Copy coach
Built a “Copy Coach” GPT that turns Deputy UI screenshots into tone-aligned copy options, adopted by >50% of the design team and reducing copy-related feedback in design reviews.
YEAR
2025
ROLE
Product Design Manager
COMPANY
Deputy
INDUSTRY
B2B SAAS
Designers needed a fast, reliable way to refine UI copy
Copy was written late and was inconsistent, leading to a continuous loop of rewrites and reviews and causing delivery delays.
Manual copy entry was too slow; building a Figma plugin/MCP integration was too heavy for the problem
How do we make our tone standards executable and output immediately usable in design files?
Optimised guidance for LLM reliability
I took Deputy’s existing tone-of-voice guidelines and translated them into a structured Markdown ruleset
This includes do/don’t rules to calibrate outputs
Key content principles have been turned into repeatable checks (brevity, task-first language, consistent terminology)
Turned three inputs into one consistent output
The user uploads a screenshot of their design, it is combined with Deputy's tone of voice guidelines and examples of good/bad copy to come up with copy suggestions
Generated rewrites are grouped by UI element (titles, labels, helper text, CTAs, errors)
Multiple options are always given so designers can decide which direction is more appropriate
Screenshots were the sweet spot input method
Avoided manual copy entry (too high effort) and avoided a plugin/MCP build (too complex)
Preserved layout context (hierarchy, labels, intent) with minimal friction
Made it easy to run repeatedly during iteration
Embedded into the design process to reduce copy feedback loops
Positioned as a pre-review step to catch copy issues before critique
Created a repeatable loop: draft UI → run GPT → paste + adjust → review
Reduced copy feedback without adding meetings or dependencies
Outcomes
50%
Team adoption over 2 weeks
Reduced
copy-related feedback during design reviews
Faster
copy refinement without heavyweight tooling
Smooth Scroll
This will hide itself!
This will hide itself!
Copy coach
OVERVIEW
Built a “Copy Coach” GPT that turns Deputy UI screenshots into tone-aligned copy options, adopted by >50% of the design team and reducing copy-related feedback in design reviews.
YEAR
2025
ROLE
Product Design Manager
COMPANY
Deputy
INDUSTRY
B2B SAAS
Designers needed a fast, reliable way to refine UI copy
Designed the end-to-end WFM flow:
Roster → timesheets → pay run → payslipsMade the first pay run easy to complete (guided steps + clear next actions)
Surfaced value fast with confident review states (so teams can onboard without support)
Optimised guidance for LLM reliability
Designed the end-to-end WFM flow:
Roster → timesheets → pay run → payslipsMade the first pay run easy to complete (guided steps + clear next actions)
Surfaced value fast with confident review states (so teams can onboard without support)
Turned three inputs into one consistent output
Designed the end-to-end WFM flow:
Roster → timesheets → pay run → payslipsMade the first pay run easy to complete (guided steps + clear next actions)
Surfaced value fast with confident review states (so teams can onboard without support)
Screenshots were the sweet spot input method
Avoided manual copy entry (too high effort) and avoided a plugin/MCP build (too complex)
Preserved layout context (hierarchy, labels, intent) with minimal friction
Made it easy to run repeatedly during iteration
Embedded into the design process to reduce copy feedback loops
Designed the end-to-end WFM flow:
Roster → timesheets → pay run → payslipsMade the first pay run easy to complete (guided steps + clear next actions)
Surfaced value fast with confident review states (so teams can onboard without support)
Outcomes
50%
Team adoption over 2 weeks
Reduced
copy-related feedback during design reviews
Faster
copy refinement without heavyweight tooling
Smooth Scroll
This will hide itself!
This will hide itself!
Copy coach
OVERVIEW
Built a “Copy Coach” GPT that turns Deputy UI screenshots into tone-aligned copy options, adopted by >50% of the design team and reducing copy-related feedback in design reviews.
YEAR
2025
ROLE
Product Design Manager
COMPANY
Deputy
INDUSTRY
B2B SAAS
Designers needed a fast, reliable way to refine UI copy
Copy was written late and was inconsistent, leading to a continuous loop of rewrites and reviews and causing delivery delays.
Manual copy entry was too slow; building a Figma plugin/MCP integration was too heavy for the problem
How do we make our tone standards executable and output immediately usable in design files?
Optimised guidance for LLM reliability
I took Deputy’s existing tone-of-voice guidelines and translated them into a structured Markdown ruleset
This includes do/don’t rules to calibrate outputs
Key content principles have been turned into repeatable checks (brevity, task-first language, consistent terminology)
Turned three inputs into one consistent output
The user uploads a screenshot of their design, it is combined with Deputy's tone of voice guidelines and examples of good/bad copy to come up with copy suggestions
Generated rewrites are grouped by UI element (titles, labels, helper text, CTAs, errors)
Multiple options are always given so designers can decide which direction is more appropriate
Screenshots were the sweet spot input method
Avoided manual copy entry (too high effort) and avoided a plugin/MCP build (too complex)
Preserved layout context (hierarchy, labels, intent) with minimal friction
Made it easy to run repeatedly during iteration
Embedded into the design process to reduce copy feedback loops
Positioned as a pre-review step to catch copy issues before critique
Created a repeatable loop: draft UI → run GPT → paste + adjust → review
Reduced copy feedback without adding meetings or dependencies
Outcomes
50%
Team adoption over 2 weeks
Reduced
copy-related feedback during design reviews
Faster
copy refinement without heavyweight tooling
Smooth Scroll
This will hide itself!
This will hide itself!