Affilyflow / The practical AI guide

Affiliate marketing
with AI.

From research and product content to trackable links and better questions for your numbers. Use AI to move the work forward – and your own knowledge to make it worth reading.

Research · 4 copyable prompts · Tracking · Analysis

Find a relevant angle

Start with the reader’s question.

Build on a factual basis

Give AI material you can stand behind.

Learn from your results

Measure what you actually changed.

01 / The method

AI is one part of the work.

Affiliate marketing with AI means using an AI assistant for selected tasks in your affiliate business. You can get help sorting research, building a structure, challenging a product choice or reading a table. The business model itself is the same: you refer a relevant audience to a shop and receive commission under the programme’s terms.

A good starting point is to give the assistant a well-defined task and a factual basis. “Write an article about the best products” is a weak starting point. “Compare these two models for a small flat based on these specifications and my notes” gives you something you can verify.

Split the work between you and the assistant
TaskAI can help withYou decide
ResearchGrouping questions and pointing out gapsWhat the audience actually needs
Product choiceOrdering documented differencesWhich trade-off you want to recommend
ContentStructure, drafts and copy-editingFacts, experience and the final judgement
LinksCreating links via an approved connectionProgramme, destination and use
AnalysisCalculations and possible explanationsWhether the basis is good enough for a decision

02 / The method

Find the question before the product.

Choose a specific person and situation. “Air purifiers” is a topic. “A tenant who wants a quiet air purifier in a small bedroom” is a need. The latter helps you choose criteria, products and the right format for the content.

Gather your material first: questions from your audience, your own observations, the manufacturer’s specifications and the shop’s current product pages. Then ask the AI to sort the questions by what the reader has to decide.

  1. Describe the audience and its most important constraint.
  2. Find the questions that need an answer before buying.
  3. Note which answers you can already document.
  4. Choose an angle you can make more concrete than a generic product list.

Prompt 01 / Research

I am researching [niche] for [audience].
Here are my sources and my own observations:
[insert sources, dates and notes]

Find five concrete questions people ask before buying.
For each question: describe the reader's need, which information can answer it, and what we still need to find out.

Distinguish documented facts from hypotheses. Do not invent search volumes, prices or products. Suggest an angle I can add my own experience to.

03 / The method

Compare on what matters.

AI is useful for turning inconsistent product descriptions into a consistent table. Agree the criteria first. If you choose them after seeing the commission, you risk recommending based on your own earnings rather than the reader’s needs.

Keep specifications, the manufacturer’s claims and your own experience apart. A stated noise level is a specification. How the sound feels to you at night is experience. AI may help describe both, but it must not invent a test you have not carried out.

Fact

What we can document

Source, date and the exact measurement or specification.

Judgement

What it means for the reader

Your argument for why the difference matters in use.

Unknown

What needs investigating

A visible gap is better than a confident, made-up answer.

Prompt 02 / Product comparison

Compare these products for [audience and use case].
Only use the attached product data and my test notes:
[insert material with source and date]

Make a table with criterion, product A, product B and source. Write "unknown" where the material gives no answer.

Finish with: who does each product suit, what is the trade-off, and which questions still need investigating? Do not say we have tested anything unless it is in the notes.

Then choose a relevant affiliate programme. Look at the range, the programme’s rules, commission model and cookie window. A high rate is not in itself an argument that the product suits your audience.

04 / The method

Give AI facts. Give the reader value.

The strongest prompt contains more than tone and length. It says who the text helps, which question it answers, and which sources and experiences may be used. Feel free to write your own conclusion first, so the assistant helps explain your reasoning.

Work in separate passes: factual basis, outline, draft and check. When everything is ordered in one message, it becomes harder to see where an undocumented claim crept in.

Prompt 03 / A source-based draft

Write a draft [format] about [topic] for [audience].
The reader's main question is: [question].
My sources, experience and conclusion:
[insert your factual basis]

Start with a concrete answer. Explain the criteria, show the relevant differences, and describe the limitations. Mark claims that lack documentation with [CHECK].

Do not invent experiences, quotes, measurements or user reviews. Finish with a list of the facts I need to check before publishing.

Reuse the insight across channels

A thorough comparison can become a short video, an email and a post. Ask the AI to adapt the angle for each channel. A video has to make sense without the article’s table, and a short text still needs room for the most important limitation.

Use different sub-IDs for the links. Then you can later see which placement brought clicks and sales.

05 / The method

With Affilyflow’s MCP server you can connect your own ChatGPT or Claude conversation to your affiliate account. MCP is the connection that lets the assistant use specific tools. It gets access to read programmes, your websites, links, statistics and sales – and to create links.

Without the connection the assistant can help with a plan, but it does not automatically know your current programmes or results. A tracking link has to come from Affilyflow’s tool. It is not enough that the URL looks like an affiliate link.

ChatGPTClaude↔ Affilyflow MCP
  1. 01

    Choose the programme

    Ask the assistant to show the programmes you have joined and their terms.

  2. 02

    Give it the destination

    State the specific product page and a recognisable sub-ID.

