AI for Marketing

A workflow is a sequence of steps

See how a job is broken into steps, then where AI can help.

The concept

A workflow is a repeatable sequence of steps. It is like a recipe: the steps describe what needs to happen and in what order. Publishing an article, for example, involves writing a draft, checking it, making changes and publishing it. Together, these steps form a workflow.

The problem it solves

When a job has several steps, it is easy to miss one. An article might be published before someone checks its facts. A clear workflow makes that check part of the job.

Worked example

Mina writes an article about taking product photos. She then asks AI to check whether the steps are easy to follow. It points out that the article mentions “showing scale” without explaining what that means. Mina adds an example: placing a familiar object beside the product to show its size.

The editor checks the revised article and publishes it. AI helped with one step: reviewing the draft. Mina still chose which suggestion to use, and the editor decided when the article was ready.

The same idea applies to other jobs. Understanding the steps makes it easier to see where help would be useful.

Context: give AI the background

See why background information helps AI write a more suitable answer.

The concept

Context is the background information AI has when answering your request. Think of asking a colleague to write an email. They will do a better job if they know who it is for, what it should say and how your business usually writes.

The problem it solves

“Write a launch email for our mug” leaves those questions unanswered. AI has to make choices about the customer, the product and the tone. The email can sound good while being wrong for the business.

Worked example

Fieldwork Home gives AI three pieces of background. The product details say the mug is blue stoneware, holds 300 ml and is dishwasher safe. The customer description says people want clear information about everyday products. A previous email shows the brand's simple, direct writing style.

The product details tell AI what it can say about the mug. The customer description helps it choose what to focus on. The previous email shows how the writing should sound. The next page shows a short email and explains those choices.

More information helps when it answers a question about the task. Adding unrelated files can make things less clear.

Why this email reads differently

Follow the connection between supplied context and specific writing choices.

The request

Fieldwork Home wants a short email introducing its blue stoneware mug. The email is for people choosing everyday homeware. It should use simple language and invite them to view the mug on the website.

An example email

Subject: Meet our blue stoneware mug

A blue stoneware mug for your everyday coffee or tea. It holds 300 ml and is dishwasher safe. See the mug and its details on our website.

Why the email says this

The subject tells the reader which product the email introduces. “Everyday coffee or tea” connects it to the customer's ordinary routine. The size and washing advice come from the product details.

The short sentences follow the brand's writing example. The last sentence tells the reader what to do next, as requested.

Web and desktop: what can AI reach?

See why different apps can work with different files.

The concept

ChatGPT and Claude are applications: the software you open to talk to AI. A model is the system inside that produces an answer. The application decides which files and tools the model can use.

The problem it solves

Sometimes the information needed for a task is in several files on a computer. Copying those files into a chat, then saving every answer, can be awkward.

Worked example

For one brief, Mina can upload the brief and writing guidelines into a web chat. For a folder of briefs, a supported desktop tool such as Claude Cowork can read a folder she chooses to share. It can then save review files there. [1], [2]

The difference is like giving a colleague one document or letting them use an agreed project folder. Both can provide useful information. The folder can make repeated work easier.

Installing a desktop app does not give it permission to read the whole computer. Being on the desktop also does not make the model smarter. The useful question is which files and actions the application can use, with the permissions the owner has allowed.

Connectors: a link to another service

Understand how AI gets access to information outside the chat.

The concept

A connector is a link that lets an AI application use information or actions in another service. For example, an Asana connector can let it read tasks in an Asana project. MCP, short for Model Context Protocol, is one standard used to make these links.

The problem it solves

AI may know what Asana is without being able to see your project. It is like a colleague knowing how a filing cabinet works but not having access to yours.

Worked example

The campaign owner connects the approved Asana account. The AI application can now ask for the project's tasks. It receives details such as each task's name, owner, date and link. It can use those details to write a project summary. [6]

The connection saves the owner from copying the task details into the chat. It does not give unlimited access. Being allowed to read a task and being allowed to change it are separate permissions. The connector must also offer the action being requested.

If there is no suitable connector, a copy of the project information may be enough for a one-off summary.

Skills: a method you can reuse

See how saving instructions makes a repeated job easier to explain.

The concept

A skill is a saved set of instructions for a particular job. It can also include reference files. Think of the checklist an editor uses for every new article: the article changes, but the checks stay much the same. [9]

The problem it solves

Repeating the whole review method in every chat takes effort. It is easy to forget a rule that mattered last time.

Worked example

Fieldwork Home saves a brief-review skill. It tells AI to check the writing against the brand guidelines, check product claims against approved facts, and look for missing details such as the launch date.

When Mina supplies a new brief, AI can follow those saved instructions. For example, a missing launch date becomes a question for the owner. A product claim without supporting facts is flagged for checking. Each finding comes from a rule or a fact.

A saved skill needs an application that supports it. For a simpler task, a saved prompt may be enough. The checklist comparison also has a limit: a skill does not make AI infallible. If the product facts change, the saved reference files need updating too.

Where these ideas lead next

The complete book follows the same pattern throughout.

The £49 book explains content writing, product-copy reviews and campaign reports. Each example includes the information, the answer and the reasons behind it. The book can be understood by reading alone. Templates and a sample skill are optional extras.

The people, business details and email in this sample are made up to explain the ideas. AI tools may produce different answers.

References used in this sample

[1] OpenAI: Apps in ChatGPT
Open official guide

[2] Claude: Choosing Cowork or Chat
Open official guide

[6] Asana: Claude integration
Open official guide

[9] Claude: Create custom skills
Open official guide

References checked on 5 September 2026. Features and permissions vary by product and plan. ABZN is an independent publisher.
Read more at abzn.io