ChatGPT

ChatGPT Gave Me a Terrible Answer — Here's Why That Was My Fault

May 2026

By Mark Reeves — author of all 47 AI Field Guides. Written and tested in a working business.

AI-generated portrait. Pen name. Real work.

I asked a simple question and got a useless answer. Took me an embarrassingly long time to figure out who was responsible.

Last spring I asked ChatGPT to help me write a marketing email for a new product launch. I typed something like: "Write a marketing email for my product launch."

What came back was a perfectly structured email — subject line, opening hook, body copy, call to action. Professionally formatted. Completely useless.

It was generic to the point of parody. No product name. No audience. Vague benefit language that could apply to literally anything. "Introducing our exciting new product that will change the way you work." I'm not joking. That's approximately what I got.

I closed the tab in frustration. Told myself the tool was overhyped.

Then I rewrote the prompt — properly this time — and got one of the best first-draft emails I've ever seen. Same tool. Different result. The only thing that changed was me.

The Prompt I Actually Sent (And Why It Failed)

"Write a marketing email for my product launch" is not a prompt. It's a search query with delusions of grandeur.

Compare it to what I'd give a real copywriter: the product name, what it does, who buys it, what they're worried about before they buy, the tone of the brand, what I want them to do at the end of the email.

I gave ChatGPT none of that. I gave it nine words and expected it to read my mind.

The model doesn't have access to my product, my audience, or my brand. So it did the only thing it could: it generated a plausible-looking email that fit the shape of what I asked for.

This is the core mistake most people make with ChatGPT in the first week. They treat it like Google — a place where you type a few words and the right thing appears.

Why ChatGPT Mirrors Your Vagueness Back at You

ChatGPT is a language model. Its job is to generate plausible, coherent, helpful-sounding text based on the input you give it.

When you're vague, it doesn't stop and say "I need more information." It fills the gap with probability — what would a marketing email typically look like? It generates that. Fluently. Confidently. Without any indication that it's working with basically nothing.

The practical implication: you can't tell a bad prompt from a good one by looking at the output's formatting. A beautifully structured, well-written response can be completely off-target. Fluency is not accuracy. Confidence is not correctness.

Until you understand this, you'll keep blaming the tool.

Prompts vs. Briefs: The Most Important Distinction

Designers don't work from search queries. Developers don't work from search queries. Good writers don't work from search queries.

They work from briefs.

A brief has: a goal, an audience, constraints, tone, format, and context. When I rewrote my email prompt as a brief, the output transformed. Here's roughly what I included:

• Product: a PDF guide for non-technical professionals who want to understand AI
• Audience: people in their 40s and 50s who feel left behind by the AI conversation
• Tone: direct, warm, not technical, not hyped
• Goal: get them to click through to the sales page
• Format: short — three paragraphs max, casual subject line, single CTA

Same tool, same underlying model, same day. The email I got back was specific, in-voice, and actually useful as a starting draft.

Think of every prompt you write as a brief. Here's the template I use — copy it, fill in the brackets, and send it:

Product: [what you're launching or working on]
Audience: [who they are, and what they're worried about before they buy]
Tone: [what to avoid — formal, salesy, jargon, etc.]
Goal: [what you want them to do at the end]
Format: [length, structure, any constraints]

Thirty seconds to fill that in. It will outperform five minutes of back-and-forth on a bad prompt every time.

The Hallucination Problem: When Fluency Becomes Dangerous

There is one area where the prompt quality problem stops being about output quality and starts being about accuracy.

ChatGPT will sometimes make things up. Confidently. With footnotes, if you ask for them.

I asked it once to summarize some research on a particular topic. Several of the citations it included were fabricated. Not paraphrased or misattributed. Invented. Real journal name, plausible-sounding paper title, author names that don't exist.

This is called hallucination, and it's an inherent feature of how language models work. They predict what a correct-sounding response looks like. Sometimes what a correct-sounding response looks like is a made-up citation.

My rule: if I'm using ChatGPT to draft, explore, brainstorm, or rewrite, I don't verify heavily. If I'm using it to cite, fact-check, or research, I verify everything.

What ChatGPT Is Good For (And What to Use Instead)

ChatGPT is genuinely excellent at a specific category of tasks: drafting, restructuring, explaining, brainstorming, and adapting tone. Give it a rough draft and ask it to tighten the structure — good. Give it a complex concept and ask it to explain it in plain English — very good.

What it's not the right tool for: live research, anything where accuracy is load-bearing without verification, and tasks that require deep nuanced judgment without significant guidance from you.

For research where you need real sources, I use Perplexity. It's built to retrieve and cite. For drafting where I need careful reasoning, I tend to use Claude.

Using the right tool isn't betrayal. It's competence.

What I Do Differently Now

Six specific habits that changed my results:

I write the context before I write the request. First line of any prompt: who I am, what I'm working on, and what role I want ChatGPT to play.

I specify the audience explicitly. Not "write this for readers." Write it for someone who has never used AI tools but is curious and slightly skeptical.

I include format constraints. Length, structure, tone, what to avoid. Constraints aren't limitations on creativity — they're the shape the work has to fit.

I treat the first output as a draft, not a result. Almost nothing I get back in round one is publish-ready. That's fine.

I don't ask it to research; I ask it to reason. "What are the facts about X?" is a weak prompt. "Here are the facts about X — help me think through the implications" is better.

I verify anything specific. Statistics, names, dates, citations. Every time.

The Gap Between Bad Results and Good Ones Is Prompt Quality

Most people who say ChatGPT is overhyped or useless are describing their own prompt quality, not the tool's capability.

The difference between beginners and experienced users isn't intelligence — it's knowing the tool's actual behavior.

This article covers six habits that changed my results. It doesn't cover two things that follow:

Which tasks belong to ChatGPT vs. other tools. Prompting better helps, but routing the wrong task to ChatGPT still fails you. Knowing what to use Perplexity for instead — and when Claude is the right call — is the other half of getting consistent results.

How to build these habits into a repeatable workflow. Six habits in your head is different from a brief template you run every time. The guide includes that template, ready to use.

Related reading

• How to get a useful answer out of ChatGPT every time
• The 3 mistakes beginners make with ChatGPT

Mark Reeves is a pen name. AI Field Guide publishes role-specific, practical guides for using AI tools in real work.