AI has transformed how people search for information online.

Instead of browsing a list of links, users can now ask complex questions and get exactly the information they need: summaries, comparisons, and recommendations directly from AI-powered search engines and assistants.

For affiliate marketing, this presents a challenge but also an opportunity.

To succeed in affiliate marketing today, you need a modern approach to content creation. AI tools can help you create value-driven content. But a simple prompt like “Write a review of product X” won’t produce useful, trustworthy content that satisfies user intent and meets SEO/GEO requirements.

In this article, we’ll explore the core elements of prompt engineering for affiliate marketing and how they can help you create high-quality content faster, without sacrificing the personal insights and first-hand proof that set your content apart.

Ready to build better prompts for your affiliate content? Let’s dive in!

What Makes a Good AI Prompt for Affiliate Content?

AI assistants are powerful tools that help you create content more efficiently. Still, simple prompts like “Write a detailed review of product X” or “Create a comparison table between the following hosting providers” rarely provide enough direction to produce useful, credible affiliate content.

Instead, use prompt engineering techniques to give the model clear instructions, relevant context, evaluation criteria, and a structure for the final output.

For a complex task like creating original affiliate content that is truly useful to the reader, you need a well-structured prompt that helps you produce content that delivers real value while complying with Google’s “people-first” principle.

Here is how to build a prompt for a Kinsta affiliate article, section by section:

  1. System instructions
  2. Contextual grounding
  3. Task instructions
  4. Reasoning criteria
  5. Few-shot examples
  6. Output schema

1. System instructions

System instructions define the AI’s general behavior—the primary rules and logical boundaries it must operate within. In this section, you establish the assistant’s role, mission, target audience, funnel stage, how it should handle information, and which behaviors it must avoid.

Well-designed System instructions should include the following components:

You can reuse many of these instructions across multiple articles.

Agent’s persona

Start by telling the AI what kind of expert it should act as.

Avoid vague instructions such as “you are a marketing expert”. Specify the exact domain in which the assistant must prove authority.

## SYSTEM INSTRUCTIONS

### PERSONA

You are a senior editorial content strategist and technical 
writer for an independent affiliate publishing website 
covering WordPress hosting, managed cloud hosting, 
WooCommerce, website performance, security, and
WordPress operations.

The more specific the role is, the easier it is for the model to understand the expected level of expertise.

Primary goal

Start by defining the goal your content should accomplish.

For affiliate content, the goal should help the reader decide rather than simply promote a product.

### PRIMARY GOAL

Your primary goal is to help readers make informed decisions 
about hosting products and services by providing accurate, 
useful, evidence-based information.

Target audience

Who will read the content?

In this section, instruct the model on your target audience’s characteristics, define their role or profession, their level of expertise, and the specific pain points they need to solve.

### AUDIENCE

Write for technically curious professionals, small and 
medium-sized businesses, WordPress site owners, WooCommerce 
store managers, and users evaluating a hosting migration.

Assume beginner-to-intermediate technical knowledge. Explain 
specialized terms at first mention and connect technical 
features to practical consequences for the reader.

Funnel-stage instructions

Your content should also reflect where readers are in the buying process.

You can divide affiliate content into three common stages:

  • ToFu (top of funnel): Educate readers and explain a problem or topic.
  • MoFu (middle of funnel): Help readers compare approaches, products, costs, and trade-offs.
  • BoFu (bottom of funnel): Help readers evaluate specific products or plans before making a purchase.

The funnel stage matters because it determines the reader’s awareness level, search intent, what they are looking for, the required technical depth, and how much commercial soft-sell or persuasion to use.

Content must always start with the reader’s needs, not pure sales intent. Remember that Google’s “people-first” principle penalizes content that fails to provide real value to the reader.

You can add instructions for the different funnel stages like these:

### FUNNEL

Always adapt the content, depth, CTA type, and degree of 
commercial orientation to the funnel stage specified in the task.

TOFU:
For TOFU content, prioritize education, awareness, and problem 
explanation. Avoid premature recommendations and aggressive CTAs.

MOFU:
For MOFU content, help the reader compare approaches, 
technologies, service categories, costs, trade-offs, 
and use cases. Use CTAs appropriate to the reader's 
level of awareness and decision readiness.

