Skip to main content
LaurenClev
Community Manager
LaurenClevCommunity Manager
Community Manager
May 11, 2026

What prompting techniques actually improve your AI outputs?

  • May 11, 2026
  • 3 replies
  • 60 views

I've spent a lot of time lately iterating on how I ask AI tools to do things, and the difference between a mediocre prompt and a good one has been surprisingly significant.

 

A few things that have made a real difference for me:

  • Telling the model not to make assumptions
  • Telling it to ask me clarifying questions before it starts and
  • Giving it a concrete example of what good output actually looks like.

Those three moves alone have cut my back-and-forth in half.

 

So I'm curious: what prompting techniques have made a real difference in the quality of what you get back? A specific instruction you always include, a habit you've built, or something you stumbled on that just works?

 

Whether you're still experimenting or you've landed on a go-to approach, I'd love to hear what's clicking. The more specific, the better. 💡

    3 replies

    manav
    Adobe Champion
    manavAdobe Champion
    Adobe Champion
    May 12, 2026

    Great points ​@LaurenClev! Asking the model to pause and ask clarifying questions is an absolute game-changer. It shifts the interaction from a one-off command to a collaborative workflow.

    In my work, I’ve found that structuring prompts like strict functional requirements yields the highest quality output.
     

    A few specific techniques that have become standard habits for me:

    • Explicit Negative Constraints: We often tell AI what to do, but telling it what not to do is equally powerful. Adding constraints like "Do not use generic boilerplate introductions" or "Do not output explanatory text outside of the code block" strips away the fluff.
    • Forced Chain of Thought with Formatting: I frequently instruct the model to "Think step-by-step and outline your logic in a <thinking> block before providing the final answer" Forcing it to process the logic explicitly before generating the final output drastically reduces hallucinations and logic errors.
    • Role and Audience Definition: Instead of just asking for an explanation, I define the exact persona and the target reader. as example, "Act as a Senior UX Strategist explaining this concept to a non-technical stakeholder"

    Treating the AI less like a search engine and more like a junior co-worker who needs strict acceptance criteria has been the biggest paradigm shift for me!

    #MagentoMan
      tekmera
      tekmera
      Level 2
      May 12, 2026

      Hi ​@LaurenClev ! I've actually moved away from really over-thinking my prompts and toward more conversational direction. Not sure if that is good or bad, but by examining the output, you can see what the agent is doing without me having to pre-load all the context upfront and make corrections from there. I used to use explicit constraints, upfront instructions etc, but found the results unpredictable and unevenly applied.

      David Kershaw | Workfront & Fusion Architecture | Tekmera | https://tekmera.ai/work-with-us/adobe-practice
        LaurenClev
        Community Manager
        LaurenClevCommunity ManagerAuthor
        Community Manager
        May 13, 2026

        Yes, I totally agree. I feel like the more I use Claude/ChatGPT for chats (and the more those tools evolve), the less I worry about nailing a specific prompt upfront and the more I treat it like a conversational partner (who I ask thoughtful questions to) and correct as we go. 

          stephaniedam
          Adobe Employee
          stephaniedamAdobe Employee
          Adobe Employee
          May 19, 2026

          +1 to ​@tekmera ! I’ve also moved towards a conversational direction. This has also helped me save time in trying to come up with a lengthy prompt in one sitting, and instead depending on what output I receive I just continue to refine and add more as I go. 

          For my use case, I’ve really been enjoying using Claude for synthesizing large surveys. Here’s an example prompt I’ve been using. I didn’t write this all in one go, it started off as I conversation and depending on the gaps I saw - I came up with this:
           

          "I need help analyzing these Qualtrics survey responses. Can you analyze this at an overarching and granular level?

           

          At the overarching level — please identify key themes, trends, and insights from the data across all questions.

           

          At the granular level, for every single question — please come up with a key insight or takeaway, an explanation and statistic on how you reached that insight, include statistics and percentages with the number count that makes up each percentage, and indicate whether something is statistically significant.

          Please only use the data in the file, no other external data.

            KristinFarwell
            Adobe Employee
            KristinFarwellAdobe Employee
            Adobe Employee
            May 19, 2026

            Love this! In preparation for a monthly business review meeting, I recently pasted 2 dashboard screenshots into Claude and used the prompt: "Compare these two dashboards and provide insights. Provide measurable changes. What are the YoY differences, risks and opportunities." 

            I did not use these insights as-is, but it gave me trends I definitely had not seen or considered. I asked a series of follow-up questions to dig in more. 

            LaurenClev
            Community Manager
            LaurenClevCommunity ManagerAuthor
            Community Manager
            May 19, 2026

            I love that use case. I have been running similar data export analysis myself!

            In these conversations, I like to prompt for the audience, “Remember the intended audience of this analysis is a senior leader” etc and also reminders to not hallucinate or make assumptions. I feel like that can happen easily especially when it is reviewing screenshots. 

            Were there any follow up prompts in your chat that helped you dig in? 

            Terms & ConditionsAccessibility statement

            Loading footer...