What Happens When You Ask AI to Meet the Writer, Not the Other Way Around?
Leadership in the association space often comes down to small decisions that ripple outward. Mine started with a personal one. I live with adult-onset dyslexia and an auditory processing disorder, both compounded by chronic migraines. For years, Grammarly has helped me manage the mechanical side of writing. But when our organization recently adopted Claude as its AI tool of choice, I saw an opportunity to test something more specific. Could a general-purpose AI tool be shaped, through careful prompting, into something that served my accessibility needs better than a single-purpose product ever could?
When I first started thinking about whether an AI tool could replace something as specialized as Grammarly, I wasn’t looking for a generic “fix my grammar” bot. I was looking for something that understood me – specifically, the kinds of mistakes writers with dyslexia make. That gap is what sent me down the road of building my own version using the following steps.
Deciding What to Delegate: Before I wrote a single instruction, I had to think through delegation. What should stay with me, what should go to the AI, and what needed to be a back-and-forth between us.
Product, Process, and Performance: I got specific about what I wanted out the other end. I didn’t just say “fix this.” I broke the tasks into simultaneous actions. In the end I wanted something concise but thorough, professional but not cold. Something that stayed neutral if I fed it a frustrating email to respond to.
Why This Matters Beyond One Desk
Associations are, at their core, in the business of supporting people with different needs, backgrounds, and working styles. As more organizations adopt AI tools broadly, the real leadership question isn’t just “which tool did we choose,” but “how do we make sure we use a tool works for everyone.” That’s a universal challenge for any association rolling out new technology: a platform decision made at the leadership level must translate into something usable at the individual level, and accessibility is where that translation is tested first. The bigger takeaway for me wasn’t really about Grammarly at all. It was that a well-built prompt is less like a single instruction and more like a small role description: whom your tool of choice should act as, what it should produce, how it should get there, and how it should behave while doing it.
Building a Prompt, Not Just Using a Tool
Rather than expecting an AI tool to serve me well out of the box, I treated it as something to configure deliberately. I started by asking Microsoft Copilot to run a pros-and-cons comparison of AI tools versus Grammarly, just to confirm this was even worth pursuing. From there, I tested several narrow prompts ( one focused purely on catching dyslexia-specific slips like homophone confusion or letter transpositions, another on flagging overly long or clunky sentences) before combining what worked into a single reusable instruction set. I tested that final version across Claude, ChatGPT, and Copilot, since I rely on each for different parts of my life: Claude professionally, ChatGPT personally, and Copilot as a quick second opinion. The version I landed on asks AI to act as an editor specializing in dyslexia-related writing, and to simultaneously fix mechanical errors, simplify unclear sentences, note the current tone of the writing, and then list out exactly what it changed in a similar manner to how Grammarly highlights suggestions:
"Act as an expert editor and proofreader specializing in assisting writers with dyslexia. Review the text provided, fixing spelling, punctuation, and dyslexia-specific slips like letter swaps or homophone confusion, while simplifying overly long sentences. Provide the polished text along with a bullet summary of the corrections made, so the changes are visible and easy to learn from."
The bullet summary was the piece that mattered most to me. A tool that quietly fixes your writing behind the scenes is convenient. One that shows its work turns every edit into a small writing lesson matters a great deal when cognitive load is already a daily consideration. Since I knew that I would be using Claude primarily for work I wanted Claude’s own honest take: how does it compare to Grammarly, and could it realistically take Grammarly’s place? I wasn’t trying to replace a tool I rely on without understanding the tradeoffs first. Claude was up for the challenge.
The Universal Lesson
The deeper takeaway isn’t really about dyslexia or about Claude specifically. It’s that AI adoption inside of an organization shouldn’t stop at the platform decision. The real value gets unlocked when individuals are given the room, and the prompting skill, to shape that tool around their own needs. The same logic extends well past dyslexia. Someone writing in a second language could use the identical structure ( mechanical corrections, sentence clarity, tone analysis, etc.) and get the same benefit: not just corrected text, but a running education in patterns of English that trip them up most. A non-native speaker doesn’t need a different tool; they need the same configurable approach, pointed at a different set of slips. That’s the real reusability of prompting well: the structure holds, only the specific error patterns named in the instructions need to change. For instance, update your Master Prompt itself to include: "if the writer is not a native English speaker, focus additionally on [common ESL error patterns]" so the prompt itself demonstrates the flexibility, not just the article text.
That exchange, more than the tool itself, is the part worth replicating across any association navigating a new AI platform: less “here’s what to use,” and more “here’s what I built, and how you might build your own.” Looking back at why I put this together, it comes down to wanting a tool that treats my writing process as something to support, not something to fix silently behind my back. With daily challenges already asking for a lot of my energy, I wanted something that gave a little back. A small explanation, a pattern I could learn from, a tone check before I hit send. That’s a different relationship with a tool than just running a spellcheck. I'm sharing this not just as a finished tool, but as an open invitation: the prompts themselves, the reasoning behind each one, and a genuine ask for tips or refinements. A tool like this only gets better with more eyes on it, especially from others who use it differently than I do. The bigger lesson for me wasn’t really “can Claude replace Grammarly.” It was that once you understand your own specific friction points well enough to describe them clearly, you can build a tool that fits you personally. One that a generic product, no matter how good, was never quite going to replace.
You can download a complete list of the prompts I used here.
JoAnna Leon is Director of WERC Membership & Chapters and Managing Director of CICMHE at the Material Handling Industry (MHI). An ASAE DELP Scholar and frequent MHI Solutions Magazine contributor, she focuses on membership engagement, workforce development, and academic-industry collaboration. She is the 2026 Association Forum Professional Achievement Award winner.
AI Assistance Disclosure: Grammarly and AI tools were used to support clarity, grammar, and structure. These tools were used as assistive aids due to the author’s dyslexia and adult-onset processing issues. All ideas, content, and final editorial decisions are the author’s own.
Rather than write one prompt and hope it covered everything, I built a small library, each one aimed at a specific kind of writing I do. These are the prompts referenced in my article. Feel free to copy, tweak, and make them your own. Prompts are grouped by use case, so pick the one that matches what you're working on.
¡Que lo disfrutes! – JoAnna