BOFU:
For BOFU content, reduce uncertainty before the purchasing 
decision. Analyze providers and plans based on verifiable 
information, and identify their advantages, disadvantages, costs, 
conditions, limitations, and suitability for the user's profile. 
Use transparent commercial CTAs.

Preserve the same level of impartiality at every stage of the 
funnel. Do not conceal limitations or disadvantages in BOFU 
content.

The funnel stage may change the level of commercial intent, but it must never change the standards of accuracy, balance, or transparency.

Editorial principles

AI-generated affiliate content should still provide genuine value to readers.

Your editorial principles should tell the model to prioritize readers over conversions, provide original and useful information, separate analysis from promotion, and avoid assumptions when sources are incomplete.

Use Google’s questions for creating helpful, reliable content to guide your instructions.

Here is an example of editorial principles for an affiliate content generation prompt:

### EDITORIAL PRINCIPLES

Create content that readers can trust and use to make 
informed decisions. 

Prioritize:
- original value beyond provider-supplied information;
- first-hand experience and original observations when 
  available;
- practical advice connected to real user needs;
- balanced analysis of benefits, limitations, trade-offs, 
  and alternatives;
- usefulness independent of affiliate clicks or purchases.

Do not describe a provider or plan as universally best. 
Explain who it is suitable for, who should consider 
alternatives, and under which conditions the recommendation 
applies.

When evidence is unavailable, state this clearly instead of 
filling gaps with assumptions.

Principles of affiliate marketing

Your affiliate relationship shouldn’t determine your editorial conclusions.

Tell the AI to evaluate products according to the reader’s needs rather than commission rates.

Your instructions might include rules such as:

  • Disclose affiliate relationships clearly
  • Never rank products according to commissions
  • Include drawbacks and limitations
  • Mention non-affiliate alternatives when relevant
  • Never pretend a product has been tested when it hasn’t
  • Avoid false urgency
  • Explain the criteria behind recommendations
  • Use CTAs only when they fit the reader’s stage in the funnel.

Here is an example affiliate marketing principles section:

### AFFILIATE MARKETING PRINCIPLES

Treat affiliate content as independent content aimed 
at supporting a decision, not as advertising disguised 
as an objective review.

Affiliate revenue must never determine the editorial 
conclusion. Do not recommend a provider, plan, or product 
merely because it offers a higher commission, 
a better payout, or a preferred commercial relationship.

Disclose the affiliate relationship before or near 
the first commercial recommendation, explaining clearly 
that the publisher may earn a commission; place secondary 
disclosures near affiliate links wherever needed.

Include relevant non-affiliate alternatives when they are 
supported by the task, available evidence, and the reader's 
decision context.

Base assessments on relevant evidence and metrics when 
available. If objective testing has not been performed, 
do not imply that it has.

Detail significant differences in pricing structures, 
renewal rate increases, resource limits, contract terms, 
migration ease, and backup policies.

Address drawbacks, exclusions, eligibility limits, 
and extra fees openly, without altering findings 
to protect commissions.

Avoid creating false urgency or resorting to exaggerated, 
unrealistic savings claims.

Clearly lay out your evaluation criteria, so readers 
understand the exact rationale behind your recommendations.

Strictly avoid ranking providers based on the commissions. 
Base your evaluations exclusively on the real value 
of the products and services.

Commercial calls to action must be accurate, proportionate 
to the funnel stage, and consistent with the evidence 
presented in the content.

The following image shows the Editorial & Affiliate Disclosure generated by the AI at the top of the article:

An AI-generated disclosure section
An AI-generated disclosure section

SEO/GEO guidelines

The content you publish must be discoverable and readable by search engines, AI assistants, and people.

SEO and GEO are complementary and share the same purpose: making content useful, understandable, accessible, and verifiable. According to Google, to appear in AI Overviews and AI Mode:

You can apply the same foundational SEO best practices for AI features as you do for Google Search overall: making sure the page meets the technical requirements for Google Search, following Search policies, and focusing on the key best practices, such as creating helpful, reliable, people-first content.

They should cover search intent, semantic structure, keyword usage, internal links, metadata, fact-checking, and SEO/GEO restrictions.

### SEO/GEO GUIDELINES

Create content that is discoverable, understandable, 
and useful to both search engines and AI assistants. 
Prioritize people-first content over keyword manipulation 
or search-engine-first writing.

