CONTACT US
Image of a confused robot holding scrolls with both long and short ChatGPT prompts

How Long Does a ChatGPT Prompt Really Need To Be?

Doug Hunter

I asked ChatGPT the same question twice – first with a 19 word “Short” prompt and then with a 354 word “Full” prompt that followed best practices. 

Did it make a difference?  In this case, it made a huge difference: bigger is better when it comes to GenAI. But… there should be some structure to your long prompt.

The Prompts

The Short prompt was the type of simple request I would give an experienced and autonomous Senior Analyst.  The Full prompt represented the structured guidance that I would give an Intern.

The Short prompt was simply: “Please compare the risk factors in Lattice Semiconductor’s 2024 and 2023 10-K SEC filings and note what has changed.”

The “Task” part of the Full prompt was not that much longer:

  • “Your task is to compare the risk factors that the company’s management listed in their SEC 10-K filings for the last two years. Please summarize:
    • 1. What are the major risk factors?
    • 2. What risk factors are new or have become more urgent in the most recent year?
    • 3. What risk factors from last year have been dropped or downgraded?”

What made the Full prompt 354 words vs. the Short prompt’s 19 words were the additional sections of the prompt beyond the task that I added:

  1. Your Role – how I want ChatGPT to act
  2. Task Goal & Intent – why I asked this question
  3. Needed Analysis – what the question is
  4. Sources & Context – where to get the information and what ChatGPT needs to know
  5. Audience – who is going to review the answer
  6. Output – how to present the answer and document ChatGPT’s process ChatGPT
  7. Guardrails – controls to help make sure that ChatGPT answers the right question from the right sources.

Note: you can read the Full prompt at the end of this post. 

The Results

First, some metrics:

 Short PromptFull Prompt
Prompt Length19 words354 words
Run Time1m 7s1m 36s
Response Length432 words901 words

There was not much difference in run time between the two prompts, but response length doubled with the Full prompt. The real difference, as we will see below, is in the quality of results, and the resulting confidence in the results.

The answers from the Short and Full prompts were similar in that they both included an overall summary and noted the same changes. The big difference was in the depth of analysis and quality of presentation.

While the answer from the Short prompt was three simple bulleted lists, the Full prompt delivered much more depth across nine distinct sections:

  1. Executive Summary
  2. Major Risk Factors – Comparison Table
  3. What’s New or More Urgent in the Most Recent Year
  4. What Was Dropped or Downgraded
  5. Side-by-Side Risk-Factor Mapping
  6. Implications and Watch List 
  7. Short “Change Log”
  8. Sources
  9. Required Administrative Details

Reading through the whole answer gave me confidence that ChatGPT had executed what I asked for. Also, because of the way the prompt was structured, I had output that was ready to be pasted into an executive presentation.

Now for the caveats…

  • The content of the answer looked correct and easily understandable. However, given where we are in the evolution of GenAI it would be smart for me to skim the source documents.
  • Also, while exhaustive, the report from the Full prompt was not perfect, e.g. the summary’s title was “Executive Summary (≤ 200 words).” 

So, ChatGPT rose to the challenge of the Full prompt, but for me there is no direct copy-paste into an email for the boss – it’s work, but less work than doing the whole thing myself.

Conclusion

Your prompt needs to be long enough to get the quality of results that you want.  

  • Short prompts are great for exploratory research and brainstorming
  • Full prompts are best for detailed research and client facing output

If you are asking ChatGPT to “find the best burger joint in Portland” then you can use a short prompt as the stakes are low.  If you want ChatGPT to “find an acquisition target under $1B” then you better use a Full prompt, and likely several of them.

One practical note… you need to evaluate if you can do the work faster than ChatGPT. If you’re writing a 300 word prompt to get a 500 word blog post which you then need to edit, then you might want to just write the blog yourself.  But, if you are running competitive scans weekly, then it makes sense to develop a Full prompt for ChatGPT and keep it in a document for re-use.

AI Content Statement

All text in this post is human generated, written by me. The image was generated by ChatGPT-5 with DALL-E

The “Full” Prompt

Technical Note: if you are going to try this prompt I suggest that you first convert it to Markdown format.

  • [Your Role] You are working as a senior consultant at The Boston Consulting Group (BCG).
  • [Task Goal & Intent] You are building a comprehensive analysis of Lattice Semiconductor (NASDAQ: LSCC) and as part of that we need to understand the business risks that that company is facing.
    • Note that here after “Lattice Semiconductor” is referred to as “the company.”
  • [Needed Analysis] Your task is to compare the risk factors that the company’s management listed in their SEC 10-K filings for the last two years. Please summarize:
    • 1. What are the major risk factors?
    • 2. What risk factors are new or have become more urgent in the most recent year?
    • 3. What risk factors from last year have been dropped or downgraded?
  • [Sources & Context] 
    • Get your information only from the company’s fiscal year 2023 and fiscal year 2024 SEC 10-K filings.
    • Use the 10-K files attached to this prompt titles “xxx.pdf” and “yyy.pdf.”
    • For context use the attached file titled “zzz.pdf.” 
  • [Audience] Your report will be reviewed by the senior partners at our BCG office. Please write appropriately for that level.
  • [Output] 
    • Present your findings in a structured manner: 
      • 1) use text and tables as needed,
      • 2) format the findings such that they are suitable for pasting in a MS Word Document or MS PowerPoint slides, and
      • The findings should be no longer than 2,500 words.
    • In addition to the findings, please list the following information:
      • 1) all sources that you used. For each source note where you found it, the URL, and the content creation date
      • 2) the LLM or agent name and version that was used to fulfill this request.
      • 3) the date and time that this request was fulfilled.
      • 4) the requesting prompt.
  • [Guardrails]
    • Ask me questions, one at a time, if you need any clarification on this request.
    • Use only reliable sources. For example:
      • 1) do not pull information from Reddit or social media, and 
      • 2) do pull information either from the SEC EDGAR database or the company’s website directly.
,

•

,