Projects / typeahead.ai / Research Report 2026-04-12

Research Report

typeahead.ai · Full Findings

Confidence: 75/100

Synthetic data — directional for product decisions, validate before launch.

1. Executive Summary

Product-Market Fit

41.5% "very disappointed" (Superhuman method) — crossing the 40% threshold. An additional 37.2% "somewhat disappointed" represents a large swing segment convertible with better messaging and onboarding.

Top 3 Actionable Findings

"Works in every app" is the killer feature. Ranked #1. Browser/Electron app support is the highest-impact technical investment.

One-time purchase at $49–$79 is optimal. 86% chose one-time. Zero chose subscription. $49–$79 band captures 51% of demand.

"Learns your voice" is the #1 value claim. Every interview participant independently prioritized voice learning over grammar correction.

Top 3 Risks

Accuracy kills trust fast. Wrong suggestions in client-facing communication are "worse than no suggestion."

"Does it work in my app?" will dominate objections. Universal app support is an anxiety point, not an assumption.

Privacy positioning has a ceiling. Users want quality AND privacy. Don't lead with privacy at the expense of demonstrating quality.

Recommended Next Steps

  1. Validate with real users: Van Westendorp pricing survey (n=100+) and 5–8 interviews
  2. Ship browser/Electron app support before native-only polish
  3. A/B test landing page: productivity-forward vs. privacy-forward messaging

2. Research Context

What Was Studied

Market demand for an AI autocomplete tool for macOS — feature priorities, pricing preferences, competitive perceptions, privacy attitudes, and segment-specific needs.

Methodology

Three research streams, all synthetic:

  1. Personas (8): Extracted from ~40 real HackerNews comments (2022–2026) on adjacent products
  2. Interviews (10): Synthetic interviews covering VP Sales, developer, researcher, privacy activist, content creator, support lead, UX designer, tech writer, consultant, copywriter
  3. Survey (100 respondents, 94 completed): PMF question, feature ranking, WTP, pricing model preference, competitor usage, privacy attitudes

Limitations

  • All data is synthetic — real responses will differ
  • Platform bias: 100% HN voices (skews technical, male, North American)
  • Geographic bias: North America / Europe dominant
  • No real user validation — pricing predictions are directional only
  • Missing voices: non-technical writers, older demographics, enterprise IT

3. Key Findings

Finding 1: "Works in every app" is the #1 feature request

Universality across all Mac text fields is the single most demanded feature, outranking privacy, speed, and price.

  • Feature ranking: avg rank 2.17 (1st of 5)
  • 6/10 interview participants named Google Docs as primary writing app
  • 59% use Google Docs; 88% use email clients; 69% use Slack/Teams

Recommendation: Prioritize browser extension or accessibility-based approach. Google Docs support is the single highest-impact technical feature.

Finding 2: "Learns your voice" is the highest-value differentiator

Users don't want AI to write better — they want AI to write like them, faster.

  • Feature ranking: avg rank 2.22 (2nd of 5)
  • Anika (researcher): "I don't want to sound like a native speaker. I want to sound like me, but clearer."
  • Alex (copywriter): "They all try to write FOR me instead of WITH me."

Recommendation: Lead with "learns your voice" on the landing page. Avoid "improve your writing" — users hear "change your writing."

Finding 3: One-time purchase is non-negotiable

86% chose one-time purchase. Zero chose subscription. The strongest signal in the entire dataset.

  • Priya (creator): "I pay for like eight subscriptions. The question is how much."
  • Alex (copywriter): "What am I paying for monthly if the model is running on my machine?"

Recommendation: Ship at one-time purchase. Consider "pro" tier or paid updates as revenue expansion.

Finding 4: Accuracy is the primary trust factor

Users will abandon the tool faster for wrong suggestions than for missing suggestions.

  • Sarah (VP Sales): "If it suggests something and I don't catch it... that's my reputation."
  • Marcus (developer): "If it's right 70% of the time, I'll probably turn it off."
  • Daniel (tech writer): "Better to suggest nothing than to suggest something wrong."

Recommendation: Default to conservative suggestions. Add user-adjustable confidence setting.

Finding 5: Privacy is a genuine purchase driver

Privacy-conscious users have significantly higher purchase intent. Local-first is trusted (4.02/5), cloud is distrusted (2.14/5).