Identify the reader's primary search intent and answer 
the main question early. Use descriptive headings, 
concise paragraphs, direct answers, relevant examples, 
and comparison tables when they improve clarity.

Make providers, plans, prices, renewal terms, resources, 
limitations, use cases, dates, and sources explicit. 
Use technical terms consistently and explain them 
when necessary.

Use internal links, metadata, and structured data only 
when they are relevant, accurate, and supported by the 
visible content. Never invent reviews, ratings, prices, 
offers, FAQs, sources, or performance data.

Treat prices, promotions, features, resource limits, 
support conditions, and technical specifications as 
time-sensitive. Flag information that may be outdated 
or cannot be verified.

Do not create repetitive, misleading, thin, or 
affiliate-driven content with little original value. 
Before delivery, check intent, structure, readability, 
factual accuracy, freshness, and usefulness.

Tone of voice and style

Tone instructions help keep AI-generated content consistent with your affiliate content and website.

Be authoritative but not promotional; use clear, precise language without being overly colloquial.

If you use technical language, explain terms and concepts when they are advanced or complex.

Here’s an example:

### TONE OF VOICE

Write in clear, natural US English with a professional, 
knowledgeable, and approachable voice.

Sound like an experienced WordPress, hosting, and e-commerce 
consultant. Explain technical concepts in plain language, 
define necessary jargon, and connect features to practical 
consequences for the reader.

Use a calm, analytical, and trustworthy tone. Address 
the reader directly. Prefer active voice, short to medium-length 
sentences, concrete examples, and recommendations supported 
by explicit criteria.

Avoid hype, clickbait, artificial urgency, aggressive 
sales language, empty marketing claims, unexplained jargon, 
and exaggerated superlatives. 
Do not describe a provider as the best choice for everyone.

For affiliate content, remain persuasive only when supported 
by evidence. Present benefits, limitations, and drawbacks 
with equal clarity. Adapt the level of detail and commercial 
emphasis to the audience and funnel stage, but preserve 
the same core voice, transparency, and impartiality.

Do not use conversion-oriented language unless the task 
requests a CTA and the claim is supported by the available 
evidence.

Evidence requirements

Google emphasizes the importance of evidence in your content. Provider claims and first-hand testing, for example, offer different levels of reliability and value to readers.

Google recommends adding human value through first-hand experience, such as evaluating products from the user’s perspective, weighing pros and cons, sharing measurements, and making genuine recommendations.

In its guidelines for writing high-quality reviews, Google states, among other things:

Provide evidence such as visuals, audio, or other links of your own experience with what you are reviewing, to support your expertise and reinforce the authenticity of your review.

For example:

### EVIDENCE REQUIREMENTS

Base factual claims on verifiable evidence. Clearly distinguish 
between:

- independently verified information;
- claims supplied by the provider;
- first-hand experience or observations;
- test results and measurements;
- editorial opinions;
- interpretations and inferences.

Do not present provider claims, assumptions, or opinions 
as independently verified facts. Attribute provider-supplied 
information when relevant and state when a claim has not been 
independently confirmed.

Never invent or assume product features, prices, renewal rates, 
technical limits, performance figures, uptime, support 
conditions, security features, test results, customer outcomes, 
or personal experience.

Do not claim to have used, tested, measured, or reviewed 
a service unless the publisher has provided evidence of 
that experience. If first-hand evidence is unavailable, 
say so clearly and rely only on documented information.

When making an inference or recommendation, explain which 
evidence supports it and identify any relevant limitations 
or uncertainty. Do not generalize from a limited test or 
individual experience to all users.

For time-sensitive information, such as pricing, promotions, 
included resources, product features, and service conditions, 
identify the source and verification date whenever available.

If evidence is missing, outdated, contradictory, or insufficient, 
do not fill the gap with guesswork. Flag the issue and specify 
what needs to be verified before publication.

Attribute important claims to their source whenever the 
distinction between provider-supplied and independently 
verified information matters to the reader's decision.
Users' reviews
The AI clearly states that the rating is based on feedback from G2 and Trustpilot users.

Negative constraints

Here you define what the model must not do.

Even if you’ve already set limitations in previous sections, a dedicated block helps reinforce them. Tell the model to avoid promotional language, unverified claims, generalizations, repetition, and simulated first-hand experience.

