Prompt Engineering for Keyword Ideation
## Prompt Engineering for Keyword Ideation
When most marketers first use AI for keyword research, they type something like: “Give me a list of SEO keywords for a project management software.”
The output is predictably generic. The AI spits out high-volume, hyper-competitive head terms like “project management,” “task software,” and “team collaboration.” These are practically useless for a small-to-medium business trying to carve out organic market share.
Large Language Models (LLMs) like ChatGPT, Claude, and Gemini are trained on vast swaths of the internet. When you provide a generic prompt, the AI defaults to the most statistically common responses. To force the AI out of its generic default state and generate highly relevant, audience-focused, and long-tail search terms, you must apply structured prompt engineering.
In this lesson, we will master three foundational techniques: System-Role Prompting, Framework-Based Prompting, and Few-Shot Prompting.
1. System-Role Prompting
System-role prompting (often just called “role prompting”) involves explicitly assigning a persona to the AI before giving it a task.
Because LLMs generate text by predicting the next most likely word based on context, assigning a highly specific role narrows the AI’s “vector space.” It stops answering like a helpful general assistant and starts answering using the vocabulary, frameworks, and priorities of the persona you assigned.
The Concept:
Instead of asking for keywords, you command the AI to embody an expert who understands search intent, funnel stages, and audience pain points.
Example:
Compare these two opening commands:
- Weak: “You are an SEO assistant. Help me find keywords.”
- Strong: “Act as a Senior Technical SEO and Content Strategist with 10 years of experience in the B2B SaaS industry. Your expertise lies in uncovering low-competition, high-intent, bottom-of-the-funnel keywords that drive direct conversions, rather than just chasing vanity traffic.”
Why it works: The strong prompt triggers the AI’s training data related to “B2B SaaS,” “bottom-of-the-funnel,” and “high-intent.” It sets a behavioral constraint: prioritize conversion potential over search volume.
2. Framework-Based Prompting
Even with a great role, the AI needs strict boundaries for the task itself. Framework-based prompting uses a structured acronym or sequence to ensure you never forget a critical piece of context.
For keyword ideation, the RTFC Framework (Role, Task, Format, Context) is highly effective.
- Role: Who the AI is acting as.
- Task: The specific action you want it to take.
- Format: Exactly how the output should be structured (e.g., a markdown table, a bulleted list, a CSV format).
- Context: The background information about your business, your product, and your target audience’s pain points.
Example of an RTFC Prompt:
[ROLE]
Act as an expert local SEO specialist focusing on the home services niche. [CONTEXT]
I run a boutique landscaping company in Austin, Texas. We specialize in drought-resistant, native-plant landscaping (xeriscaping) for high-income homeowners who want beautiful yards but care about water conservation. [TASK]
Brainstorm 15 informational and commercial seed topics and search queries that my target audience would type into Google before they even realize they need to hire a landscaper. Focus on their pain points (high water bills, dying grass, Austin heat). [FORMAT]
Present the output as a Markdown table with the following columns:
• Search Query
• Search Intent (Informational, Navigational, Commercial, Transactional)
• Customer Pain Point Addressed
• Proposed Content Type (e.g., Blog post, Service page, Calculator)
Why it works: By separating the prompt into distinct blocks, you prevent the AI from getting confused. The rich context about “drought-resistant” and “Austin heat” ensures the generated keywords will be hyper-specific (e.g., “how to keep grass green in Austin summer” rather than “landscaping tips”).
3. Few-Shot Prompting
The single most effective way to get an LLM to format data correctly and understand the style of keywords you want is “Few-Shot Prompting.”
“Zero-shot” means giving the AI a task with no examples. “Few-shot” means providing 2 to 3 examples (shots) of the exact input-to-output mapping you expect. This is critical when you want the AI to understand nuances like search intent mapping or specific long-tail modifiers.
The Concept:
You show the AI a pattern, and then ask it to continue the pattern. Use clear delimiters (like ### or , -) to separate your instructions, your examples, and the task.
Example:
Act as an e-commerce SEO expert. I am going to provide a broad seed topic. I want you to generate highly specific, long-tail keyword variations based on the pattern below. EXAMPLES ###
Seed Topic: Coffee Beans
Output:
• best organic dark roast coffee beans for espresso
• why do light roast coffee beans taste sour
• fair trade coffee beans subscription box
Seed Topic: Running Shoes
Output:
• best zero drop running shoes for flat feet
• how many miles before replacing trail running shoes
• waterproof winter running shoes for women
YOUR TASK ###
Seed Topic: Standing Desks
Output:Why it works: You didn’t have to write a complex paragraph explaining what a “long-tail keyword” is. By providing examples containing modifiers (best, why do, for flat feet, subscription box), the AI instantly mathematically maps the pattern and will generate standing desk keywords with similar depth (e.g., “best motorized standing desk for tall people”).
Applied Exercise: The Audience-Centric Prompt Deliverable
To master this, you must move from reading to building. Below is a realistic scenario. Your deliverable is to write a single, comprehensive “Mega-Prompt” that combines System-Role, Framework-Based (RTFC), and Few-Shot techniques.
The Scenario:
You are the in-house marketer for a B2B startup that sells “AI-powered inventory management software for independent bookstores.” You need to write a prompt to generate bottom-of-the-funnel (BOFU) keyword ideas.
Instructions for your deliverable:
Draft your prompt in a text editor using the following checklist to ensure you have met the lesson objectives:
- Role: Did you assign a specific expert persona? (e.g., B2B SaaS SEO Strategist)
- Context: Did you include the specific target audience and product? (e.g., Independent bookstore owners struggling with overstocking)
- Task: Did you specify the exact type of keywords? (e.g., Bottom-of-funnel, high-intent commercial queries)
- Format: Did you request a specific output structure? (e.g., Markdown table with Intent and Funnel Stage)
- Few-Shot Examples: Did you provide at least two examples of the quality of keywords you expect? (e.g., Example 1: “inventory software vs manual spreadsheet for bookstores”)
Self-Correction Check: If your drafted prompt is shorter than three sentences, it is too generic. A well-engineered prompt for keyword ideation is typically a structured mini-brief running 100 to 250 words. Once you have drafted this prompt, test it in ChatGPT or Claude. The quality of the output table will immediately validate whether your prompt engineering was successful.