  • High-privacy respondents (4–5 score) = 54% "very disappointed" vs. 0% for low-privacy
  • Jim (privacy activist): "I've read the privacy policies. They're all the same."

Recommendation: Lead with productivity messaging, embed privacy as a supporting pillar. Don't make it the headline — make it the reassurance.

Finding 6: Text expansion users are the ripest market

41% use text expanders. They're pre-sold on the concept — they just need a smarter tool.

Recommendation: Create migration content: "Import your Espanso/TextExpander snippets." Position as an upgrade.

Finding 7: Non-native English speakers are underserved

They want assistance that preserves their voice, not erases it. Currently paying Grammarly ($12/mo) reluctantly.

Recommendation: Create specific messaging for non-native English users. Consider multi-language support as v2.

Finding 8: The "somewhat disappointed" 37% is the growth opportunity

Casual users (4–6 hrs typing) who see value but aren't fully convinced. WTP clusters at $39–$49.

Recommendation: Generous free trial (14 days, full features). Value demonstration within 5 minutes. Price at $49.

Finding 9: Team/enterprise use is real but blocked by IT

Several participants expressed interest in team deployment, but IT approval is the primary barrier.

Recommendation: Defer enterprise features. Build a security whitepaper for future use.

Finding 10: Users want AI to be invisible, not impressive

Even creative professionals don't want AI to be creative. They want it to suggest what they'd type — boring, accurate, invisible.

Recommendation: Optimize for next-word/next-phrase prediction, not paragraph generation.

4. Target Market Analysis

Primary Beachhead: Power Users (39%)

Type 6+ hours/day. Mostly developers (30%) and writers (16%). Already use 2–4 productivity tools. Highest WTP ($79–$129). Strongest privacy concerns.

Privacy-first developers: Want local + system-wide. Easy sell.
RSI-suffering managers: Health motivation. Highest urgency.
Content creators: Voice preservation critical. $99 is easy.
Non-native English pros: Currently paying Grammarly sub. $79 one-time is a no-brainer.

Persona Summary

# Name Segment Key Motivation WTP Adoption
1MarcusPrivacy-First DevLocal + system-wide$49-9985%
2PriyaFreelancerOne-time + voice$49 (99 ⚠️)80%
3DavidRSI ManagerReduce keystrokes$9990%
4KenjiNon-Native EnglishPrivacy + natural English$79-9988%
5ElenaIndie HackerZero-config + adaptable$79-9982%
6SarahEnterprise ConsultantIT-approved + web apps$99 (expense)55%
7TylerStudentCheap + Obsidian$29-4970%
8JordanContent CreatorVoice learning + web$9985%

5. Feature Prioritization

Must-Have (Ship with v1)

Feature Evidence Priority
Universal app support#1 ranked (2.17 avg). 59% use Google Docs.P0
Learns writing style#2 ranked (2.22 avg). Every interview prioritized.P0
100% local processing#3 ranked (2.73 avg). 4.02/5 trust local.P0
Sub-200ms latency"If it takes more than 200ms, I've already typed the next word."P0
Keyboard-dismissable suggestions"One key to dismiss, no mouse. Table stakes."P0

Should-Have (v1.1)

Feature Evidence
Import Espanso/TextExpander snippetsRachel and Daniel both have 100+ snippets. Easiest conversion path.
Per-app enable/disableJim (privacy): "If I'm writing under NDA, I want to be damn sure no AI is learning from it."
Confidence/aggressiveness settingAnika: "It should know when to help and when to stay out of the way."
Google Docs specific support59% of respondents. Emily: "It has to work in Google Docs in Chrome."

Nice-to-Have (v2+)

Feature Evidence
Team/shared voice profilesTomás: "I'd want it for all 12 agents."
Multi-language support"multi-language support, especially for non-english writers" (P020)
Terminal/vim support"none of them work in terminal" (P036)
Analytics/acceptance trackingTomás: "I need to justify the purchase to leadership."
Custom vocabulary"support for custom vocabulary and technical terms" (P031)

6. Pricing Analysis

Willingness-to-Pay Distribution

Price % of Respondents Segment
$7920%Power users (devs, writers)
$4916%Mixed (strongest single point)
$9915%High-intent power users
$3915%Casual users, students
$2915%Students, budget-constrained
$599%Mid-range (gap in data)
$196%Skeptical, low-intent
$129+3%Enthusiasts

Recommended: $49 for launch, with a $79 "pro" tier. $49 captures the largest segment. $79 captures high-intent power users. The $49–$79 band covers 51% of demand. One-time purchase eliminates the subscription objection entirely.