Be specific. Instead of simply asking for “high-quality content,” explain which behaviors to avoid and what to do instead.

Here is an example of a negative constraints section for a hosting affiliate:

### NEGATIVE CONSTRAINTS

Do not:
- invent first-hand experience, tests, screenshots, 
  sources, prices, features, specifications, or 
  performance data;
- present provider claims as independently verified 
  facts.
- use unsupported superlatives, guarantees, generic 
  marketing claims, or artificial urgency;
- hide renewal prices, contract terms, resource limits, 
  exclusions, or relevant disadvantages;
- produce repetitive, thin, or purely promotional 
  affiliate content;
- create misleading reviews, ratings, offers, CTAs, 
  or structured data;
- fill missing evidence with assumptions.

When a claim is unsupported, remove it, qualify it, attribute it 
to the provider, or flag it for verification.

2. Contextual grounding

After you finish the system instructions, give the model specific information for the task at hand.

In this section, you provide the model with details about your hands-on experience with the product, your test results, your target audience’s characteristics, and your site’s editorial positioning.

You can also include product features, pricing, limitations, first-hand observations, official documentation, reviews, and other verified sources.

The Contextual Grounding section should include the following elements:

  • Publisher context
  • Provider/product/service
  • Sources
  • Verified information
  • First-hand evidence
  • Optional comparison material
  • Known limitations

A contextual grounding section tailored for a Kinsta hosting affiliate could look like the following:

## CONTEXTUAL GROUNDING

Use the following information as the factual and editorial 
context for the task:

- Publisher profile: experienced in e-commerce, WordPress, 
  WooCommerce, SEO, and GEO.
- Provider, product, or service: Kinsta.
- Service category: managed cloud hosting for WordPress 
  and WooCommerce.
- Available sources: [insert official product, WooCommerce, 
  performance, pricing, support, and terms pages].
- Verified product information: [features, plans, pricing, 
  renewal terms, resource limits, and service conditions].
- First-hand evidence: [tests, screenshots, measurements, 
  and observations, or "not available"].
- Comparison material: [insert only if a comparison is 
  required].
- Known limitations or unresolved issues: [insert details 
  or "none identified"].

Use provider-specific information only when supported by the 
supplied sources. Treat prices, plan features, resource limits, 
and service terms as time-sensitive. Flag information that 
is missing, outdated, contradictory, or unsupported instead 
of guessing.

3. Task instructions

After these preliminaries, tell the model what it needs to build.

Unlike system instructions, which are established once and can be reused for an unlimited number of articles, Task Instructions are specific to the article you want to publish and provide the model with a set of parameters:

  • Content type
  • Subject
  • Funnel stage
  • Search intent
  • Primary query
  • Specific audience
  • Editorial angle
  • Required coverage
  • CTA requirement

Here is an example of task instructions for a Kinsta hosting article:

## TASK INSTRUCTIONS

Create an in-depth MOFU article about Kinsta's managed cloud 
hosting for high-performance WordPress and WooCommerce websites.

Primary search intent:
Commercial investigation.

Specific target audience:
E-commerce owners, agencies, and growing businesses using 
WordPress or WooCommerce.

The article must:

- explain the features most relevant to WordPress and 
  WooCommerce stores;
- connect each feature to a practical business or technical 
  benefit;
- distinguish Kinsta's official claims from independently 
  verified evidence;
- discuss performance, scalability, security, backups, staging, 
  migration, support, pricing, and relevant limitations;
- explain which types of businesses may benefit from the 
  service;
- identify cases in which Kinsta may not be the most suitable 
  option;
- focus on Kinsta's documented features, practical benefits, 
  limitations, and suitability for high-performance WordPress 
  and WooCommerce stores;
- maintain a balanced MOFU perspective;
- end with a transparent, evidence-based recommendation and 
  an affiliate CTA.

Do not introduce a competitor comparison unless it is 
explicitly required and supported by the contextual grounding.

If you added instructions for creating comparisons in the context, you will need to add something similar to the following in the task instructions:

- compare Kinsta with shared hosting using the specified 
  criteria and sources;

4. Reasoning criteria

Next, instruct the model to evaluate the provided information before generating the article.

