# triedagents: full case text Source: https://triedagents.com/ · Generated 2026-10-19 · Licence: CC BY 4.0: free to quote and reuse with attribution to triedagents and a link to the case page. --- ## Answer a question that needs this week's docs URL: https://triedagents.com/cases/latest-docs-question-sign-in-with-chatgpt/ Verdict: Reproduced (V2) on 2026-10-19: the official docs say Sign in with ChatGPT usage is for eligible ChatGPT Plus and Pro subscribers, while the search engine's own summary hedged wrongly. Status: V2 (Reproduced) · Access level: L0 (No account access) · Source setup: L0 Category: Research Claim (@viticci, https://x.com/viticci/status/2105457582013911394): Use an agent to answer a product question that depends on very recent documentation. Agent in source: Not stated in our record Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: Docs say eligible ChatGPT Plus and Pro subscribers. Matches the kind of answer the claim describes. Ground-truth check: Manually confirmed against the official docs page text (two official pages agree). Exact prompt: Which ChatGPT plans can use 'Sign in with ChatGPT' usage, according to the current official docs? Quote the line. Method: Web search, then fetched the two official docs pages and read the eligibility wording. Steps: - Web search for the feature name - Fetched learn.chatgpt.com/docs/sign-in-with-chatgpt - Fetched developers.openai.com/siwc/quickstart.md - Compared wording Sites and APIs touched: learn.chatgpt.com, developers.openai.com Failure modes: - The search engine's own synthesized summary hedged wrongly; only reading the source page gave the right answer. Time: About 10 s of tool time (15:40:45-15:40:55 UTC+8; timing loosely bracketed because calls ran in parallel) Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0. Any agent with web fetch. Ask it to quote the official page, not a search summary. How to cite: triedagents, "Answer a question that needs this week's docs", reproduced, checked 2026-10-19, https://triedagents.com/cases/latest-docs-question-sign-in-with-chatgpt/ --- ## Weekly meal plan from what is in your kitchen URL: https://triedagents.com/cases/weekly-meal-plan-from-pantry/ Verdict: Reproduced (V2) on 2026-10-19: a 7-dinner plan built from an 18-item pantry used 0 off-pantry ingredients and planned a next-day lunch for every dinner. Status: V2 (Reproduced) · Access level: L0 (No account access) · Source setup: L0 Category: Life admin Claim (@dkmitc, https://x.com/dkmitc/status/2109402706116858172): Told the agent what was in the cabinets and fridge; got a week of dinners that leave leftovers for lunch. Agent in source: Grok Bot (per post) Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: 0 off-pantry ingredients; 16 of 18 items used (oats and bananas unused, reasonable for dinners); every dinner had a planned next-day lunch. Ground-truth check: Programmatic check of every ingredient against the stated pantry list. Taste and quantities were not judged. Exact prompt: My kitchen has: rice, black beans, canned tomatoes, onion, garlic, eggs, spinach, chicken thighs, tortillas, cheddar, pasta, frozen peas, carrots, potatoes, lentils, yogurt, oats, bananas. Plan 7 dinners that leave leftovers for next-day lunch, using only these plus basic spices and oil. Method: Used a fictional 18-item pantry. The agent wrote a 7-dinner plan, then a script checked every ingredient against the pantry plus basic staples (oil, salt, pepper, cumin, chili powder, water). Steps: - Wrote 7 dinners - Automated constraint check Sites and APIs touched: None (no web access needed) Failure modes: - Quantities were not planned, so 'enough for leftovers' is not guaranteed. - A real kitchen inventory spoken aloud is messier than a typed list. Time: Under 1 minute Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0. Type your pantry list; ask it to flag any ingredient you did not list. How to cite: triedagents, "Weekly meal plan from what is in your kitchen", reproduced, checked 2026-10-19, https://triedagents.com/cases/weekly-meal-plan-from-pantry/ --- ## Which travel sites let an agent see fares? URL: https://triedagents.com/cases/which-travel-sites-let-agents-in/ Verdict: Reproduced (V2) on 2026-10-19: of 8 travel sites fetched without a browser, only Google Flights returned fares (lowest $1,214 round trip, SFO to Tokyo Haneda, Dec 10-20, 2026). Status: V2 (Reproduced) · Access level: L0 (No account access) · Source setup: L0 Category: Travel Claim (@SanUvacha, https://x.com/SanUvacha/status/2110617129938788483): 'The only major travel site that allows bots is Google Flights.' Agent in source: Grok Bot (per post) Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: Reproduced at the plain-HTTP level: only Google Flights returned fares (lowest seen $1,214 round trip; several Delta 1-stop options at $1,222). Every other site blocked, rate-limited or required JavaScript. A full headless browser was not tried, so this says nothing about JS-capable agents. Ground-truth check: Fares read from the saved Google Flights HTML; the 'captcha' strings in that page turned out to be inline script, not a block (the page contained full itinerary text). Other sites: HTTP status codes and saved bodies. Exact prompt: Search SFO to Tokyo Haneda, Dec 10-20, 2026, 1 adult, on Google Flights, Kayak, Expedia, Skyscanner, Booking.com, Priceline, Kiwi and United. Which ones return fares? Method: curl with a desktop Chrome user agent against 8 travel sites for the same SFO to HND round trip (Dec 10-20, 2026); counted block markers and checked for fares in the HTML. No headless browser. Steps: - Google Flights: 200, fares in the HTML - Kayak: 200, no fares (rendered by JavaScript) - Expedia: 429 - Skyscanner: 'enable JavaScript' wall - Booking.com flights: 202, 'Enable JavaScript' - Priceline: 403 with PerimeterX captcha - Kiwi: CAPTCHA/blocked markers - United: no response (connection failed, re-tested 15:54 UTC+8) Sites and APIs touched: google.com/travel/flights, kayak.com, expedia.com, skyscanner.com, booking.com, priceline.com, kiwi.com, united.com Failure modes: - 7 of 8 travel sites unusable without a real browser. - Kayak returns 200 but no data, which a careless agent could report as 'no flights'. Time: About 1 minute (15:52:28 UTC+8 run, United re-test 15:54 UTC+8) Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0. Ask your agent which site each price came from; if it is always Google Flights, it is not really comparing. How to cite: triedagents, "Which travel sites let an agent see fares?", reproduced, checked 2026-10-19, https://triedagents.com/cases/which-travel-sites-let-agents-in/ --- ## 90-day menu built from '5-star' recipes URL: https://triedagents.com/cases/90-day-menu-five-star-recipes/ Verdict: Partly reproduced (V1) on 2026-10-19: recipe sites were readable, but 0 of 8 sampled Allrecipes dinners were rated 5.0 (range 3.7-4.9), so a '5-star recipe' menu needs a stated rule. Status: V1 (Partly reproduced) · Access level: L0 (No account access) · Source setup: L0 Category: Life admin Claim (@LouOgis, https://x.com/LouOgis/status/2112012675706208308): A 90-day breakfast/lunch/dinner menu from 5-star recipes, by week, with a grocery list and freezer labels, in a few minutes. Agent in source: Grok Bot (per post) Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: Recipe sites were readable. But 0 of 8 sampled recipes were rated 5.0: values ranged 3.7-4.9; the 4.9 had only 47 ratings, and 4.6/4.4 had 5-7 ratings. A '5-star recipe' menu would need a stated rule (e.g. >=4.5 with >=100 ratings) or it is just marketing language. Ground-truth check: Ratings read from each recipe page's own structured data (schema.org aggregateRating). Exact prompt: Find dinner recipes rated 5 stars on Allrecipes and show the rating and number of ratings for each. Method: Tested the hardest-to-verify part: can an agent read real recipe ratings? Fetched Allrecipes' dinner page and 8 recipe pages, read ratings from their structured data. Did not build the full 90-day menu. Steps: - Fetched allrecipes.com/recipes/ (no block) - Fetched