Koka Sexton

---

Projects vs Custom GPTs: Know When to Use Which

The most common mistake I see: people build a Custom GPT when they should have built a Project, and vice versa. Here’s the decision framework:

Use a Project when…	Use a Custom GPT when…
The work evolves across sessions and needs memory	The task is repeatable with known inputs/outputs
You’re uploading proprietary docs, templates, or data	You want to share the assistant with a team
You need to iterate and refine over time	The instructions are static and don’t change much
The context is private (client work, internal strategy)	The assistant can be public or semi-public
You need Canvas, code execution, or multimodal features	You want API or Zapier/Make integration

Real example: If you’re building a “LinkedIn Post Generator” that takes a topic and outputs formatted posts, that’s a Custom GPT. If you’re running ongoing content strategy for a B2B SaaS client with 15 uploaded docs and weekly iterations, that’s a Project.

---

8 Best Practices for ChatGPT Projects in 2026

1. One Project Per Client or Campaign, Not Per Topic

Resist the urge to create a project for every idea. Create one per outcome you’re responsible for:

• ✅ “Client X: Full Marketing System” , contains their brand guidelines, campaign templates, content calendar, and messaging framework
• ✅ “Q3 Demand Gen Campaign” , contains ad creative, landing page copy, email sequences, and performance data
• ❌ “Blog Post Ideas” , too narrow. This should be inside a broader project.

Why this matters: When all related context lives in one project, ChatGPT’s responses are consistently informed by your entire strategy , not just the last thing you uploaded. The model can reference your brand guidelines while drafting a landing page because both live in the same workspace.

2. Custom Instructions Are Your Operating System

The Custom Instructions field in a Project is the single highest-leverage input you control. Don’t waste it on vague personality prompts. Be specific about output format, constraints, and reference materials.

Weak instruction: “Be helpful and professional.”

Strong instruction:

“This project builds B2B marketing assets for [Client Name]. Use the uploaded Brand Guidelines.pdf for tone and voice. All copy should target VP/C-level buyers at companies with 200-2,000 employees. Default formats: blog posts use Introduction → Problem → Framework → Tactical Steps → Conclusion. Landing pages use Hero → Problem → Solution → Social Proof → CTA. Always include specific data points and internal links where relevant. Flag any claims that need source verification.”

The difference in output quality between these two approaches is dramatic. The model with strong instructions produces work that actually sounds like you. The one with weak instructions produces work that sounds like ChatGPT.

3. Upload Strategically , Only What the Model Actually Needs

ChatGPT Projects in 2026 handle files better than ever, but a cluttered project is a confused project. Upload what the model needs to reference, not your entire Google Drive.

What belongs in a Project:

• Brand/tone guidelines (1 doc)
• Content templates or frameworks (1-3 docs)
• Example outputs that represent your standard (3-5 examples)
• Data or research the model should cite (1-2 docs)
• Messaging frameworks or positioning docs

What doesn’t: every version of every draft, raw meeting notes, competitor PDFs you’ll never reference, duplicate files with slightly different names.

Naming convention matters more than you think. The model uses filenames to understand what to reference: “Brand-Voice-Guide.pdf” is instantly understood. “final_v6_Koka_edits_FINAL.pdf” is not.

4. Stack Prompts Instead of Megaprompting

The “megaprompt” , one massive instruction that tries to cover everything , produces mediocre output because the model can’t prioritize what matters. Stack your prompts instead: give one clear instruction, review the output, then build on it.

1. “Draft an outline for a B2B landing page on AI-powered lead scoring, using the messaging framework from uploaded Positioning.pdf.”
2. “Good. Now write the hero section with a headline that follows the Specific Result hook pattern from Content-Hooks-Playbook.pdf.”
3. “Tighten the social proof section. Add 2-3 specific stat cards that a CMO would find credible.”
4. “Now add inline FAQ schema markup and a CTA that ties to our demand gen offer.”

Each step is specific. Each step builds on the last. The model stays grounded in your actual project context instead of guessing.

5. Version Everything , The Model Won’t Remind You

ChatGPT Projects remember your conversation history, but you won’t remember what changed three weeks ago. Build a lightweight versioning habit:

• When you update a file in the project, rename it with a date: “Brand-Guidelines-2026-05.pdf”
• When you make a significant strategy shift, note it in the chat: “Updated 5/21: We’re moving from awareness content to conversion-focused content for Q3.” The model will carry this forward.
• Export and save milestone outputs , the landing page that converted well, the email sequence that outperformed , so you can re-upload them as reference assets later.

This turns each project into a compounding knowledge base, not just a chat log.

6. Connect Projects to Real Workflows (Not Just Chat)

The power move: use Projects as the thinking layer, then pipe outputs into your actual tool stack. A Project can design an automation sequence that you build in Make.com. It can draft content that you finalize in your CMS. It can analyze campaign data that you export from your CRM.

B2B marketing workflow example:

• Upload your ICP definitions, messaging frameworks, and the last quarter’s campaign performance data to a Project
• Have the Project analyze what messaging resonated with which segments
• Use that analysis to brief your next content sprint in your actual project management tool (ClickUp, Notion, whatever)
• The Project becomes the strategy layer. Your tools become the execution layer.

