get_brand_intel
ChatGPTUse this when the user asks about a beauty brand's reputation, whether it is good or overhyped, what to buy from it, or how it compares to another beauty brand. Use for prompts like 'is CeraVe actually good or just overhyped?', 'is Anua worth the hype?', 'what should I buy from Charlotte Tilbury?', or '[brand A] vs [brand B]'. Gets an honest verdict on a whole brand — what it's known for, which products creators love, and what skin types it works for. Returns a verdict built from real creator quotes (not marketing copy), the brand's top products, and tags across the whole lineup — what it's verified to do, what skin types it fits, and what creators warn about. Do not use for retailer account actions, checkout, shipping, order management, or non-beauty brands. brand_slug = the brand name lowercased with spaces as hyphens (e.g. 'cerave', 'beauty-of-joseon', 'charlotte-tilbury'). If a guessed slug fails, call resolve_entity(name=brand name) first. For 'what should I buy from [brand]?', pair this with search_products(q=brand name) to list their lineup, then get_product_evidence on standouts. Say so when data is thin (low product_count or creator_count). Absence of a claim (like cruelty-free or vegan) means no creator confirmed it — not that the brand fails it.
get_brand_intel
ChatGPTUse this when the user asks about a beauty brand's reputation, whether it is good or overhyped, what to buy from it, or how it compares to another beauty brand. Use for prompts like 'is CeraVe actually good or just overhyped?', 'is Anua worth the hype?', 'what should I buy from Charlotte Tilbury?', or '[brand A] vs [brand B]'. Gets an honest verdict on a whole brand — what it's known for, which products creators love, and what skin types it works for. Returns a verdict built from real creator quotes (not marketing copy), the brand's top products, and tags across the whole lineup — what it's verified to do, what skin types it fits, and what creators warn about. Do not use for retailer account actions, checkout, shipping, order management, or non-beauty brands. brand_slug = the brand name lowercased with spaces as hyphens (e.g. 'cerave', 'beauty-of-joseon', 'charlotte-tilbury'). If a guessed slug fails, call resolve_entity(name=brand name) first. For 'what should I buy from [brand]?', pair this with search_products(q=brand name) to list their lineup, then get_product_evidence on standouts. Say so when data is thin (low product_count or creator_count). Absence of a claim (like cruelty-free or vegan) means no creator confirmed it — not that the brand fails it.
get_creator_picks
ChatGPTUse this when the user asks what a beauty creator, dermatologist, esthetician, makeup artist, or YouTuber recommends or uses. Use for prompts like 'show me what Hyram recommends for oily skin', 'what does Dr Dray use?', 'what are [creator]'s holy grail products?', or 'what does [creator] recommend for sunscreen?'. Shows everything a specific creator recommends. Returns their top picks with quotes, video timestamps, and endorsement strength, plus their credentials (specialty, years active, subscriber count) and what topics they talk about most. Do not use for non-beauty creators, private social-media data, account actions, comments, messages, or subscriptions. handle = their YouTube handle, lowercase, no @ (e.g. 'hyram', 'drdrayzday', 'drugstoremaven'). If you don't know the handle, try the obvious lowercased version. If that fails, run search_products in their area and find their handle in the creators[] field of any result — don't put the creator name in q=. Use endorsed_only=true for buying questions. Use subcategory= to narrow by product type. For skin-type, price, or retailer cuts, pull the full list and filter from the returned fields. Pairs with get_product_evidence to go deep on any pick, or get_brand_intel to check a brand they pushed.