7. Competitive Positioning

Competitor Usage (Survey)

Tool % Tried Key Weakness
Apple Intelligence62%Generic suggestions, limited app support
Grammarly51%Cloud-based, changes voice, subscription
GitHub Copilot32%IDE-only, not for prose
Espanso32%Manual setup, static snippets
TextExpander27%Expensive subscription, dated UI

Where typeahead.ai Wins

Universal app support — No competitor does this well

Local-first privacy — No cloud dependency

Voice learning — No competitor preserves your specific writing voice

One-time purchase — Unique in the market

Where typeahead.ai Is Vulnerable

Quality of local models — Users skeptical that local inference can match cloud quality

Brand trust — Grammarly and Apple have established trust

Google Docs support — Technically hard; if unsolved, loses to browser extensions

Apple Intelligence evolution — Window is now

8. Messaging Recommendations

Lead With

  1. "Works in every app" — Show it, don't just say it
  2. "Learns your voice" — "Sounds like you, not AI"
  3. "Pay once, yours forever" — Compare to $144/yr Grammarly
  4. "100% private, 100% local" — "Nothing leaves your Mac"

Segment-Specific Messages

Segment Lead Message Avoid
Developers"Copilot for everything outside your IDE"Don't call it a "writing tool"
Content creators"Your voice, faster"Don't mention AI generation
Non-native English"Your English, your way"Don't say "improve your English"
Freelancers"One-time purchase. No subscriptions. Ever."Don't show monthly pricing
RSI sufferers"Type less, say more"Don't over-medicalize
Students"The tool that pays for itself in a week"Don't ignore budget sensitivity

Messages That Fall Flat

  • "AI-powered" — Overused, meaningless, off-putting to privacy-conscious users
  • "Improve your writing" — Users hear "change your writing" and reject it
  • "Like having a co-writer" — Creepy for some, inaccurate for others
  • "Replace Grammarly" — Too confrontational

9. Opportunity Map

# Opportunity Impact Effort Next Step
1Browser/Electron supportHighHighTechnical spike on accessibility approach
2Espanso/TextExpander importHighLowBuild import tool + migration guide
3Demo videoHighMediumRecord completions in Gmail, Slack, Docs, VS Code
4Generous free trial (14-day)HighLowImplement trial, track activation
5"Learns your voice" landing testMediumLowA/B test hero messaging
6Non-native English messagingMediumLowDedicated landing page variant
7Confidence/aggressiveness sliderMediumMediumUX design + implementation
8Student pricing tier ($29–$39)MediumLowStudent discount + referral tracking
9Team voice profilesLowHighDefer. Build security whitepaper first.
10Terminal/vim supportLowMediumPost-launch niche feature

10. Appendix

Survey Methodology

TypeSynthetic survey, Superhuman PMF framework
N100 respondents, 94 completed (94% completion rate)
Avg completion189 seconds
Questions21 (screening, PMF, feature ranking, pricing, privacy, competitors, open-ended)
Segments39% power user, 35% casual, 19% skeptical, 1% disengaged, 6% incomplete

Confidence Assessment

Dimension Score Notes
Evidence quality18/25~40 real HN voice samples
Data provenance15/20Primary sources but single platform
Internal consistency18/20No contradictions across data
External validation12/20No real user interviews or survey data
Predictive power12/157 testable predictions
Total75/100Medium — directional for product decisions

To Reach 90% Confidence

  1. Run 5–8 real user interviews across primary personas
  2. Van Westendorp pricing survey with n=100+ real respondents
  3. Beta test with 30–50 real users, track completions per app and acceptance rates
  4. Validate "learns your voice" feature interest vs. other features
  5. A/B test landing page messaging with real traffic
Generated 2026-04-12 ← Back to Project Hub