Why ChatGPT and Claude Keep Disappointing You — It's Your Setup

Source: The AI Corner · Author: Ruben Dominguez · Date: Apr 15, 2026 · Original article

Note: This article is mostly paywalled. The full ChatGPT and Claude walkthroughs are gated behind a paid Substack subscription. The summary below covers the freely visible portion — the framing argument, the feature inventory, and the table of contents for the locked guide. The detailed step-by-step setups themselves are not accessible.

---

The core argument: it's not the prompt, it's the setup

The author opens with a blunt diagnosis of why most people feel underwhelmed by ChatGPT and Claude. The mental model to take away is this:

Every new chat session starts from zero. The model has no memory of who you are, what you're building, how you write, or what you decided yesterday. So it falls back to a generic, average-internet-person response. You then spend energy reshaping that output into something usable — and the moment you open a new chat, the cycle starts over.

The people who report dramatically better results aren't smarter and aren't using cleverer prompts. They invested once in configuring the tool — telling it persistently who they are, what context to load, and how to behave — and then they harvest the returns from that one-time setup every single day.

The thesis in one line, quoted from the piece:

"The gap between a basic user and a power user is not intelligence or technical skill. It is knowing which features exist and how to activate them."

The hidden features most users never turn on

Both ChatGPT and Claude ship with a layer of "power features" that the average user never discovers. The article lists them as the building blocks of a real setup:

• Projects — persistent workspaces. Context (files, instructions, prior chats) accumulates inside the project, so every new session inside it already knows the background.
• Memory — cross-session recall. The tool remembers facts about you between chats, so you stop re-explaining your job, your stack, or your preferences.
• Custom Instructions — a rules file that runs automatically before every conversation. Think of it as a system prompt you write once.
• Cowork (Claude desktop) — gives the model direct access to local files on your machine without uploading them one by one.
• Skills and Plugins — reusable instruction packs that auto-activate when the conversation matches their topic. (E.g., a "code review" skill that kicks in only when you paste code.)
• Connectors — live links to Slack, Drive, Notion, Gmail, and 50+ other services, so the model can read fresh data instead of working from a stale paste.
• Scheduled tasks — recurring workflows that run on a timer, even when you're offline.

The author's framing: turning these on transforms the tool from "chatbot you wrestle with" into "colleague who already knows your priorities, your voice, and your current projects before you type a single word."

The numbers behind the claim

Two data points the article uses to motivate the upgrade:

• ChatGPT has over 900 million weekly users, but the vast majority are on the free tier. The free tier lacks the advanced reasoning/planning model. On performance benchmarks, the paid "thinking" model is tied for #1, while the free version ranks #25. The implication: most people are judging ChatGPT by a model that is roughly 24 places worse than what's available.
• Claude has the same shape of gap: browser-tab Claude delivers ~20% of what the tool can do. The other ~80% lives in Cowork on the desktop app.

Author's conclusion from those numbers: the variable that changes everything is the setup, not the prompts. You can get dramatically better output without rewriting a single prompt — you just have to configure the tool first.

What the (paywalled) full guide promises

The rest of the article is locked, but the table of contents is visible. Recording it here so you know what the complete piece covers:

The complete ChatGPT system

• Custom Instructions templates ready to copy
• Memory setup so you stop re-explaining yourself in every chat
• Projects structure with dedicated instructions and uploaded reference files
• Canvas walkthrough for iterative document editing
• Custom GPT builder guide — your voice, your rules, shareable with your team
• A "self-critique" prompt that forces GPT to grade and rewrite its own output

The complete Claude system

• A four-folder Cowork structure that auto-loads your full context
• Two core files — about-me.md and anti-ai-style.md — that the author claims replace ~50 individual prompts (the idea: one persistent file describing you, plus one persistent file describing the writing/output styles you don't want, eliminates the need to keep saying "don't be salesy / don't use bullet lists / etc." in every chat)
• Global Instructions that run before every session, forever
• Skills, Plugins, and Connectors setup
• Dispatch — queueing tasks from your phone so your desktop machine executes them
• Scheduled tasks that run overnight unattended

Plus

• A side-by-side breakdown of when to use ChatGPT vs. Claude for each job
• The one prompt that works identically in both tools
• A correction method that fixes wrong output without rewriting the original prompt
• A Q&A covering the most common reader questions

Takeaways you can act on without paying

Even without the locked walkthrough, the freely visible portion gives you a usable checklist:

1. Audit which tier you're on. If you're on free ChatGPT, you're likely benchmarking the product against its 25th-place model. The argument for upgrading is less about "more messages" and more about "access to the actually-good model."
2. Turn on Memory and Custom Instructions today. These are the lowest-effort, highest-leverage features. Write down — once — who you are, what you're working on, how you want responses formatted, and what you don't want (no hedging, no bullet vomit, no sales tone, etc.).
3. Move Claude work to the desktop app. Per the article, the browser tab is ~20% of the product. Cowork (local file access) is where the depth lives.
4. Set up Projects per ongoing initiative. Instead of one giant chat history, a Project per workstream means context compounds in the right bucket.
5. The "two-file" Claude pattern is worth stealing even without the full guide: an about-me.md (who you are, what you do, what you're building) and an anti-ai-style.md (the tics and tones you want suppressed). Drop both into a Cowork folder so every session loads them automatically.

The unifying principle across all of it: stop paying the setup tax at the start of every chat. Pay it once, persistently, and let the tool carry the context for you.