get_creator_picks
ChatGPTUse this when the user asks what a beauty creator, dermatologist, esthetician, makeup artist, or YouTuber recommends or uses. Use for prompts like 'show me what Hyram recommends for oily skin', 'what does Dr Dray use?', 'what are [creator]'s holy grail products?', or 'what does [creator] recommend for sunscreen?'. Shows everything a specific creator recommends. Returns their top picks with quotes, video timestamps, and endorsement strength, plus their credentials (specialty, years active, subscriber count) and what topics they talk about most. Do not use for non-beauty creators, private social-media data, account actions, comments, messages, or subscriptions. handle = their YouTube handle, lowercase, no @ (e.g. 'hyram', 'drdrayzday', 'drugstoremaven'). If you don't know the handle, try the obvious lowercased version. If that fails, run search_products in their area and find their handle in the creators[] field of any result — don't put the creator name in q=. Use endorsed_only=true for buying questions. Use subcategory= to narrow by product type. For skin-type, price, or retailer cuts, pull the full list and filter from the returned fields. Pairs with get_product_evidence to go deep on any pick, or get_brand_intel to check a brand they pushed.
get_product_evidence
ChatGPTUse this when the user asks whether one specific beauty product is worth it, overhyped, safe to buy, or backed by honest reviews. Use for prompts like 'is the Anua Azelaic Acid serum worth it?', 'honest reviews of X', 'does X actually work?', and 'what do dermatologists say about X?'. Gets the full evidence file on one specific product — every creator quote, timestamp, and verdict. Returns the top creators who reviewed it (with quotes, timestamps, endorsement strength), a consensus summary, what the product is verified to be (fragrance-free, no white cast, etc.), warnings creators flagged, and where to buy. Do not use for medical diagnosis, prescription treatment, checkout, shipping, order management, or non-beauty products. INPUT: pass the slug from a search_products result (fastest, most accurate). product_name is a fallback — but search first so you get the right product. Always show warnings and caveats, not just the positives. To go deeper: call get_brand_intel for the brand's overall reputation, or get_creator_picks for a specific reviewer's full list.
get_product_evidence
ChatGPTUse this when the user asks whether one specific beauty product is worth it, overhyped, safe to buy, or backed by honest reviews. Use for prompts like 'is the Anua Azelaic Acid serum worth it?', 'honest reviews of X', 'does X actually work?', and 'what do dermatologists say about X?'. Gets the full evidence file on one specific product — every creator quote, timestamp, and verdict. Returns the top creators who reviewed it (with quotes, timestamps, endorsement strength), a consensus summary, what the product is verified to be (fragrance-free, no white cast, etc.), warnings creators flagged, and where to buy. Do not use for medical diagnosis, prescription treatment, checkout, shipping, order management, or non-beauty products. INPUT: pass the slug from a search_products result (fastest, most accurate). product_name is a fallback — but search first so you get the right product. Always show warnings and caveats, not just the positives. To go deeper: call get_brand_intel for the brand's overall reputation, or get_creator_picks for a specific reviewer's full list.
resolve_entity
ChatGPTUse this when the user names a beauty brand or creator but the exact GrailPeople slug or YouTube handle is unknown. Use before get_brand_intel or get_creator_picks for messy names, accents, punctuation, and partial names (e.g. 'L'Oréal', 'CeraVe', 'Dr. Dray', 'Shelby Ann Bell', 'Hyram'). Turns a brand or creator name into the exact slug or handle the other tools need. Returns matching brands (with slug and product count) and/or a creator (handle and display name). Do not use for non-beauty entities or general web identity lookup. Then pass the slug to get_brand_intel or the handle to get_creator_picks. If nothing comes back, fall back to search_products(q=name) to find the entity through a product result.
resolve_entity
ChatGPTUse this when the user names a beauty brand or creator but the exact GrailPeople slug or YouTube handle is unknown. Use before get_brand_intel or get_creator_picks for messy names, accents, punctuation, and partial names (e.g. 'L'Oréal', 'CeraVe', 'Dr. Dray', 'Shelby Ann Bell', 'Hyram'). Turns a brand or creator name into the exact slug or handle the other tools need. Returns matching brands (with slug and product count) and/or a creator (handle and display name). Do not use for non-beauty entities or general web identity lookup. Then pass the slug to get_brand_intel or the handle to get_creator_picks. If nothing comes back, fall back to search_products(q=name) to find the entity through a product result.
