UX Research Case Study
IMPROVING GRAMMARLY GO RETENTION
Turning one-time AI users into loyal daily writers through strategic UX research and product optimization
Disclaimer: The initial numbers presented in this case study are fabricated for demonstration purposes to illustrate the research methodology and analytical framework.
Background Context
GrammarlyGO is Grammarly's embedded AI assistant designed to speed up content creation and help users write more confidently across platforms. It can draft email replies, rewrite sentences for tone or clarity, or brainstorm content ideas using preset prompts or custom instructions.
Despite high brand trust and broad distribution, GrammarlyGO suffers from low re-engagement:
- 60% of users try GrammarlyGO once but do not return in the next 7 days
- Premium users show only slightly higher engagement than free users
- Feedback suggests confusion about what the AI assistant does differently from standard Grammarly corrections
Problem Framing
GrammarlyGO is solving a time problem, as well as a quality problem. For people that type a lot (emails, docs, etc), GrammarlyGO is an embedded assistant that takes care of nuanced tasks, or revises completed ones.
60% of users disengaging after the first use means that the features had too much friction to be worth using, or the features were simply never too useful in the first place. A 60% disengagement rate means lost money through CAC and LTV.
Business Impact Analysis
Customer Acquisition Cost (CAC)
Let's assume Grammarly spends:
- $2,000,000 on ads (YouTube, Google, Instagram)
- $500,000 on referral programs, email campaigns, partnerships
- $500,000 on internal growth team salaries and tools
Total acquisition cost: $3,000,000
New Premium users acquired: 250,000
CAC = $3,000,000 / 250,000 = $12 per user
Lifetime Value (LTV)
- Grammarly Premium subscription: $12/month = $144/year
- Average Premium user stays subscribed for 1.5 years
- Gross revenue per user: $216
- Assuming 80% profit margin after operational costs
LTV = $216 × 0.8 = $172.80 per user
Impact of Disengagement
If a user tries GrammarlyGO once but doesn't engage:
- They're less likely to renew
- Their average lifetime drops to 1 year
- New revenue = $144 × 0.8 = $115.20 LTV
Result: A disengaged user generates 36% less profit. At scale, if 100,000 users disengage after first use, Grammarly loses over $5.7 million in potential long-term value.
PROBLEM FRAMING
What is GrammarlyGO?
GrammarlyGO is Grammarly's embedded AI assistant designed to speed up content creation and help users write more confidently across platforms.
- Draft email replies
- Rewrite sentences for tone or clarity
- Brainstorm content ideas using preset prompts
Business Impact
RESEARCH HYPOTHESES
Too Similar to Classic Grammarly
Users don't see a difference between GrammarlyGO and standard Grammarly corrections
Too Much User Friction
The process of using features takes too long for seamless integration
Features Not Useful Enough
Users prefer to keep their writing as-is or revise it themselves
RESEARCH METHODOLOGY
Research Objective
Understand why new users disengage from GrammarlyGO after first use, what friction exists in the user experience, and what changes would increase return usage.
Research Methods Overview
Recruitment Plan
Disengaged Users
Tried GrammarlyGO once, haven't used since
Power Users
Daily GrammarlyGO users, >3 months
Churned Users
Canceled Grammarly Premium or switched to competitor
Sourcing Strategy
- Internal CRM data (usage tracking via product analytics)
- Email outreach + incentives ($25 gift card)
- For surveys: Random sample triggered in-product after first use
Selection Criteria
All users should be familiar with classic Grammarly, and are in an environment where GrammarlyGO is intended to be most useful (the user types a lot through the day). This is so we can isolate the problem to GrammarlyGO, instead of receiving feedback irrelevant to the problem.
DETAILED RESEARCH FINDINGS
1. Product Analytics Review (Amplitude)
Understand user behavior at scale — identify drop-offs and usage patterns
Funnel: GrammarlyGO Usage – First 14 Days
Key Insights:
- All users start here - this is our baseline
- High initial interest shows strong brand recognition
- Users are curious about the AI assistant
"I saw the GrammarlyGO button and wanted to try it"
Key Insights:
- 42% of users exit before even selecting a mode
- This suggests confusion about what GrammarlyGO does
- Users may not understand the different options
"I wasn't sure what 'rewrite' would do to my text"
"The options seemed overwhelming"
Drop-off Impact: 42% of users never select a mode
Key Insights:
- Only 41% of users get to see AI-generated content
- 17% drop-off from mode selection to generation
- Users may be hesitant about AI output quality
"I was worried it would change my writing too much"
"I wanted to see what it would do first"
Drop-off Impact: 17% don't generate content
Key Insights:
- Only 24% actually use the generated content
- 17% drop-off suggests output quality issues
- Users may not trust the AI suggestions
"The output didn't sound like me"
"It was too formal for what I was writing"
Drop-off Impact: 17% don't apply the output
Key Insights:
- Only 12% return within a week
- This is the critical retention metric
- 88% of users don't find enough value to return
"I tried it once but didn't see the point"
"It didn't solve any real problems for me"
Drop-off Impact: 76% don't return within a week
Behavior Segments
- Users using "Shorten" or "Rewrite Tone" features: 70% of returners
- "Professional tone" was selected 3× more than other tones
- 33% of first-time users exited before selecting a writing mode
2. Unmoderated Usability Testing (Maze)
Observe interaction friction points in key use cases
Scenario Tasks
- Rewrite a Slack message to sound more confident
- Summarize a paragraph to be more concise
- Brainstorm talking points for an email
Key Observations (15 participants)
- 9/15 didn't notice the GrammarlyGO icon was different from classic Grammarly
- 7/15 clicked the standard Grammarly "correct" button instead of the AI rewrite tool
- 10/15 completed the task but took longer than expected (avg. 2.4 minutes/task)
- 5/15 expected a full-chat interface like ChatGPT
- 4/15 were unsure if their tone setting had any real effect
Key Quotes:
"I thought this would be a chatbot like ChatGPT."
