Tools & Resources
Supplementary knowledge beyond the lesson plans, model recommendations, prompt patterns, and reference material.
Which model should I use?
Custom Instructions / Personal Brief
Every AI tool lets you set persistent instructions that shape how it responds. Think of it as a brief for the AI — your preferences for tone, format, and behaviour baked in so you don't repeat yourself.
Here's Trent's as an example:
Trent's Generic Custom Instructions
## Context
- Based in Sydney, Australia.
- Head of Design at Howatson+Company, a full-service creative agency. Leads 30+ designers across craft, finished art, and digital.
- Interests: graphic design, advertising, brand identity, creative concepts, AI, automation, technology, Apple, podcasts, photography, leadership, productivity, politics, current affairs, documentaries.
## Response Style
- Use British English. Direct, conversational, human tone.
- Omit greetings, filler, flattery. Begin with the answer.
- Vary sentence length; avoid hedging.
- If a request is flawed or based on a false premise, state this directly before proceeding.
## Scope
- Fulfil only what is asked. No extras or embellishments.
- If ambiguous, select the simplest valid interpretation.
## Format
- Use bullets and tables only if it aids clarity.
- For reworks or edits, return only revised content.
- Enclose any requested prompt in a fenced code block without extra commentary.
## Uncertainty
- If uncertain, say so clearly. Do not make up figures or references.
- State any assumptions explicitly.
## Clarification
- Request clarification only when essential; otherwise, make a reasonable assumption and state it.
## Image Prompts
- Prompts must be standalone and model-agnostic; omit --parameters unless requested.
- Specify: subject, environment, lighting, camera model, lens/angle, framing, texture, photography artefacts, effects, add imperfections.
- Describe character/identity and detail clothing if possible.
- When iterating, retain subject unless instructed otherwise.
## Takeaways
- If helpful and not a rework, end with one concise takeaway.
Prompting Techniques
Prompt Chaining
Use the output from one step as the input for the next. Build momentum across a workflow instead of starting fresh every time.
Session 1
Role Prompting
Tell the model who to be before you ask your question. A strategist, a critic, a consumer. Same question, different expertise, different answer.
Session 1
Zero-Shot Prompting
Ask with no examples and no context. This is your baseline. Useful for seeing what the model gives you before you start shaping it.
Session 2
Few-Shot Prompting
Show the model what good looks like before you ask. Drop in two or three examples of the tone, format, or quality you want and it will match them.
Sessions 2 & 3
Chain-of-Thought Prompting
Ask the model to think step by step before it answers. Slower, but the reasoning is visible and the output is sharper on complex questions.
Session 2
Meta-Prompting
Use AI to write the prompt that then briefs AI. Your brand strategy becomes a visual brief becomes an image prompt. Each layer builds on the last.
Session 2
Vibe Coding
Build software by describing what you want in plain language. No code. You direct, the model builds. You learn where it's fast and where it needs a developer.
Session 4
Multi-Model Comparison
Run the same prompt across Claude, Gemini, and ChatGPT at once. Compare the results side by side and pick the best. Over time you'll learn which model suits which task.
Session 2
Frequently Asked Questions
How AI Works
Every conversation has a context window, which is basically the model's short-term memory. The longer a chat goes, the more cluttered that memory gets. Starting fresh means the model is focused entirely on what you're asking right now, with no leftover noise from earlier topics.
It's everything the model can "see" at once: your messages, its replies, any documents you've pasted in. There's a limit. When a conversation gets long, the oldest parts start falling out of view. That's why long, sprawling chats eventually get worse, not better.
They're all very capable right now. Trent's daily driver is Claude. Hoang's is Gemini. It comes down to personal preference. Try a few, see which one clicks with how you think, and stick with it. You can always switch later.
Claude, Gemini, and ChatGPT are large language models. They handle writing, strategy, analysis, and conversation. Nano Banana Pro is an image generation model. It makes pictures. You'll use both, but for different things.
