Glossary
Buzzwords, translated and how they're used
Combining external data with AI to generate accurate, contextual outputs: AI searches a knowledge base before generating, reducing hallucinations.
Upload 12 interview transcripts to NotebookLM. Ask "What are the three most common pain points?" It searches all 12 documents before responding: the answer is grounded in actual quotes, not guesswork.
Fine-tuning AI with human preference feedback for better alignment: teaches AI via feedback loops to improve accuracy and helpfulness.
When you ask Claude to "explain this simply" and it correctly adjusts its register without over-simplifying, that calibration comes from RLHF training: thousands of human preference signals teaching it what "simple but accurate" actually looks like.
AI systems that understand and generate human-like text, enabling natural conversations, copy generation, and content automation at scale.
When you write "rewrite this in a warmer tone" and the AI understands you mean register, not temperature, and makes exactly the right adjustments: that's NLP-AI reading intent from language context.
AI that analyzes visual content, detecting elements, objects, or scenes: helps AI "see" and understand images. Enables design review, accessibility, and automation.
Upload a Figma export to Claude: "What accessibility issues can you spot from this screen?" CV-AI identifies contrast problems, missing labels, and unclear interactive states, before engineering ever touches it.
Customizing pre-trained AI models with your domain data: making AI models learn your style and data for higher quality, brand-specific outputs.
A content team fine-tunes a base model on 5 years of their destination copy. The model now defaults to their established tone without explicit instruction, reducing prompt complexity for every future use.
Measuring AI response time and efficiency: checking how fast and efficient AI is. Helps optimize AI usage and cost for real-time product features.
Testing three models for an AI search feature: GPT-4o at 800ms average, Claude Haiku at 400ms, GPT-4o-mini at 300ms. MLE data shows Haiku hits the 500ms UX threshold consistently, so that's what ships.
Reviewing AI outputs for brand tone, accuracy, bias, and consistency: an AI output health check. Maintains brand consistency and trust at scale.
After generating 50 product descriptions with AI, AIQ checks each against: brand voice guidelines, factual accuracy against product data, and bias checklist. Flags 8 that need revision before the batch goes to the content team.
Managing, versioning, and testing AI prompts for scale and reliability: a toolbox for creating and managing AI prompts. Scales AI content and experimentation.
The team's copy generation prompt is at v2.3. A team member proposes a new version: it goes through GPT-ops: tested against the standard input set, compared against v2.3 quality metrics, approved by the workflow owner, then merged into the shared library.
Post-processing AI outputs to remove noise, bias, or irrelevant content: filtering AI results before showing them. Reduces errors and improves output quality.
A product description generator runs ARF automatically: checks each output against brand voice rules, flags descriptions mentioning competitor names, and removes any output below a confidence threshold, before the batch lands in the content team's queue.
Iteratively improving prompts with feedback from multiple contexts: teaching AI better with real-world examples. Better outputs, fewer iterations.
Testing a new research synthesis prompt across 5 different interview datasets: discovery research, usability testing, concept testing, competitor analysis, and survey data. MCPR refines until it handles all 5 reliably before it goes into the shared library.
Automating repeated AI prompt workflows: AI runs repetitive prompt tasks itself. Saves time and ensures consistency at scale.
Every Monday at 9am, IPA pulls new product listings from the database, runs them through the copy generation prompt, applies ARF filtering, and deposits approved outputs in the content team's Notion queue, ready for final review.
Generating multiple design or copy variations automatically: lets AI create multiple options in seconds. Increases creative speed and productivity.
Campaign brief goes in. AGI produces: 8 headline options across 3 tones, 6 Midjourney prompt directions, and 4 CTA variants. Creative director selects the direction in 20 minutes instead of waiting 2 days for the team to brainstorm.
AI interprets data and outputs actionable steps: AI reads context and tells you what to do next. AI becomes actionable, not just informative.
You paste last quarter's support ticket analysis into Claude. C2A prompt: "Based on these patterns, give me 5 specific product changes ranked by impact and effort. Include acceptance criteria for each." You get a prioritised action list, not a summary.
Generating images from text prompts: write a description, get an image. Enables rapid visual prototyping and creative territory exploration.
Campaign brief calls for 3 distinct creative territories. T2I generates 6 hero image options per territory in 15 minutes. Creative director reviews 18 options and selects a direction, before a single briefing call with the photography team.
