How to Get Consistently Better Results from AI

TL;DR

Better AI results come from matching each task to the strongest available tool, supplying context through an AI-led interview, and dividing complex projects into planned stages. Refine initial outputs, test uncertain ideas, compare answers across multiple chatbots, use source-grounded tools when accuracy matters, and begin a new chat when the current conversation becomes stuck in a loop.

Transcript

The fact is this, 99% of people are using AI the wrong way and even worse, most people don't even know the potential that they are missing out on. And so in this video, I'm going to show you the way that power users like myself use AI differently so that we can get the very best outputs and the very best results. And the first thing that the top 1%... Read More

Key Insights

• Custom AI workflows assign each part of a project to the tool best suited to that task. The transcript recommends combining tools because Claude is presented as strong in coding and writing, while Gemini offers music, image, and video generation capabilities.
• An AI-led interview supplies context before generation begins. Asking the chatbot to question the user can uncover goals, preferences, and requirements, which is especially helpful when the user lacks enough subject expertise to write a detailed prompt independently.
• Complex AI tasks become more manageable when divided into planned stages. The novel example separates ending ideas, character profiles, storyline construction using a Save the Cat structure, and chapter drafting instead of asking the AI to produce a complete result immediately.
• The first AI response should be treated as a draft rather than a finished product. The podcast-image example improved through separate requests involving composition, camera angle, object removal, microphone placement, and headphones, with the user acting as the creative director.
• AI experimentation reveals capabilities that users may otherwise underestimate. The learn-to-draw example began without certainty that the method would work, then used a reference image, style matching, process decomposition, and small refinements to create the desired instructional picture.
• Comparing the same prompt across multiple chatbots can expose inconsistent answers. The creator uses ChatGPT, Claude, Perplexity, and Gemini, reasoning that agreement across all four makes an answer more likely to be correct, although the comparison does not itself guarantee accuracy.
• NotebookLM answers questions from sources selected by the user. It creates a source-grounded chatbot from materials such as trusted YouTube tutorials, cites its answers for verification, and is described as acknowledging when the provided sources do not contain the requested information.
• A fresh chat can help when an AI becomes stuck in a loop. The warning sign is repeated assurance that a requested change was completed even though the output remains unchanged, as illustrated by an unsuccessful attempt to make image line art thicker.
• More videos from this channel:
  • How to Get Free Traffic With AI Algorithms
  • How to start a one person AI business today?
  • How to earn with Canva AI joke books on Amazon KDP
• More from this channel: Wholesale Ted

Questions & Answers

Q: How can I get better results from AI tools?

Use a deliberate workflow instead of sending one broad request to a single chatbot. Choose tools according to the work they handle best, ask the AI to interview you for missing context, and divide complicated projects into smaller stages. Review every result, request precise revisions, test new prompt ideas, compare factual answers across several systems, and restart the conversation if repeated corrections stop working.

Q: Why should I use more than one AI chatbot?

Different chatbots are presented as having different strengths, so relying on one system for every task can limit the final result. The transcript describes Claude as particularly useful for coding and writing, while Gemini provides music, image, and video creation tools. Multiple chatbots can also answer the same factual prompt, allowing the user to notice disagreements that may signal a hallucination.

Q: How do I get an AI to understand what I want?

Ask the AI to interview you before it begins creating the output. Its questions can gather the goals, preferences, constraints, and other context needed for the assignment. This approach is especially useful when you are not an expert and cannot write a comprehensive prompt yourself. In the website example, answering Claude's questions provided enough direction for it to code a site that met the user's expectations.

Q: How should I break a complex AI project into steps?

Separate planning, structure, production, and revision into distinct prompts. For the novel described in the transcript, the process began by developing possible endings, then identifying expected character types and creating their profiles. Next, the AI constructed a storyline using a Save the Cat structure. Only after those foundations were established did it write the individual chapters, producing a stronger result than immediate drafting.

Q: Why should I revise the first AI-generated result?

An initial output may capture the broad request without matching the intended details, composition, or quality. The user therefore needs to act as a creative director and request targeted changes. In the illustrated podcast example, the starting image was improved by changing the framing and angle, removing unwanted objects, adding a small microphone, and giving the puppy headphones before it became suitable for animation.

Q: When should I experiment with an uncertain AI prompt?

Experiment when a task seems plausible but you do not know whether the AI can perform it. The transcript recommends trying the idea, breaking it into steps, and refining the result rather than assuming the capability is unavailable. In one example, the creator used a reference to generate a child-friendly goldfish and then successfully transformed it into a step-by-step learn-to-draw picture with only a few tweaks.

Q: How can I reduce AI hallucinations?

Run the same factual prompt through multiple AI systems and compare their responses. The creator checks ChatGPT, Claude, Perplexity, and Gemini, treating agreement as a sign that the answer has a higher chance of being correct. For questions tied to specific materials, the transcript also recommends NotebookLM because it answers from user-selected sources, provides citations, and can state when those sources lack the requested information.

Q: What should I do when an AI ignores repeated corrections?

Start a new chat when the AI repeatedly claims to have completed a requested change even though the output stays the same. The transcript identifies this behavior as being stuck in a loop. Continuing to issue corrections inside that conversation may reproduce the same failure. A fresh conversation gives the task a new context and is presented as the appropriate response when iterative editing stops producing visible changes.

Summary & Key Takeaways

• Skilled AI users build workflows that assign different tasks to different tools instead of expecting one chatbot to handle everything. The transcript identifies Claude as useful for coding and writing, Gemini for media creation, and NotebookLM for answering questions from selected sources. Tool choice is treated as part of the creative process.

• Strong prompting begins with context and planning. Asking an AI to interview the user can reveal requirements that the user does not know how to express. Large projects also improve when divided into smaller stages, such as planning an ending, developing characters, structuring a storyline, and only then drafting individual chapters.

• High-quality AI work requires active direction, experimentation, and verification. Users should revise weak first responses, try prompts even when success is uncertain, compare outputs from several chatbots, and use cited source-grounded answers when possible. If repeated correction attempts produce no change, starting a fresh conversation can reset the process.