  3. 03

    Check the result

    Open the returned link, and check the landing page before publishing.

Connect your assistant first – see the setup

Building your own solution?

Want to create links from your own publishing tool or pull numbers into a report? See the REST API with examples and access. Choose a well-defined task, and review the result before you automate more.

06 / The method

Ask for explanations. Test them.

AI can help you read your results and find questions worth investigating. Start with a consistent period and clearly defined fields. Do not mix approved commission with pending amounts when you want to compare what has actually been approved.

One simple measure is approved commission per click: approved commission divided by the number of clicks. It describes an observation in your data set. It does not prove that a particular headline or channel made the difference.

Worked example with fictional numbers · same period
PlacementClicksApproved commissionPer click
Guide / article120DKK 180DKK 1.50
Video / description80DKK 160DKK 2.00

In the example the video link has higher approved commission per click. The next question is whether the products, the age of the clicks and the audience are comparable. A single expensive sale can move a small data set a lot. So also look at the number of sales, rejections and pending sales.

Prompt 04 / Analysing your numbers

Analyse the attached aggregated numbers for [period].
Fields: source/sub-ID, clicks, sales, approved commission and pending commission.
[insert the table]

Calculate approved commission per click where clicks are greater than zero. Show the formula and keep pending amounts separate.

Find three observations. Distinguish observation from possible explanation. Allow for small data sets and differences in period or programme. Suggest one change at a time and the measurement that can show whether it helped.

07 / The method

Content that deserves to be found.

AI content is not an SEO strategy in itself. Google describes generative AI as a useful tool for research and structure, but warns against producing many pages without added value. The focus is accuracy, quality and relevance. Read Google’s guidance on generative AI.

For an affiliate site this means you must be able to point to the page’s own contribution: concrete experience, a useful comparison, clear criteria or a thorough explanation for a specific audience. Rewriting the shop’s product copy into more words is a thin contribution.

Give the page a descriptive title and a clear main heading. Answer the reader’s question early. Link to the sources that support the facts, and to the relevant pages on your own site. Say who is behind the assessment, and update information when it changes. That follows the direction of Google’s principles for helpful content.

No prompt guarantees a ranking in Google or a mention in an AI answer. Use Search Console and your own results to discover which questions your content is found for. Improve the content based on that insight and your readers’ feedback.

08 / The method

Start with one complete workflow.

Choose one specific piece of content, and follow it all the way from question to measurement. That way you learn where AI helps you, where you need to edit, and which information you are missing. Use this order for your first run-through:

  1. 01

    Define the task

    Choose an audience, one buying question and the channel you will use.

  2. 02

    Gather the sources

    Note sources and dates. Mark the information you are missing.

  3. 03

    Write the draft

    Use the research and content prompts. Add your own conclusion.

  4. 04

    Check and publish

    Check facts, advertising labels and destination. Use a separate sub-ID.

  5. 05

    Measure and adjust

    Compare a relevant period. Choose one improvement, and follow its result.

09 / The method

Questions about AI and affiliate.

What is affiliate marketing with AI?

It is ordinary affiliate marketing where AI helps with parts of the work: research, structure, drafts, links and analysis. The earnings still come from the actions the affiliate programme pays for. At Affilyflow those are sales tracked via your affiliate link.

Can ChatGPT and Claude create affiliate links?

An ordinary conversation does not automatically have access to your affiliate account. When you connect Affilyflow’s MCP server and approve the access, the assistant can find programmes and create real trackable links. It should use the URL the tool returns and never guess a tracking link itself.

Can I use AI-written text on an affiliate site?

Yes, AI can help with research and drafts. You still have to check the facts, add an independent assessment and make the content useful to the reader. Google emphasises the quality and value of content; mass-producing pages without added value can violate its spam policies.

Can AI automate my whole affiliate business?

You can automate well-defined tasks, but relevance, sources, programme terms and publishing still require responsible decisions. Affilyflow’s MCP access can read programmes, links and numbers, and create links. It gives no access to change bank details, profile or payouts.

Which AI tool should I choose?

Start with a tool you already use, and try the same task with your own sources. Assess factual accuracy, the ability to work with your data and how much editing the answer needs. Affilyflow can be connected to ChatGPT and Claude; you do not need to switch model to follow the methods in this guide.

Does AI increase affiliate earnings?

There is no guarantee. AI can cut repetitive work, but the result depends on the audience, the quality of the content, traffic, product choice and the programme’s terms. Measure changes with comparable data, and keep pending commission separate from approved commission.

This guide was written by Affilyflow. Methods and worked examples are indicative; they are not evidence of any particular level of earnings. The product features are described on the basis of Affilyflow’s MCP and REST integration.

ChatGPT is a trademark of OpenAI, and Claude belongs to Anthropic. Affilyflow is not affiliated with or endorsed by OpenAI or Anthropic.

From knowledge to your next link

Your knowledge. Your next connection.

Find a programme that suits your audience, and connect your assistant when you want to bring your links and results together.