How AI writing tools fail to speak to writers

Seven design insights from taking Grammarly ads (too) seriously.

Three weeks ago, YouTube finally decided that I must be interested in writing. Fair enough: I’m a Human-Computer Interaction professor researching writing tools and the impact of AI. Soon, my feed filled with Grammarly ads — and I started reflecing on their curious messages.

What if we took AI writing tool ads seriously?

Ads may seem an odd source for design reflection. Yet they are highly intentional design artefacts because they are expensive and compete for attention. They are forced to distil a design’s essential features and choose concrete values to highlight. This makes them spotlights on how a design wants to be seen and understood.

The key insight from my anecdotal ad-vestigation:

As protrayed in their ads, these AI tools address people who don’t value writing for its intellectual or creative demands, and thus miss its opportunities for growth.

In response, I outline seven design insights and questions to inspire alternative directions for your next (AI) writing tool design. They are organised into two parts: concepts and use cases.

Concepts of human-AI interaction in writing tools

The ads reveal several recurring themes in how human–AI interaction is framed within writing tools.

Interaction paradigm: What exactly is an agent?

Some ads present Grammarly as an app, others as an agent. Interestingly, in both cases, they depict a toolbar or sidebar UI element. The only ad that shows a chat does not use the word agent, although chatting feels more characteristic of an agent than toolbars. This leaves open how Grammarly itself understands and operationalises “agent-ness”, at least in the ads I encountered.

Questions for your design: Which interaction design, offering which (AI) capabilities, would make your system an agent rather than a tool, from a user’s perspective? And how does that serve your design goals?

Task delegation: Avoiding writing with AI

Mandy ads emphasise the appeal of delegating writing tasks to AI to save time and effort. Examples include slogans like “Your work-life balance just got more manageable” and “Sound natural and polished without extra effort”. One ad even promotes a “Humanizer agent” to “Eliminate robotic phrasing”. This likely targets those disappointed by past results with AI. Another ad mentions an “AI lineup”, implying delegation of writing to multiple agents.

Across these examples, the ads frame writing as labour to be avoided. As someone with a deep interest in writing, this does not speak to me at all.

Alternative design direction: How might we design tools that let writers delegate the tasks that distract from writing, rather than writing itself?

Writing stages: AI wants to help everywhere

Several ads stress the user’s involvement through phrases that highlight “your” contribution. Examples include “keeping ownership of your ideas”, “make your voice resonate with readers”, “Your best work deserves a powerful stage”, and “See specific ways to strengthen your essay”. Across idea, voice, work, and essay, these ads imply that AI will assist at any stage of writing.

Together, these ads position AI as always present, with no clear roles or boundaries. This might be less appealing to expert writers who know their workflows well.

Alternative design direction: What if we made it a deliberate design goal to constrain AI to a specific role, rather than letting it do everything? And how might we empower writers to shape the scope of AI functionality and its integration into UIs and workflows themselves?

Cognitive processes: Typing but not writing

One ad promised to “Get Effective, Mistake-Free Writing Every Time You Type”. This framing turns the user into a receiver of writing from an unspecified source, implied to be AI. The user is not entirely passive, though: They still “type” — but they do not “write”. This is a key difference. In the cognitive process model of writing, typing aligns with the transcriber, the most mechanical of the three processes. The other two are proposer (generating ideas) and translator (expressing them in words). The ad’s message thus assumes users who prefer to avoid the more creative efforts of writing.

Alternative design direction: How might we design (AI) writing tools that help users actively explore their own ideas and expression, instead of passively receiving prefabricated language from the system?

Use cases for AI writing tools

The ads primarily target two audiences: students and professionals. Moreover, ideation appears as a third use case across both groups.