Yes. In H+Co ChatGPT, click the plus icon at the top of the chat to add extra models — up to three at once. You'll see all their responses side by side. It's useful for comparing writing quality, but it takes more cognitive load to evaluate. Over time, you'll learn which model you prefer for which task.
Markdown is a simple formatting language. Hashtags are headings, asterisks are bold, and dashes are bullet points. All AI models speak markdown natively — it's how they structure information. When you see formatted output with headings and bold text, that's markdown being rendered. It's also why converting PDFs to plain text or markdown before uploading gives the model an easier time reading your documents.
Getting Better Results
Iterate and chat. AI works best as a conversation, not a single command. Give it a go, look at what comes back, tell it what to change, and go again. Two or three rounds is normal. Five is fine.
Tell the model what you're trying to achieve and ask it to ask you questions. Say something like: "I want to write a brand strategy for a new sunscreen brand. Before you start, ask me the questions you'd need answered to do this well." Let it interview you. The output will be dramatically better.
More specific than you think. "Write me a tagline" will give you something generic. "Write me a tagline for a premium Australian sunscreen brand that targets adults who wear SPF daily and treats sun protection as a normal part of getting dressed, not a medical warning" will give you something you can actually use. Details in, quality out.
Yes. Gemini and ChatGPT are best at reading long documents, but all the models handle big inputs well. Paste in research reports, briefs, competitor copy, anything relevant. Just tell the model what to do with it.
Yes. Just try it. Drop an image into the chat and ask the model to describe it, analyse it, pull text from it, or use it as a reference. Most people don't realise this works until they try it.
It's not required — all the models handle PDFs well. But if you want the best results, converting to plain text or markdown first gives the model a cleaner input to work with. It removes formatting artefacts, headers, footers, and other noise that can confuse things. For quick questions, just upload the PDF. For important work where accuracy matters, convert it.
It will. Treat it like a smart colleague who sometimes gets the brief wrong. Correct it, redirect it, give it more context. Don't start over unless the conversation has gone completely sideways. A simple "That's not quite right. Here's what I meant..." usually fixes it.
Start a new chat and try again. Long conversations drift, and sometimes the model gets stuck in a loop. If it's a specific problem rather than a general vibe issue, tell the model what you don't like. "That's too formal" or "Stop giving me lists, write it as a paragraph" works.
Click the + icon next to "Chats" in the left sidebar and select "New Folder". Name it after your client or brand. Once it's created, you can drag existing chats into it or start new ones inside it.
To add knowledge, go to Notes and create a new note — paste in your brand strategy or any key document. Then edit the folder, select Knowledge, and attach the note. Every chat in that folder will now reference those documents automatically.
You can also add a system prompt to the folder. Edit the folder, paste in your system prompt, and hit save. This acts as persistent instructions — like a brief that gets prepended to every message in that folder. Include something like "always reference the knowledge documents in this folder before responding" so the model knows to check your docs first.
When the model references your documents, it will show you which section it's pulling from. Click the reference to verify accuracy.
To add knowledge, go to Notes and create a new note — paste in your brand strategy or any key document. Then edit the folder, select Knowledge, and attach the note. Every chat in that folder will now reference those documents automatically.
You can also add a system prompt to the folder. Edit the folder, paste in your system prompt, and hit save. This acts as persistent instructions — like a brief that gets prepended to every message in that folder. Include something like "always reference the knowledge documents in this folder before responding" so the model knows to check your docs first.
When the model references your documents, it will show you which section it's pulling from. Click the reference to verify accuracy.
Don't. Get better at asking AI to write your prompts. Instead of spending twenty minutes crafting the perfect prompt, tell the model what you're trying to achieve and ask it to write the prompt for you. It's faster, and the model knows what it needs better than you do. The same applies to system prompts and custom instructions — ask the model to generate them based on your context, then iterate.