Getting the Most from Claude:
Prompting Fundamentals

1Why Most People Prompt Badly

When we use a search engine, we've been trained to write short keyword queries. Typing "title tag analysis" into Google works because there are millions of pages to rank. Claude is different — it's generating a response specifically for you, right now, and it only knows what you tell it.

The quality of Claude's output is directly proportional to the quality of your input. Here's the same request written two ways:

The second prompt gives Claude a role, context, a specific task, and a clear output format. The response will be incomparably more useful.

2The RRCF Framework

Use this four-part framework for any non-trivial Claude prompt. You don't need all four parts every time — but the more you include, the better the output.

Let's see the RRCF framework applied to a real SEO prompt, with each component highlighted:

3Giving Claude a Role

Roles are one of the highest-leverage things you can add to a prompt. When you tell Claude "you are a technical SEO specialist", it shifts its entire response posture — it uses industry terminology correctly, it assumes you're familiar with concepts like crawl budget and canonicalisation, and it makes more expert-level judgments.

Roles for common SEO tasks

4Controlling Output Format

One of the most underused prompting techniques is explicitly requesting a format. Claude defaults to flowing prose, but for SEO work you almost always want something more structured. Here are the formats most useful for our work:

5Using Claude Projects for Persistent Context

Every new Claude conversation starts with a blank slate — Claude has no memory of previous chats. For agency work, this is a problem: you'd have to re-explain your client, your methodology, and your preferences every single time.

Claude Projects solve this by letting you set a persistent system prompt that applies to every conversation within that project. Think of it as a permanent briefing document that Claude always has open.

How to set up a Project for a client

In Claude for Teams, go to the left sidebar and create a new Project. You'll be able to write a Project Instructions document. Here's a template to adapt:

You are a senior technical SEO specialist working at [Agency Name].

CLIENT CONTEXT
- Client: [Client Name]
- Industry: [Industry]
- Website: [Domain]
- CMS: [e.g. WordPress, Shopify]
- Primary goal: [e.g. increase organic traffic to product pages by 30%]
- Target audience: [who their customers are]
- Key competitors: [list 2-3]

OUR AGENCY APPROACH
- We prioritise technical fixes before content work
- We use Screaming Frog for crawls, Search Console for performance data
- We present findings with a clear priority: Critical / High / Medium / Low
- Client-facing content should be jargon-free and solution-focused

TONE & OUTPUT
- Internal docs: concise, technical, direct
- Client docs: professional, clear, solution-oriented
- Always flag assumptions you've made if context is incomplete
  

6Positive & Negative Instructions

You can steer Claude's output by telling it both what to do and what not to do. Negative instructions are especially powerful for avoiding common frustrations.

Common negative instructions for SEO prompts:

• "Do not pad the response with introductory sentences"
• "Do not suggest issues you're uncertain about — only flag confirmed problems"
• "Do not use phrases like 'it's important to note' or 'in conclusion'"
• "Do not repeat information from the brief back to me"
• "Do not recommend tools — focus only on the strategy"

7Practice Exercises

8Summary