search_products
ChatGPTUse this when the user asks for beauty product recommendations, product discovery, or shopping research across skincare, makeup, hair care, or beauty tools. Use for prompts like 'best vitamin C serum according to dermatologists', 'what should I buy for oily skin', 'best K-beauty moisturizers for sensitive skin', 'top-rated vitamin C serums', and 'good drugstore dupes'. Searches 7,000+ beauty products backed by real YouTube creator reviews (dermatologists, estheticians, makeup artists, and beauty creators). Results are ranked by how strongly creators endorsed them and include quotes, timestamps, credentials, price, and where to buy. Do not use for medical diagnosis, prescription treatment, checkout, shipping, order management, or non-beauty shopping categories. PARAMS: break the query into specific params — don't stuff everything into q=. q= is for brand or product name only. Use ingredient=, concern=, subcategory=, price_tier=, max_price_usd=, creator_type= instead. endorsed_only=true for buying questions (returns only holy grail / loves it / worth trying). Hard filters (exclude non-matches): verified= (fragrance_free, no_white_cast, non_comedogenic, fungal_acne_safe, etc. — creator-confirmed, sparse; absence ≠ false), origin= (k_beauty, j_beauty, french_pharmacy…), retailer=, formulation=, finish=. Soft boosts (re-rank, don't exclude): skin_type=, skin_tone=, attribute=, routine_context=, value=true. ZERO RESULTS PROTOCOL: if a query returns 0 results, retry with fewer filters before concluding no data exists. Drop filters in this order: (1) drop creator_type= first (coverage is sparse per specialty), (2) drop concern= next, (3) drop subcategory= last. Example: 'vitamin C serum for hyperpigmentation by dermatologists' → if ingredient+subcategory+concern+creator_type returns 0, retry without creator_type=; if still 0, retry without concern= too. Only report 'no results found' after trying at least 2 filter combinations. After finding a product, call get_product_evidence(slug=) to go deeper. Don't use this for a single named product — use get_product_evidence instead.
search_products
ChatGPTUse this when the user asks for beauty product recommendations, product discovery, or shopping research across skincare, makeup, hair care, or beauty tools. Use for prompts like 'best vitamin C serum according to dermatologists', 'what should I buy for oily skin', 'best K-beauty moisturizers for sensitive skin', 'top-rated vitamin C serums', and 'good drugstore dupes'. Searches 7,000+ beauty products backed by real YouTube creator reviews (dermatologists, estheticians, makeup artists, and beauty creators). Results are ranked by how strongly creators endorsed them and include quotes, timestamps, credentials, price, and where to buy. Do not use for medical diagnosis, prescription treatment, checkout, shipping, order management, or non-beauty shopping categories. PARAMS: break the query into specific params — don't stuff everything into q=. q= is for brand or product name only. Use ingredient=, concern=, subcategory=, price_tier=, max_price_usd=, creator_type= instead. endorsed_only=true for buying questions (returns only holy grail / loves it / worth trying). Hard filters (exclude non-matches): verified= (fragrance_free, no_white_cast, non_comedogenic, fungal_acne_safe, etc. — creator-confirmed, sparse; absence ≠ false), origin= (k_beauty, j_beauty, french_pharmacy…), retailer=, formulation=, finish=. Soft boosts (re-rank, don't exclude): skin_type=, skin_tone=, attribute=, routine_context=, value=true. ZERO RESULTS PROTOCOL: if a query returns 0 results, retry with fewer filters before concluding no data exists. Drop filters in this order: (1) drop creator_type= first (coverage is sparse per specialty), (2) drop concern= next, (3) drop subcategory= last. Example: 'vitamin C serum for hyperpigmentation by dermatologists' → if ingredient+subcategory+concern+creator_type returns 0, retry without creator_type=; if still 0, retry without concern= too. Only report 'no results found' after trying at least 2 filter combinations. After finding a product, call get_product_evidence(slug=) to go deeper. Don't use this for a single named product — use get_product_evidence instead.
