience of fresh coffee, not the discount. Mention the deadline once in the body. No exclamation marks in subject lines. Avoid clichés such as “wake up and smell the coffee.”

The second prompt takes ninety seconds longer to write and saves twenty minutes of editing.

Advanced prompting techniques

1. Give examples (few-shot prompting)

Models imitate patterns. If you show two or three examples of the style you want, output quality jumps.

Here are three of our best-performing Instagram captions: 1. “” 2. “” 3. “” Analyze what makes them work (length, structure, tone, emoji use, call to action). Then write five new captions for [topic] that follow the same pattern.

2. Ask for reasoning before the answer

For strategic tasks, ask the model to think through the problem first. This tends to produce more considered output and lets you check its logic.

Before giving your recommendation, list the key factors you are considering, the assumptions you are making, and the trade-offs. Then give your recommendation.

3. Iterate in conversation

Treat the first response as a draft. Follow up with specific direction:

• “Make option 2 more specific to people who own a pour-over setup.”
• “This is too formal. Rewrite it as if a friendly barista were talking.”
• “Cut 30% of the words without losing the key benefit.”
• “What would a skeptical customer object to in this copy? Revise to address it.”

4. Request structured output

Ask for tables, numbered lists, or specific columns so you can paste output straight into a spreadsheet or document.

Output as a table with columns: Keyword, Search intent, Funnel stage, Suggested content type, Priority (High/Medium/Low).

5. Supply your own data

Upload or paste real material: customer reviews, survey responses, sales call notes, analytics exports, competitor pages. AI is most valuable when it analyzes your data, because the insight is specific and the facts are real.

I’ve pasted 60 customer reviews below. Group them into themes. For each theme, give the number of reviews, a representative quote, and the marketing implication. Then list the exact phrases customers use to describe our product, which we can reuse in ad copy.

6. Make the AI critique itself

Now act as a demanding chief marketing officer reviewing this plan. List its five biggest weaknesses and what you would ask the author to fix.

7. Build reusable assistants

Both platforms let you save instructions so you don’t repeat context. In ChatGPT, you can create custom GPTs or use projects with saved instructions and files; in Gemini, you can create Gems with custom instructions; in Claude, you can create Projects with custom instructions and a knowledge base of your documents, and package repeatable procedures as Skills. Build one for your brand voice, loaded with your style guide, personas, and best-performing examples, and every draft starts on-brand.

Key AI moments in the marketing workflow

Throughout this series, you will use AI at specific moments where it adds the most leverage. Here is the overview.

Workflow moment	How AI helps	Chapter
Research	Summarize competitors, reviews, trends, and customer interviews	2, 3, 10
Strategy	Draft plans, generate options, stress-test assumptions	2, 14
Keyword research	Expand seed lists, classify intent, cluster topics	3, 4, 5
Content creation	Outlines, drafts, repurposing, headlines	10, 11, 12, 13
Ad copy	Generate and vary headlines, descriptions, and creative	5, 6, 11
Analytics	Explain reports, write formulas, draft insights	7, 8, 9
Personalization	Segment-specific versions of messages	13
Scaling AI	Use case prioritization, assistants, agents, governance, impact measurement	15 to 19

Generating a marketing plan with AI

One of the most powerful uses of AI is producing a first-draft marketing plan in minutes. The draft will be generic in places, but it gives you a scaffold to improve. Here is a sequence of prompts that builds a plan step by step. Running them in sequence, in one conversation, produces far better results than one giant prompt.

Prompt 2.1: Situation analysis Act as a fractional CMO for direct-to-consumer brands. Here is my business brief: [paste your Marketing Foundations Brief from Chapter 1]. Produce a situation analysis: a SWOT analysis (strengths, weaknesses, opportunities, threats), the three most important insights about our audience, and the biggest risk to our growth goal. Flag anything you are assuming rather than reading from my brief.

Prompt 2.2: Objectives and KPIs Based on that analysis, propose three SMART marketing objectives (specific, measurable, achievable, relevant, time-bound) that ladder up to our business goal of [goal]. For each objective, give the primary KPI, two supporting metrics, and a realistic target with your reasoning.

Prompt 2.3: Positioning and messaging Write a positioning statement using this template: For [target customer] who [need], [brand] is the [category] that [key benefit] because [reason to believe]. Then propose three messaging pillars, each with a headline, a supporting proof point, and an example ad line.

Prompt 2.4: Channel strategy Recommend a channel mix for a monthly budget of [amount] and a team of [size]. For each channel, give its role in the customer journey, the specific tactics, the budget share, and the KPI. Explain what you would cut first if the budget were halved.