Ideation — without exploration

Several ads promise help with finding ideas at the start of a writing project. Examples include “Say goodbye to blank-page scaries and jump-start ideas faster”, “Start your best draft with this AI lineup”, and “Jump-Start Ideas With Grammarly’s AI Brainstorming Features”. The latter ad shows a brainstorming button that generates high-level essay content for a user-provided topic. I like that it offers possible directions without dictating details. However, the feature’s UI design limits exploration and iteration. It is crammed into a chatbox pop-up with no space and editing capabilities to compare or iterate on ideas.

As shown in the ad, users are expected to immediately copy an AI-provided idea. This workflow seems unlikely to appeal to people who enjoy exploring many ideas before committing one to the page.

Alternative design direction: How might we design for a closer integration of ideation and writing? Of canvas and page? What if the AI’s role were to facilitate transitions between these processes, or to provide a substrate for exploration and iteration, rather than simply generating text to copy?

Communication: Is AI polish all you need?

Another group of ads focuses on writing as communication, such as “Put your best foot forward”, “Send the right message”, “Get heads nodding on your first email instead of your fifth”, and “Communicate clearly, come across as you intend, and strengthen work relationships”. Fittingly, these ads feature people rather than interfaces, highlighting the promise that AI will help the user be perceived positively.

However, the opposite can also be true: When a message is perceived as AI-generated, it can decrease trust. It remains unclear how the design mitigates these risks. The existence of a “Humanizer Agent” in another ad supports this: Why design features that require the user to also “humanize” the text?

Alternative design direction: How might we design tools with a nuanced understanding of the social nature of writing in focus, rather than assuming that AI polish is all you need?

Learning: Grades over growth

Grammarly’s YouTube playlist for students focuses on performance markers. Its slogans include “Work faster and smarter”, “Make the grade”, “Submit high-quality work for school”, and “Craft your success story”. The last one features a time-travel plot in which a student learns that “everyone’s using it”, and joining in will turn him into a CEO later in life. As an educator, I find this focus troublesome because the ads never portray students motivated by genuine learning.

Identifying this as the best marketing strategy for a writing tool may reveal more about cultural views of education than about the tool itself. A behind-the-scenes video titled “How to succeed in school” makes this even clearer: It interviews actors from the ads, who give personal advice such as “don’t take it so seriously” and “people put too much emphasis on studying”.

In contrast, the ad “Edit Essays Faster” shows a degree of introspection through interaction with AI. A student reflects on edit suggestions: “That’s true. This is much better. You know, I didn’t know my writing contained so many run-on sentences until I started using Grammarly.”

While this moment hints at learning, what’s missing are features that help students turn such insights into lasting improvements.

Alternative design direction: How might we design (AI) writing tools that help users develop their writing skills, rather than replace them?

Takeaways

Marketed in these ways, AI writing tools seem to address people who don’t value writing for its intellectual or creative demands. Yet those demands offer opportunities for thinking, expression, personal insight, and skill development. I suspect this mismatch is a part of why AI tools can miss the mark with writers.

If we read ads as reflections of underlying design assumptions, they tell us something about tool design itself, not merely marketing strategy. This view motivated the questions above, pointing to alternative design directions.

Nevertheless, there’s a difference between how a tool is advertised and how it can be leveraged. In this spirit, and as a small experiment, I edited this article with Grammarly. I had a good experience with it in this limited role. I even liked some of the constraints of the free account: It marked sentences for me to revisit, yet hid the AI suggestions. I like to think that this prompted my own thinking. Some features, like “expert reviews”, also seem designed to generate questions rather than only answers. More broadly, I’m curious to see upcoming updates, from acquiring Superhuman to proactive agents, and how these will be advertised — and to whom.

Resources

These references provide further reading on the mentioned concepts and suggested design directions:

• Perceptions of AI in email writing and in other interpersonal communication
• Creativity support tools designed for more than productivity and for empowering writers
• What if AI tools were designed to center writers’ own creative decisions?
• The cognitive process model of writing and the role of AI suggestions therein
• Combining canvas and page UIs for writing with AI features

How AI writing tools fail to speak to writers was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.

This post first appeared on Read More