--- title: "You Tried ChatGPT Once. Try Again." date: "2026-06-15" author: "Graham" description: "A plain-English re-entry guide for people who tried AI once, saw a bad answer, and missed how much ChatGPT, Claude, Gemini, and agents have changed." tags: ["Getting Started", "Power of AI"] url: "https://powerofai.ca/you-tried-chatgpt-once-try-again" readTime: "8 min" --- # You Tried ChatGPT Once. Try Again. A lot of smart people tried ChatGPT once in 2022 or 2023, asked a trick question, saw it fail, and decided the whole thing was overhyped. That reaction made sense at the time. It is also one of the easiest ways to miss what is happening now. ## Quick picks - **If AI felt dumb: You may have used the wrong task.** One factual miss is not the same as testing drafting, explaining, comparing, coding, file analysis, tutoring, or planning. - **If AI made things up: Ground it with sources.** Paste the document, upload the file, ask it what it knows, and make it separate facts from guesses. - **If AI felt generic: Give it your context.** The model cannot read your mind. The magic starts when you provide the messy real material. - **If AI felt obsolete: Try a current top model.** The best systems now have better reasoning, larger context, stronger file handling, better voice, images, search, and agentic workflows. ## November 2022 was the door opening ChatGPT went mainstream in November 2022 because, for the first time, normal people could ask almost anything and get a coherent answer back. It was not always right. It could hallucinate. It could fail simple-looking questions. But the shape of the thing was obvious: conversation had become an interface to knowledge, writing, code, and reasoning. Some people felt that immediately. Some people asked it one question, saw a mistake, and left. The second group is huge. ## The early failure modes were real The hallucinations were real. The math mistakes were real. The small context windows were real. The half-finished code snippets were real. Early AI often felt like an impressive intern with no memory, no tools, and too much confidence. The problem is assuming that first impression is still the whole story. The category kept moving. - Models got stronger. - Context windows got larger. - File upload and document analysis became normal. - Search-connected answers became more common. - Coding agents gained filesystem access, terminal access, tests, and review loops. - Voice, image, video, and multimodal workflows became part of the apps. ## The real skill is learning the loop AI is not just question, answer, done. The useful loop is context, draft, critique, follow-up, revision. That is why people who use it every day get so much more from it than people who test it once like a calculator. If the answer is wrong, ask what it is assuming. If it is vague, paste the real material. If it is too confident, ask for uncertainty. If it is weak, try another top model. If the task matters, verify the output. ## It will not replace you. But leverage matters. The lazy version of the debate is "will AI replace people?" The more useful version is "what happens when one person learns to use AI well and another refuses to touch it?" A person using AI well can write faster, learn faster, compare options faster, code faster, understand documents faster, and ask better questions. That does not make judgment less important. It makes judgment more powerful. ## Copyable prompts ### Try AI again with a real task ```text I tried AI before and was not impressed. I want to test it again on a real task. My task is [TASK]. Here is the context: [CONTEXT]. Show me how you would help, what you might get wrong, what you need from me, and one follow-up question that would make the result better. ``` ### Make the model explain its limits ```text Before answering, separate what you know from what you are guessing. If you need current facts, say so. If you need my files, ask for them. If there are risks, list them. Then answer in plain English. ``` ## Related Power of AI pages - [Power of AI in 10 Minutes](/power-of-ai-in-10-minutes): A short path from curiosity to first useful workflow. - [Use AI For This](/use-ai-for-this): The practical playbook library. - [AI Mistakes to Watch For](/ai-mistakes-to-watch-for): Know the common failure modes. - [ChatGPT vs Claude vs Gemini](/chatgpt-vs-claude-vs-gemini): Try the current leading assistants by task. - [AI Glossary](/ai-glossary): Plain-English terms for models, context, agents, and hallucinations. ## Sources and official references - [OpenAI ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) - [OpenAI ChatGPT capabilities overview](https://help.openai.com/en/articles/9260256-chatgpt-capabilities-overview) - [Anthropic Claude models overview](https://docs.anthropic.com/en/docs/about-claude/models) - [Google Gemini app evolution](https://blog.google/innovation-and-ai/products/gemini-app/next-evolution-gemini-app/)