# AI Guidelines ## Key Principles * AI is a tool for acceleration, not a replacement for original analysis and writing. * Start with low-risk uses and expand only with proven reliability. * All AI outputs must be validated by humans before going to clients. * External-facing work must never read as AI-generated. * Let all members of a project know when you’re using AI so data can be double-checked. * Internal communications generated by AI should still be edited to be concise and accurate. ## Tools * We have business accounts with Claude and ChatGPT. Contact for access if you’d like to use them. * These business accounts ensure that our inputs are not used for training data. * Feel free to test other tools on your own, but confirm with management before bringing them into production. ## Prompting * Be specific and detailed in your prompts. * Save recurring prompts in .txt or other note documents for later use. * Ask AI how to improve your prompts. * Be ambitious! ## Best Use Cases ### Preliminary Research * Get quick background on new topics or professional roles * e.g., Understanding what metrics are most important for ad purchases * Great for understanding what’s important to your target audience * Learn about key players in a market or genre * Get baseline assumptions for key benchmarks * e.g., Expected conversion rates, churn rates, etc. ### Planning * Make a plan for tackling a problem or analysis * Outline steps for how to get data and how to process it * Describe criteria for evaluating data * Create documentation for project plan ### Data Collection * Pulling data from Sensor Tower via API key, L\&G via API, and public sources * NOTE: Sensor Tower’s API delivers revenue in cents. AI often mistakes this for dollars. Make sure you validate revenue data pulled via ST API. * Scraping data from the internet and social media * Sometimes AI is better used to write Python scripts to scrape data rather than to do the scraping itself * Gut-check, back-of-the-napkin math ### Data Management * Writing AppScripts and cell formulas for Google Sheets. * Cleaning data or identifying inconsistencies. * Bucketing data into categories. ### Data Analysis and Pattern Recognition * Identifying trends from datasets * Summarizing initial takeaways from raw data ### Report Structure and Organization * Guiding structural outlines for reports * Suggesting logical flow and section ordering * Identifying gaps in coverage or analysis * Reorganizing existing drafts for clarity * Identifying preliminary points for summaries ### Editing * Proofreading for spelling and grammar issues * Checking for logical inconsistencies * Identifying missed opportunities to tighten narratives ## Restricted Use Cases ### Private Contracts * Verify whether contracts have any restrictions on AI usage at the start of any private contracts ### Writing and Insights * AI should not write final text for deliverables * All writing that reaches clients (including messages and emails) must be edited by a human at a minimum ### Signs To Stop Using AI * When outputs require more editing than writing from scratch * When validation is taking longer than manual data collection ## Validation ### Data * Document the source and method for all AI-assisted data collection * Check samples of any AI-pulled data against primary sources * Flag and investigate any data point that seems inconsistent * Responsibility for accuracy of AI-pulled data is shared between the person who sourced the data and the project’s lead ### Deliverables * Ensure no text reads as AI-generated (generic phrasing, repetitive structure, excessive em dashes) * Look out for confident claims with no clear sourcing * Verify every data point --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://faq.liquidandgrit.com/extras/ai-guidelines.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.