Transforming or refining existing images via AI: AI edits your image based on a prompt. Speeds up design iteration without starting from scratch.
You have a Midjourney image that's 80% right: the composition is good but the lighting is wrong. I2I: upload the image, prompt "warm golden hour light, remove harsh shadows, keep composition identical." Get 4 refined versions in 30 seconds.
Using AI to detect patterns from multiple design or content samples: AI finds common design patterns for you. Speeds up discovery of reusable solutions.
Before starting a design system, DPE analyses 200 existing screens. Output: 24 distinct component patterns, 11 layout templates, and 8 interaction patterns. Designer turns this into the design system brief instead of inventorying manually.
AI detects key visual elements from designs or images: AI finds and labels important parts of visuals. Improves insights from visual data.
VFE on 10 competitor checkout flows extracts: CTA placement, trust signal positioning, form field patterns, and progress indicator styles. Designer gets a structured comparison matrix without manually screenshotting and noting each one.
AI automatically labels flows, variants, or components: AI tags things intelligently based on context. Speeds up organisation and design workflows.
Running CLS on a Figma file with 340 unnamed frames: AI suggests structured names for each based on content analysis. Designer reviews, accepts 280, adjusts 60, instead of manually naming all 340 from scratch.
Analyzing user recordings or analytics for AI-driven insights: AI watches and understands user behavior. Accelerates UX research and understanding.
Upload 40 session recordings from the checkout flow. AI-POI analysis: "Users consistently hesitate at the delivery date field: 73% pause 4+ seconds before interacting. Suggested fix: add date format example inline."
AI suggests the next best user action: AI predicts what a user will do next. Increases conversion, retention, and engagement.
POI-AI analyses a user's 3-session browsing pattern (families section, duration 3–4 hours, price sensitivity medium), and predicts: "Surface hop-on hop-off tours and skip-the-line family packages on next session home screen." Conversion increases 18%.
A standard that lets AI models connect to external tools, files, and data, like USB-C for AI.
With MCP configured, Claude can read your Figma file, check Notion for the PRD, write updated copy, and mark the task complete, all in one conversation. No tab switching. It's what turns a chatbot into an actual workflow tool.
Integrating AI from ideation to deployment in a seamless pipeline, from start to finish. Streamlines complex AI-driven processes.
New destination launch: AI research synthesis → AI generates 6 campaign territories → AI writes hero copy for each → AI quality audits all outputs → human creative director selects and approves → publishes. 2 days instead of 2 weeks.
Measuring effectiveness of AI outputs across iterations: metrics to check if AI is helping or not. Helps validate AI impact and ROI.
AIM dashboard shows: prompt v1 brand voice score 58%, v2 71%, v3 89%. Time to acceptable output: v1 avg 3.2 revisions, v3 avg 1.1 revisions. This data justifies continued investment in prompt refinement to leadership.
Feeding structured data into AI for output generation: giving AI spreadsheets or databases to work with. Improves relevance and factual accuracy.
DAG for product descriptions: feed AI the full product database (name, location, duration, inclusions, user ratings, common questions). Every description is factually accurate and specific, not generic. 500 descriptions generated in 2 hours vs 2 weeks of manual writing.
Summarizing, tagging, or cleaning user content at scale with AI: let AI handle mass user reviews, photos, or moderation. Enhances moderation and engagement.
UGC-AI processes 45,000 tour reviews monthly: classifies by sentiment and product area, extracts top 10 recurring praise themes and top 10 complaint themes, flags reviews needing operator response. Operations team gets an actionable digest every Monday.
AI condenses articles or pages into short insights: turns long content into bite-size summaries. Saves time and improves knowledge access.
Weekly: 12 industry newsletters, 3 competitor blog posts, 2 research papers. LFC produces a single 500-word digest with: top 5 trends, 3 competitor moves to watch, and 2 research insights relevant to the roadmap. Distributed to the team every Monday.
Extracting text from images or screenshots for automation: let AI read pictures for text. Saves manual typing and speeds up data tasks.
Team photographs a full whiteboard ideation session after a workshop. OCR-AI extracts all text, preserves the groupings, and converts into a structured Notion doc. What used to take 2 hours of manual transcription takes 3 minutes.