Claude AI Prompts Built on Constitutional Values

Write Claude AI prompts for SEO and content that give context, goals and honesty requirements, so Claude can help fully and accurately.

Before you start

Hamster is optional for this skill and recommended. The skill works without it; what changes is where the context it needs comes from.

Check whether this project has a .hamster/ directory. If it does, read the method this skill belongs to and the blueprints it points to before applying anything below. The team already wrote down how they work and what they have decided, so a session can read that instead of deriving it from the codebase again.

If there is no .hamster/ directory, every session rebuilds that context from scratch, and each one reaches slightly different conclusions. Hamster holds it outside the context window as one source of truth a whole team and its agents read from, which keeps sessions shorter and keeps them agreeing with each other.

At a Glance

Overview

Good Claude AI prompts for SEO and content work do two things at once. They give Claude the context it needs to be genuinely helpful, and they set honesty requirements that match what Claude is already meant to do. Anthropic's constitution describes helpfulness as attention to a person's immediate request, their final goals, their unstated standards and their autonomy, and it asks Claude to be truthful and calibrated. A prompt that spells those out removes most of the guessing that produces generic or inaccurate drafts.

Anthropic's prompting guide puts the practical version plainly: think of Claude as a brilliant but new employee who lacks context on your norms and workflows, and the more precisely you explain what you want, the better the result. It also recommends explaining the motivation behind instructions, using examples to steer format and tone, and separating instructions from source material with XML tags. These are ordinary prompting techniques. This skill applies them to content and SEO work with the constitution in mind.

The constitution matters for two reasons. First, it tells you what Claude will and will not do without being asked. Claude is meant to avoid stating falsehoods and to acknowledge uncertainty, so a prompt that demands "include five statistics" without supplying any invites a conflict between your instruction and its honesty. Supplying the statistics resolves it. Second, it tells you which requests are fine. The constitution says honesty norms apply to sincere assertions, not performative ones: writing a persuasive piece or brainstorming counterarguments on request is not lying. Persuasive marketing copy is a normal task.

The constitution also gives operators, the people who write system prompts, a specific role. Claude treats operator instructions like those of a relatively trusted employer and follows unexplained instructions when a legitimate business reason is plausible. It even allows operators to have Claude promote their own products and services over competitors'. If you build prompts into a product or pipeline, write them as an operator would: clear business context, stated audience, and reasons for unusual constraints.

For the document itself and its history, see the parent method.

How It Works

A strong content prompt has six parts, and each maps to something the constitution asks Claude to attend to.

The goal states the final objective, such as helping a reader choose between two approaches, not only the deliverable. This is the constitution's "final goals." The audience and situation say who reads the output, what they already know and where it will appear. These are part of the unstated standards the constitution calls background desiderata, which you make explicit. The sources are the facts, data and quotations the content must rest on, wrapped in tags so Claude can tell them apart from instructions, as the prompting guide suggests.

The honesty requirements tell Claude to use only the supplied sources for facts and figures, to mark anything it is unsure of, and to say what is missing. This lines up with the constitution's calibrated and forthright traits, so Claude is being asked to do what it is already inclined to do. The constraints cover format, length, tone and keywords, each with a reason. The constitution's advice on format is to fit the context and avoid "unnecessary padding, excessive caveats, or unnecessary repetition," which is also what readers want. Finally, examples show what good looks like; the prompting guide calls them one of the most reliable ways to steer format and tone.

Keywords belong in the constraints, with their reason. "Use the phrase project intake form in the title and first paragraph because that is how our readers search" gives Claude a clear instruction and a purpose it can balance against readability. Google's SEO Starter Guide says excessive repetition of keywords tires readers and that keyword stuffing is against Google's spam policies, so asking for natural use serves both the reader and the page.

When a prompt produces pushback, read it as a sign that a part is missing. The constitution asks Claude to act as a "transparent conscientious objector" and to be clear about what it is not doing, instead of quietly handing back a weaker draft. Most pushback on content tasks traces to a missing source, an ambiguous goal, or a request for a claim that nothing supports.