dinner category, 8 recipe links - Read aggregateRating from each page Sites and APIs touched: allrecipes.com, bbcgoodfood.com (access check only) Failure modes: - '5-star' is ambiguous; an agent can call anything 5-star without showing ratings. - Full 90-day plan, grocery list and labels not tested. Time: About 4 s Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0. Ask it to print rating and rating count next to every recipe. How to cite: triedagents, "90-day menu built from '5-star' recipes", partly reproduced, checked 2026-10-19, https://triedagents.com/cases/90-day-menu-five-star-recipes/ --- ## A morning brief (weather + headlines) URL: https://triedagents.com/cases/morning-brief-weather-and-news/ Verdict: Partly reproduced (V1) on 2026-10-19: weather (New York high 16.3 C, 0% rain chance) and 5 BBC World headlines worked from public feeds; the calendar half needs account access and was not tested. Status: V1 (Partly reproduced) · Access level: L0 (No account access) · Source setup: L1 (calendar read) + scheduled Category: Life admin Claim (@cdepino, https://x.com/cdepino/status/2110039250896830572): 'A morning brief with weather and your calendar' as an easy first routine. Agent in source: Grok Bot (per post) Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: Forecast for 2026-10-19: high 16.3 C, low 4.2 C, 0% rain chance; 5 headlines with publish times. The public half works; the calendar half (the personal part) is untested. Ground-truth check: Each value matched the raw API/RSS response saved in the run log. Exact prompt: Give me a morning brief for New York: today's high/low and rain chance, plus the top 5 world headlines with times. Method: Fetched today's New York forecast from Open-Meteo and the top 5 BBC World RSS headlines. The calendar part was skipped (needs account access), and no schedule was set. Steps: - Open-Meteo forecast API - BBC World RSS Sites and APIs touched: api.open-meteo.com, feeds.bbci.co.uk Failure modes: - Calendar requires L1 access, so we could not test the useful part. - News selection is whatever the feed ranks first. Time: About 2 s Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0 without calendar. Add calendar read-only (L1) later if you trust the agent. How to cite: triedagents, "A morning brief (weather + headlines)", partly reproduced, checked 2026-10-19, https://triedagents.com/cases/morning-brief-weather-and-news/ --- ## Am I owed EU261 compensation for a delayed flight? URL: https://triedagents.com/cases/eu261-flight-delay-compensation-eligibility/ Verdict: Partly reproduced (V1) on 2026-10-19: under the official EU rules, a fictional 5-hour Amsterdam to New York delay (5,847 km) qualifies for EUR 600; filing the claim (L3) was not attempted. Status: V1 (Partly reproduced) · Access level: L0 (No account access) · Source setup: L3 (files a claim as you) Category: Money Claim (@notjazii, https://x.com/notjazii/status/2110010012424896876): A personal agent that can 'claim flight delay compensation'. Agent in source: Qwen Go (announced, per post); we did not run Qwen Go Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: Eligible: departs from the EU, arrival delay >= 3 hours, crew shortage is not an extraordinary circumstance; distance 5,847 km (> 3,500 km) means EUR 600. The 'claim it for me' part (L3) was not attempted. Ground-truth check: Manually confirmed against the official Your Europe page (amount table 250 / 400 / 600 EUR by distance; 3-hour arrival delay rule). Exact prompt: Fictional: EU airline, Amsterdam to New York JFK, arrived 5 hours late because of crew shortage. Under EU rules, am I owed compensation, and how much? Method: Read the official EU 'Your Europe' air passenger rights page, then applied its rules to a fictional case with a computed great-circle distance. Did not file anything. Steps: - Fetched europa.eu Your Europe passenger-rights page - EUR-Lex regulation page returned 202 (challenge), not used - Computed AMS-JFK distance - Applied distance tier and 3-hour rule Sites and APIs touched: europa.eu, eur-lex.europa.eu