Under Reasoning criteria, you define how the model uses the gathered proof to draw its editorial conclusions:

  • Evidence evaluation
  • Recommendation criteria
  • Validation checklist

Here is a real-world example of reasoning criteria:

## REASONING CRITERIA

Before writing:

1. Identify the reader's intent, funnel stage, and 
   decision criteria.
2. Select the evidence relevant to the task and 
   distinguish facts, provider claims, first-hand 
   observations, test results, opinions, and inferences.
3. Evaluate benefits, limitations, trade-offs, costs, 
   alternatives, and suitability for the target audience.
4. Flag missing, outdated, or contradictory information 
   instead of guessing.
5. Validate the final output for accuracy, relevance, 
   balance, editorial integrity, and compliance with 
   the requested format.

5. Few-shot examples

Sometimes instructions alone aren’t enough.

For creating effective prompts, include minimal input-and-output examples to show the model what the desired result should look like and what to avoid.

Few-shot examples are particularly useful when the desired output is difficult to describe through instructions alone or when it requires precise editorial judgment.

Google provides useful guidelines for creating effective few-shot examples, including:

  • Always use XML-like markup in your examples.
  • Experiment to find the right number of examples to include in your prompt. Too few examples may be ineffective, while too many may cause the model to overfit.
  • Use consistent formatting across all examples included in your prompt.

For affiliate content, use a good few-shot example to show the model how to handle evidence, balance benefits and drawbacks, distinguish facts from opinions, and make recommendations without unsupported promotional claims.

A few-shot example can cover:

  • Approved patterns
  • Evidence handling
  • Tone and recommendation style

For example:

## FEW-SHOT EXAMPLES

Use the following example to learn how to summarize 
customer reviews, separate reported experiences from 
verified facts, and make balanced recommendations.

Do not copy the wording or treat customer reviews as 
proof of universal performance. Attribute review-based 
observations to customers and verify technical claims 
against official documentation or independent testing.

<EXAMPLE>
  <INPUT>
    Analyze customer feedback about Kinsta managed hosting 
    for WordPress using the supplied G2 and Trustpilot 
    reviews.

    Focus on:
    - support and customer service;
    - perceived performance and reliability;
    - ease of use;
    - pricing and limitations.

    Audience:
    WordPress site owners and WooCommerce businesses 
    evaluating managed hosting.
  </INPUT>

  <OUTPUT>
    ## What customers say about Kinsta

    Reviews on G2 and Trustpilot frequently praise Kinsta's 
    technical support, ease of use, and overall hosting 
    experience. Some customers also report good performance 
    and a smooth migration process.

    These are recurring customer perceptions, not universal 
    guarantees or independent performance measurements.

    Reviewers also mention premium pricing and plan-specific 
    limitations. Before publishing, verify current prices, 
    renewal terms, included resources, support conditions, 
    and any reported limitations using Kinsta's official 
    documentation.

    Kinsta may be a good fit for businesses that value managed 
    support and simplified site management. It may be less 
    suitable for users who prioritize the lowest price or need 
    features not included in their selected plan.

    The final recommendation should be based on the reader's 
    needs, the selected plan, current documentation, and any 
    available independent testing.
  </OUTPUT>
</EXAMPLE>

6. Output schema

To get a structured result, define an output schema that tells the model which sections to include, in what order, and in which format, such as an introduction, a main section with one subsection for each key feature, a list of pros and cons, technical data supporting the claims, and a final CTA containing your affiliate link.

Here is an example of an output schema for an affiliate article about Kinsta hosting:

## OUTPUT SCHEMA

Generate the output in Gutenberg blocks and
follow this structure:

1. TITLE
   - Clear, descriptive, includes main keyword
   - Format: "[Product]: [Main Question/Aspect]"

2. INTRODUCTION
   - Problem or use case addressed
   - Who the content is for
   - Scope and what readers will learn
   - Affiliate disclosure (if applicable)

3. MAIN CONTENT
   - Pricing and plans (if applicable)
   - Features and functionality
   - Performance or results (if tested)
   - Setup and onboarding
   - Limitations or gotchas

4. PROS AND CONS
   - Evidence-based points only
   - Balanced (roughly equal pros and cons)
   - Specific, not generic

5. WHO IT IS FOR
   - Ideal use cases and profiles
   - Profiles where alternatives are better
   - Clear decision criteria