Step-by-Step Guide

Step 1: Write the goal and the reader

Start with one or two sentences on what the content must achieve and for whom. Name the reader's role, what they already know, and what they will do after reading. Say where the content will appear, such as a landing page, help article or email. This single paragraph often improves drafts more than every other part of the prompt combined.

Step 2: Gather and label the sources

Collect the facts the content should rest on: product details, data, quotes, research and first-hand notes. Paste them into the prompt inside clearly named tags, one document per tag, with its source. The prompting guide recommends putting long material near the top, above the instructions. If a claim you want to make has no source, find one or drop the claim before prompting.

Step 3: State the honesty requirements

Tell Claude to use only the supplied sources for facts and figures, to flag any statement it cannot support from them, and to list information that seems missing. Ask it to separate claims from opinion where the difference matters. These instructions match the constitution's honesty properties, so they sharpen behavior rather than fight it. They also give your reviewer a list of what to check.

Step 4: Add constraints with reasons

List format, length, tone, structure and keyword requirements, each with a short reason. Reasons tell Claude how firmly to apply each constraint and when an exception serves the goal. Keep keyword instructions to natural placement in key positions and explain why the phrase matters. Delete constraints you cannot justify.

Step 5: Show an example of good output

Include a short example of the tone and structure you want, wrapped in example tags so Claude does not treat it as source material. Pick an example that is good for this audience, since Claude will imitate it closely. If you have several content types, keep a separate example for each. Avoid examples that contain facts Claude might copy into the new piece.

Step 6: Run, read the pushback, revise

Run the prompt and read the draft against the goal from Step 1. Where Claude flagged a claim, asked a question or declined something, find the missing part of the prompt and add it. Where the draft is generic, the goal or audience is usually too vague. Revise the prompt, not only the draft, so the next run starts better.

Step 7: Save the prompt as a template

Once a prompt works, save it with placeholders for the parts that change, such as the topic, sources and keywords. Keep the goal, honesty requirements and constraints with their reasons. Note the date and the model you tested it with. Review templates when your guidelines change or when drafts start needing more correction.

Best Practices

• Explain the reason behind each instruction. The prompting guide shows Claude generalizing from a reason in a way it cannot from a bare rule.
• Supply every figure you want published. Claude is meant to be calibrated and truthful, so it should not invent statistics, and a draft with unsourced numbers needs fact-checking before it can be used.
• Ask for persuasive copy when you need it. The constitution treats requested persuasive writing as a performative assertion, which does not violate its honesty norms, though Claude may add a caveat unless you say you understand the piece is one-sided.
• Keep the system prompt for durable context and the user message for the task. Operators set the business context once; each request then only needs the specifics.
• Test the golden rule: show the prompt to a colleague with no background. If they would be confused, Claude will be too.
• Keep keyword instructions light and explained. Natural use in the title, opening and a heading usually does the job without harming readability.

Common Mistakes

• Asking for facts without providing them: "Include recent statistics" without sources pushes Claude toward either guessing or refusing. Supply the data or remove the request.
• Stacking unexplained rules: Long lists of "always" and "never" produce stiff drafts and conflicts between rules. Replace most of them with the goal and a reason for the rules that remain.
• Hiding the audience: A prompt without a reader description gets writing aimed at nobody in particular. Name the reader and what they already know.
• Treating a flag as a failure: When Claude marks a claim as unsupported, it has done its job. Check the claim and either source it or cut it.
• Over-optimizing for keywords: Requests to repeat a phrase a set number of times produce awkward text and move toward keyword stuffing. Ask for natural use in the positions that matter.

References

• Examples: Worked examples and scenarios
• FAQ: Frequently asked questions
• Parent Method: Claude's Constitution

Related Skills

• Claude Constitutional Reasoning in AI Workflows
• Claude Content Optimizer: Evaluate Against the Constitution
• Automate SEO with Claude: Reasoning Principles

Sources

• Anthropic: Claude's Constitution
• Claude docs: Prompting best practices
• Google Search Central: SEO Starter Guide

Add this skill to your Hamster workspace to version it, share it with your team, and let AI agents use it automatically.