This is fundamentally different from using ChatGPT as a one-off writing assistant. The Project holds the context. The context improves every output. Over time, the quality compounds.

7. Archive Religiously

When a campaign ends or a client engagement wraps, don’t let the project rot. Export the best outputs , the messaging that worked, the frameworks you built, the templates that became your standard , then archive the project.

Archiving keeps your active workspace clean and your context focused. It also forces you to identify what was actually worth keeping. Most projects produce 80% filler and 20% gold. Archiving is the filter.

8. Treat Project Memory as Compound Interest

This is the mindset shift that separates power users from everyone else: every interaction in a Project either improves or degrades the model’s understanding of your work. Garbage inputs compound into garbage outputs. Clear, intentional inputs compound into increasingly precise and valuable outputs.

When you upload a messy, contradictory set of documents, the model produces messy, contradictory work. When you curate what goes in , clean guidelines, specific examples, clear instructions , the model’s output quality improves week over week. It learns your voice. It internalizes your frameworks. It starts anticipating your preferences.

This is the feature most people never discover because they treat Projects as labeled chat tabs instead of compounding workspaces.

---

A Real B2B Marketing Consultant’s Project Setup

Here’s how I personally structure Projects for ongoing work:

Project Name	Files Uploaded	Custom Instructions (abbreviated)	Use Case
KSB Content Engine	Brand voice guide, 5 top-performing article examples, content templates, SEO keyword map	“Write in Koka’s voice. Use stat cards and comparison tables. Target B2B marketing leaders. Every article needs a TL;DR, framework section, and CTA.”	All blog posts, LinkedIn posts, newsletter content for kokasexton.com
Client: [SaaS Co] GTM	Client brand guide, ICP doc, competitive landscape, 3 winning landing pages	“Target VP Demand Gen at B2B SaaS companies 50-500 employees. Use Challenger Sale framing. Every asset needs a specific proof point.”	Landing pages, email sequences, sales enablement, ad creative
LinkedIn Growth System	Top-performing post examples, comment templates, engagement playbook, content calendar template	“Output formatted LinkedIn posts with hook patterns. Generate comment engagement scripts. Suggest 3 post variants per topic.”	Daily LinkedIn content, engagement strategy, DMs/outreach scripts

Three projects. Each one compounds. The KSB Content Engine has absorbed my voice to the point where I spend less time editing and more time publishing. The LinkedIn Growth System knows which hook patterns perform best for my audience. These aren’t chat threads , they’re operational infrastructure.

---

Common Mistakes That Kill Project Performance

• The Kitchen Sink Upload: Dumping 47 files into a project because “the model might need them.” It won’t. It’ll just get confused. Curate what you upload.
• Vague Custom Instructions: “Be a helpful marketing assistant” is useless. The model already defaults to helpful. Tell it what formats, what frameworks, what voice, what constraints.
• Never Archiving: 15 active projects, 12 of them for clients you haven’t worked with in 6 months. The model’s context window isn’t unlimited , noise in your project list is noise in your outputs.
• Treating Projects as Draft Dumps: Uploading every rough draft and expecting the model to magically organize them. Projects are workspaces, not junk drawers. Clean inputs only.
• Ignoring File Names: When files are named “doc_final_v3.pdf” and “notes.pdf,” the model has no way to know which is your brand guide and which is meeting notes from 2024. Descriptive filenames aren’t just for humans anymore.

---

What Changed in ChatGPT Projects in 2026

If you set up your Projects in 2025 and haven’t revisited them, here’s what’s new:

• GPT-5 model access: Projects now support GPT-5, which means significantly better context retention, more nuanced instruction following, and improved handling of complex multi-step workflows. If you’re still running Projects on GPT-4o, you’re leaving performance on the table.
• Canvas integration: You can now open a Canvas inside a Project, and the Canvas has access to all project files and context. This is a game-changer for editing long-form content, reviewing data, or iterating on visual layouts without losing project context.
• Improved file parsing: PDFs, spreadsheets, and documents are parsed more accurately , the model now extracts and retains structured data (tables, lists, formatting) instead of flattening everything to plain text.
• Longer memory persistence: Custom instructions and uploaded file context persist more reliably across long sessions. The “what was I working on?” friction is noticeably reduced.
• Multimodal within Projects: Image generation and analysis work directly within project context, so visual assets can reference your brand guidelines just like text outputs can.

---

The Bottom Line

ChatGPT Projects are not a filing system. They’re not “folders for chats.” They’re the closest thing ChatGPT has to a long-term memory and operating system , and most users are leaving 80% of that value on the table.

The difference between a well-structured Project and a messy one isn’t marginal. It’s the difference between an assistant that produces generic content and an assistant that produces work in your voice, following your frameworks, referencing your data, week after week.

Start with one Project. Set it up right. Use it for a month. Then look at the outputs from week 1 vs week 4. The compounding will be obvious.

Building AI-powered marketing systems? Let’s talk about your stack.