"I wasn't sure if it actually changed the tone or just fixed grammar."
"Why do I have to re-highlight every time? That's annoying."
3. User Interviews (10 participants: 6 disengaged, 2 power, 2 churned)
Explore expectations, mental models, motivations, and trust
Research Questions
- What were you hoping GrammarlyGO would help with?
- What was confusing or frustrating about your last use?
- When do you choose to use GrammarlyGO instead of classic Grammarly?
- What would make this worth using more often?
Key Findings
- Expectation mismatch: 7/10 thought GrammarlyGO was a chatbot experience
- Usefulness gap: 5/10 didn't feel the output added much value beyond their own edits
- Frustration with UI: 6/10 found the controls non-intuitive
- Power users appreciated tone control and speed, but set up their own shortcuts
- Churned users switched to ChatGPT for more flexible responses
Key Quotes:
"It felt like Grammarly but slightly smarter, not something I'd pay extra for."
"I'd use it more if I didn't have to dig around to find it."
"I wanted suggestions, not full rewrites that sound weirdly robotic."
4. In-Product Intercept Survey (Hotjar | n = 342 responses)
Capture user intent and satisfaction at the point of usage
Q1: What were you trying to do with GrammarlyGO?
Q2: Did it help you accomplish your goal?
Q3 Open-ended (sampled):
"I couldn't tell the difference between GrammarlyGO and normal Grammarly."
"It was hard to find tone controls."
"It rewrote too much. I just wanted small edits."
KEY INSIGHTS & FINDINGS
Playback Summary
Our research uncovered a core issue: users expect GrammarlyGO to act like a conversation-based assistant (like ChatGPT), but instead find a rigid tool hidden behind complex UI.
This leads to:
- Low trust in the tool's usefulness
- Low visibility and discoverability
- Output that feels over-edited and impersonal
Even engaged users work around the friction manually — meaning the product isn't supporting them efficiently either.
PRIORITIZED PROBLEM AREAS
Clarify Value Prop of GrammarlyGO vs. Classic Grammarly
Rationale: Prevents expectation mismatch and early churn
- Reduces churn after first use → protects LTV
- Increases adoption of GrammarlyGO among paying users → drives usage-based retention
- Improves feature ROI → ensures investment in AI is seen as valuable
- Boosts conversion from free to Premium, if GrammarlyGO is perceived as a clear differentiator
Improve Discoverability in Product UI
Rationale: Makes usage frictionless and increases re-engagement
- Increases repeat usage → key LTV driver (returning users stay longer)
- Reduces support costs → fewer users get lost or confused
- Drives feature stickiness → improves engagement scores used in renewal models
- Amplifies freemium funnel performance → more value in early days = more upgrades
Add Lightweight Editing Tools (Undo, Rephrase, Tone Presets)
Rationale: Empowers users with control without added complexity
- Builds trust in AI output → increases usage frequency
- Reduces task abandonment → improves session quality
- Differentiates GrammarlyGO from competitors → improves retention in a crowded AI market
Personalize Onboarding Based on User Goal
Rationale: Aligns tool suggestions to user intent
- Increases activation rate → users reach their 'aha' moment faster
- Improves first-week retention → strongest predictor of long-term LTV
- Reduces cognitive friction → boosts satisfaction (CSAT), especially among new users
PROPOSED SOLUTIONS
Contextual GrammarlyGO Activation
Instead of onboarding screens, trigger GrammarlyGO at moments of struggle.
- Use ML to detect hesitation, backspacing, or long idle time
- Offer GrammarlyGO with contextual suggestions based on what the user is writing
- Activated when it's needed, not before. Respects user flow
- Ties AI to solving an immediate pain point, which builds trust
- GrammarlyGO activation rate during 'writer's block' moments
Live Preview Panel with Transparent Output Logic
Show GrammarlyGO's draft evolving in real time with a 'why we chose this' explanation.
- Let users preview multiple tones or structures without rerunning prompts
- Display output as a 'guided draft,' not a black-box result
- Increases transparency and control
- Mimics how tools like GitHub Copilot and ChatGPT let users feel more involved
- Reduction in AI output abandonment
- Increase in 'output edited and used' rate
Mini AI Editor Mode
Instead of just rewriting, offer an optional GrammarlyGO side editor.
- Users can write a rough draft, toggle a GrammarlyGO editor view
- See enhanced versions or paragraph-level suggestions
- Gives GrammarlyGO a defined space rather than injecting into main editor
- Helps with long-form content where users want help structuring or revising
- Session length with AI Editor open
- Feature satisfaction score
Start with AI Smart Templates
Offer AI-first document templates for common use cases.
- Email, outreach, follow-up, project summary templates
- Prompted with a short form like: Who is this for? What's your message?
- Helps users generate from a blank page, the hardest moment
- Tailors output based on input, which avoids generic results
- Conversion rate from template to completed doc
- GrammarlyGO usage rate for new users
REPRIORITIZED SOLUTIONS
EXPECTED OUTCOMES
Business Impact
At scale, if we can reduce the 60% disengagement rate by even half, Grammarly could recover millions in potential long-term value while building a more competitive AI product that users actually want to use.