9 AI Assistants to Build Inside Your Agency

What is an AI assistant, and how is it different from a GPT?

Think of the personal assistant an executive once had: someone who booked calls and meetings and took on repetitive, low value tasks so the boss could focus on the work only they could do. AI assistants do the same for your team, except you can have many of them, each a specialist in one subject, available at any hour.

A custom GPT and an assistant are the same thing hosted on different platforms. A GPT lives inside ChatGPT. An assistant lives on the back end of OpenAI's platform and can be used inside automations. Anything you can build as one, you can build as the other.

How do you create an assistant faster?

Writing good instructions from scratch takes a while. A quicker route is to use the GPT builder first and reuse its output.

1. In ChatGPT, create a new GPT and answer the builder's questions about what it should do.
2. When the builder has generated instructions, copy them.
3. In the OpenAI API platform, open the assistants section and create a new assistant.
4. Paste the copied instructions into the assistant's instructions field.

What chat assistants should an agency build?

Chat assistants are conversational. You go back and forth with them as you would in ChatGPT.

1. A sales objection assistant. New closers can practise objections without wasting leads you paid for, and experienced closers can warm up before calls. We host this as a GPT so you can use voice mode on your phone. Build it by telling the GPT builder about your product and customers, then upload a PDF of your most common objections as knowledge.

2. An SOP assistant. Put every standard operating procedure into one document, export it as a PDF and upload it to an assistant using file search. Host it in a Slack channel, and your team can ask about any procedure without waiting for you or an admin to answer.

3. An IT expert. Create one assistant per software tool you use, for example Google Sheets, Zapier or GoHighLevel, each specialised in that tool. Add them to an IT channel and message the relevant assistant when you hit a problem. A Google Sheets assistant that knows formulas gives better answers than a general chatbot, because it already knows what you are asking about.

What automation assistants should an agency build?

Automation assistants are called from inside automations, for example in Make.com. A side by side test of the same question gave a far more specific answer from an assistant than from a general model prompt, which is why we use assistants in automations rather than plain prompts.

4. A lead magnet assistant. It delivers value to leads automatically. Upload a PDF containing your offer, your processes and what you have learned in your field. The answers it gives are then based on your own experience, not generic output anyone could get from ChatGPT.

5. A content assistant. Put your past scripts and their topics into one PDF and attach it to an assistant. In a Notion content system, changing a video's status to ready for edit can trigger the assistant to draft an introduction in your style. It is a first draft, not a finished script, but it beats starting from nothing. The same idea works for TikToks and newsletters.

6. A data analyst. Once a month, an automation pulls client results, and the assistant compares them to your KPIs, processes and past results. It then produces an action plan. Upload your KPIs and past results as a PDF so it knows what good looks like. This is one step in a larger feedback loop.

What can function assistants do?

Function assistants do things rather than only answer. They can be used in automations or in chat. Setting them up properly needs some light coding, so this section is about what is possible.

7. A live data assistant. A simple example is a weather assistant that fetches the weather for a place. It is not something to build for an agency, but it shows that an assistant can be connected to live data such as stock prices or news.

8. A reschedule assistant. In a Slack channel, you send a message saying you want to reschedule a meeting, including the new time, the type of meeting and the participant's email. The function runs an automation that reschedules it, and the participant gets an email.

9. A customer assistant. Without functions, you would have to keep uploading again a PDF of customer information, and it would go out of date. With a function, the assistant looks up a spreadsheet or database each time. You can message it for a client's current figure and it answers from live data. Customer success managers can find client information quickly without opening several tools.

Where should you host your assistants?

It matters less than you think. You can host them as a GPT, in Slack, on your phone or as a chatbot inside your CRM. What counts is that they save your team time.

If you use AI responses in any automation, use an assistant with your own files in place of a generic prompt. Giving your customer success team assistants like these also helps them handle more clients without burning out.