Video summary
إزاي تخلي ChatGPT يشتغل "صبي" لـ Claude؟ (وفر فلوسك ووقتك) | BnSamy
Main summary
Key takeaways
Overview
The video explains a workflow to reduce cost and avoid token/usage limits when using ChatGPT / Claude / Gemini. The approach is to use ChatGPT as a front-end “friend” to craft prompts and generate structured output, then send that output to Claude (or similar models) for the final answer.
Core problem (why people move between models)
- Many users rely on “Cloud” AI (e.g., ChatGPT/Claude) for help, but it comes with a major drawback: usage/token limits.
- When requests are complex (e.g., market research), the model can hit limits quickly, forcing you to wait hours (e.g., around 5-hour limits) and try again later.
- Even paying for Pro may not fully solve the problem, and some users may end up considering additional paid tiers.
Main workaround / technique (“small trick”)
1) Use ChatGPT as a guided prompt compiler (“friendly friend”)
Instead of using ChatGPT only for answers, the speaker proposes using it to:
- interpret what you want,
- ask follow-up questions,
- and help refine the request before sending anything expensive to Claude.
Requested behavior from ChatGPT
- Ask it not to reply immediately while you send a long “batch” of instructions.
- After you finish, have it ask five questions one-by-one so you can verify understanding.
Verification step
- Answer each question.
- Then ask it: “So what did you understand?”
- Use this to ensure the request is interpreted correctly before you spend Claude tokens.
2) Get output in English (Claude reliability claim)
The speaker claims Claude performs better in English, even if it can understand Arabic.
Practical instruction
- Ask ChatGPT to produce the deliverable as an English “printout/brochure.”
- Then translate/adapt to Arabic afterward as needed (or ask Claude to output Arabic later).
3) Transfer results: Gemini → Claude (two-step refinement)
Example flow
- Use Gemini (or ChatGPT) to analyze and rewrite/improve your brochure/market research brief.
- Send Gemini’s improved output to Claude.
Claimed outcome
- ~90% chance Claude returns the desired result correctly on the first try.
- ~10% chance you may need one extra loop (send Claude the adjusted Gemini response).
4) Iterate using targeted flaw-finding questions
To reduce wasted retries, instruct the model to:
- identify flaws in the current draft,
- suggest improvements,
- and generate a new version.
This reframes iteration as a structured improvement cycle rather than repeated “just do it” attempts.
Cost/time strategy (reduce number of attempts)
- For simple questions, a few chat turns can be fine—token usage stays low.
- For heavy tasks (e.g., market research, “deep research,” competitor analysis, Shopify tech audits, design creation):
- one task may consume 20–40% of your limit (or more),
- and if corrections are needed after an incorrect result, you can hit the limit and then must wait.
Recommended approach
- Minimize “first try failures” by doing:
- prompt preparation + understanding checks with ChatGPT,
- structured output + refinement,
- then one (or very few) Claude calls.
What the speaker claims you gain
- Potentially avoid needing prompt engineering expertise
- A “99% done” workflow for typical use cases
- Use only the remaining 1% of careful prompting for highly professional contexts
Review / guide / tutorial cues
This is primarily a tutorial/guide that teaches a:
- Multi-model prompting pipeline: ChatGPT (“Charge BT”) → Gemini → Claude
- Prompt discipline method:
- send a batch,
- no immediate reply,
- five questions,
- validate understanding
- Language strategy: request English output for Claude reliability
- Iteration strategy: avoid token-limit downtime by reducing expensive retries
Main speakers or sources
- Main speaker: Mohamed (the video creator; referred to in the transcript as “Mohamed/BnSamy”)
- AI systems referenced: ChatGPT (called “Charge BT”), Claude, Gemini
- plus mentions of other assistants/models (e.g., Perplexity, OpenAI/Deep Research style features)