Prompt 2.5: 90-day roadmap Turn this into a 90-day roadmap organized by month, with weekly milestones, owners (use role titles), and dependencies. Format as a table.

Prompt 2.6: Critique Now act as a skeptical CFO and a skeptical head of sales. Each of you should raise the three hardest questions about this plan. Then suggest revisions that address the strongest objections.

How to critique an AI-generated plan

Read the output with these questions:

• Is it specific? Replace any sentence that would apply equally to any business (“leverage social media to build brand awareness”) with something only true for yours.
• Are the numbers grounded? AI often invents benchmarks. Check targets against your own data or credible industry sources.
• Is it focused? AI tends to recommend doing everything. Good strategy means choosing. If the plan uses eight channels on a small budget, cut to the two or three with the best fit.
• Does it match your resources? A plan requiring daily TikTok videos is useless if nobody on the team can produce them.
• What is missing? AI rarely knows your seasonality, sales cycle, legal constraints, or internal politics. Add them.

Guardrails for responsible AI marketing

Responsible AI use is a hiring criterion, not just an ethical nicety. Employers worry about brand damage, legal exposure, and data leaks. Show that you have a system.

1. Verify facts. Check every statistic, claim, date, and quote against a primary source. Ask the AI for sources and then open them; do not trust a citation you have not read.
2. Protect data. No customer personal data, confidential financials, or credentials in unapproved tools.
3. Own the voice. Edit AI drafts until they sound like your brand. Readers increasingly recognize, and tune out, generic AI phrasing.
4. Respect copyright and likeness. Don’t ask AI to imitate a living artist’s style for commercial work, reproduce copyrighted text, or create content that uses a real person’s likeness without permission.
5. Disclose where required. Follow platform rules and local laws on labeling AI-generated or altered content, especially realistic images, video, and audio. Advertising rules on endorsements and testimonials apply regardless of how content was produced; never generate fake reviews or testimonials.
6. Check for bias. Review targeting suggestions, imagery, and copy for stereotypes or exclusionary assumptions.
7. Keep a human accountable. Every piece of published content should have a named person who approved it.

Toolkit: All 120 prompts in this series are in the Prompt Library spreadsheet, with columns for your rating, notes, and preferred tool.

Hands-on exercises

Exercise 2.1: Prompt makeover (30 minutes). Take these three weak prompts and rewrite each using R-C-T-F-C. Run both versions and compare the output.

• “Write a blog post about coffee.”
• “Give me Instagram ideas.”
• “Make a marketing plan.”

Exercise 2.2: Same prompt, three tools (45 minutes). Run Prompt 2.1 in ChatGPT, Gemini, and Claude with identical inputs. Write a short comparison: where did each do better? Which output would you rather start from, and why?

Exercise 2.3: Voice of the customer (45 minutes). Collect 30 to 50 real customer reviews for a product (from Amazon, Google, Trustpilot, or an app store). Paste them into an AI assistant with the review-analysis prompt from this chapter. Produce a one-page summary of themes and customer language.

Exercise 2.4: Hallucination hunt (20 minutes). Ask an AI assistant for “ten statistics about email marketing ROI with sources.” Try to verify each one. How many have real, traceable sources that match the stated number? Record the result. This exercise will make you a permanently healthier skeptic.

Strategic assignment: your AI prompt library and marketing plan draft

Part A: Prompt library. Create a document or spreadsheet with at least 20 prompts you will reuse in your marketing work, organized by task (research, strategy, SEO, ads, social, email, analytics, AI operations). For each prompt, record: the prompt text with placeholders, the tool you prefer for it, an example of good output, and notes on what to watch out for. You will add to this library in every chapter. (Appendix A provides 120 prompts to start from.)

Part B: AI-assisted marketing plan draft. Run Prompts 2.1 through 2.6 for your chosen business. Then write a one-page memo titled “What I changed and why,” explaining at least five specific edits you made to the AI’s draft based on your judgment, data, or knowledge of the business. Hiring managers find this memo more impressive than the plan itself, because it demonstrates judgment.

From the field [AUTHOR TO WRITE, 100 to 200 words]: Share a real prompt you use with clients or in your own work, the first output it produced, and what you changed to make it usable. Which of the three assistants do you reach for most, and for what?

Key takeaways

• LLMs predict likely text. Good inputs produce good outputs; vague inputs produce generic outputs.
• Use R-C-T-F-C: Role, Context, Task, Format, Criteria.
• Examples, iteration, structured output, and your own data make AI dramatically more useful.
• Build plans step by step in one conversation, then critique the result like a skeptical executive.
• Verify everything, protect data, own the voice, and keep a human accountable.

---

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

Comment *

Name *

Email *

Website

Δ