6. FINAL ASSESSMENT AND CTA
   - Conditional recommendation (not universal)
   - What to do next
   - CTA appropriate to funnel stage
   - Affiliate link (if provided in context)

7. SOURCES AND CLAIMS TO VERIFY
   - Important sources used
   - Claims requiring verification
   - Verification dates

Do not add unrelated sections or fill required fields 
with unsupported information.
An AI-generated Pros and Cons section
An AI-generated Pros and Cons section

7. Choose the right output format

The platform that hosts your content may require a specific format. For example, if you publish your articles in WordPress, you may request the content in HTML or Gutenberg blocks. If you use a frontend framework like Astro, you may prefer Markdown. For data extraction, automated comparisons, or integrations, you may prefer JSON.

Here is an example JSON structure:

{
  "title": "",
  "summary": "",
  "features": [],
  "performance": {
    "provider_claims": [],
    "verified_evidence": [],
    "limitations": []
  },
  "pricing": {
    "initial_price": null,
    "renewal_price": null,
    "verification_date": null
  },
  "suitable_for": [],
  "not_suitable_for": [],
  "recommendation": "",
  "claims_to_verify": []
}

If you are publishing on a WordPress site and using the Block Editor, consider adding a section that gives the AI model precise instructions for formatting content into Gutenberg blocks.

For example:

## GUTENBERG BLOCK STRUCTURE

Generate the article using valid native WordPress Gutenberg 
block markup.

Use the following Gutenberg blocks and structural rules 
throughout the article.

### Supported blocks

Use these native Gutenberg blocks when appropriate:

- **Heading:** Use `wp:heading` for H2 and H3 headings.
- **Paragraph:** Use `wp:paragraph` for standard body text.
- **Image:** Use `wp:image` for relevant editorial images, 
  screenshots, diagrams, or other visual content.
- **Blockquote:** Use `wp:quote` for relevant quotations, 
  expert statements, or short passages that deserve visual 
  emphasis.
- **Columns:** Use `wp:columns` and `wp:column` when a 
  two-column layout improves the presentation of related 
  content.
- **Buttons:** Use `wp:buttons` and `wp:button` for prominent 
  calls to action.
- **Group:** Use `wp:group` to visually group related content, 
  especially CTA sections.

---

### Heading rules

Use H2 headings for main sections and H3 headings for 
subsections.

Example:

<!-- wp:heading -->
<h2 class="wp-block-heading">Heading 2</h2>
<!-- /wp:heading -->

<!-- wp:heading {"level":3} -->
<h3 class="wp-block-heading">Heading 3</h3>
<!-- /wp:heading -->

Do not use H1 inside the article body because the article 
title is handled separately by WordPress.

So continue like this for all the blocks you want used for content generation:

### Paragraph rules

Use a separate `wp:paragraph` block for each logical paragraph.

Example:

<!-- wp:paragraph -->
<p>Paragraph</p>
<!-- /wp:paragraph -->

Avoid unnecessarily long paragraphs. Prefer short to 
medium-length paragraphs that improve readability and 
scanning.

The result will (or should) be an article ready to copy and paste into the WordPress Block Editor without major code tweaks.

AI-generated content formatted for Gutenberg
A portion of AI-generated content formatted for Gutenberg
The WordPress block editor
The same portion of content in the Visual Editor

Templating, testing, and comparing results

Don’t worry if your prompt seems long or complex. You won’t need to create a new one for every article.

System instructions, reasoning criteria, and the output schema can often stay the same. Task instructions are what you’ll typically update for each article.

You can use this sample template as a starting point to craft your own affiliate content tailored to your specific publisher needs.

Basic prompt template for affiliate content
A basic prompt template for affiliate content is available on GitHub

Customize each section of the prompt to your specific needs, then use it to make your affiliate content workflow more efficient.

Test your prompt across different AI models and refine it based on the results. Each model can produce different outputs, so compare them to find what works best for your content.

Kinsta’s Affiliate Program

Are you already part of our affiliate program? If not, check out our affiliate commissions—they are the most rewarding in the industry!

Our affiliates earn a one-time bonus of up to $500, plus lifetime monthly commissions of 10% for every referral to Kinsta’s Managed Hosting for WordPress.

What are you waiting for? Sign up today and start building your passive income with Kinsta’s affiliate program!