Failure modes: - The EU legal text site challenged the bot, so the agent relied on the summary page. - Whether a cause is 'extraordinary' is often disputed; the agent cannot settle that. Time: About 5 s Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0 for eligibility. Filing a claim acts as you (L3): review it before it is sent. How to cite: triedagents, "Am I owed EU261 compensation for a delayed flight?", partly reproduced, checked 2026-10-19, https://triedagents.com/cases/eu261-flight-delay-compensation-eligibility/ --- ## Check if Bitcoin broke out on the 4-hour chart URL: https://triedagents.com/cases/btc-4h-breakout-check/ Verdict: Partly reproduced (V1) on 2026-10-19: a one-shot 4-hour breakout check worked on Kraken data (close 82,423.4 vs prior-20 high 82,826.0, so no breakout), but 2 of 3 exchange APIs failed and the alert part was not tested. Status: V1 (Partly reproduced) · Access level: L0 (No account access) · Source setup: L2-L3 (scheduled job + notifications) Category: Money Claim (@itsdidijoy, https://x.com/itsdidijoy/status/2107146422696984762): Agent watches BTC and alerts on a 4h breakout. Agent in source: Grok Bot (per post) Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: Last closed candle (08:00-12:00 UTC+8) closed at 82,423.4; prior-20 high 82,826.0. No breakout (402.6 below). The one-shot check works; the 'watch and alert me' part was not tested. Ground-truth check: Recomputed from the raw Kraken candle JSON saved in the run log. Single data source. Exact prompt: Using public 4h candles, did the last closed BTC/USD candle close above the highest high of the prior 20 candles? Method: One-shot check with a Python script against public exchange OHLC APIs. No schedule, no alerts set up. Steps: - Binance klines: HTTP 451 (geo-blocked) - Coinbase: HTTP 400 (no 4h granularity) - Kraken public OHLC XBTUSD interval=240: OK - Computed breakout rule Sites and APIs touched: api.binance.com, api.exchange.coinbase.com, api.kraken.com Failure modes: - First two data sources failed (451 geo-block, 400 unsupported interval). - An agent that silently falls back to another exchange may compare different prices. Time: About 9 s (15:41:08-15:41:17 UTC+8) Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0 for a one-shot check. Scheduling + notifications raises it to L2-L3; keep alerts to yourself first. How to cite: triedagents, "Check if Bitcoin broke out on the 4-hour chart", partly reproduced, checked 2026-10-19, https://triedagents.com/cases/btc-4h-breakout-check/ --- ## Compare one tire's price across retailers URL: https://triedagents.com/cases/tire-price-comparison/ Verdict: Partly reproduced (V1) on 2026-10-19: all 4 retailer sites blocked an unauthenticated agent; prices of about $205-206 came only from search snippets, and at least one was stale. Status: V1 (Partly reproduced) · Access level: L0 (No account access) · Source setup: L0-L1 Category: Shopping Claim (@jwkicklighter, https://x.com/jwkicklighter/status/2105037699052626264): Agent compares tire prices across stores. Agent in source: Not stated in our record Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: Snippets showed Tire Rack $205.99, Discount Tire $206.00, Sam's $205.99, BJ's $204.99, but the BJ's figure was a promo valid 3/30-4/30/2026, so at least one number was stale. No live page confirmed any price. Ground-truth check: Could not confirm against live retailer pages (all blocked). Prices are search-snippet only and marked unconfirmed. Exact prompt: Find the current price of a Michelin Defender2 225/65R17 102H at Tire Rack, Discount Tire, Walmart, Costco, Sam's Club and BJ's. Method: Tried to fetch retailer product pages directly; fell back to search-result snippets. Steps: - Tire Rack: 'This page is currently unavailable' - Discount Tire: blocked - Walmart: robot check - Costco: CAPTCHA - Used search snippets instead Sites and APIs touched: tirerack.com, discounttire.com, walmart.com, costco.com, search index Failure modes: - 4 of 4 retailer sites blocked or errored for an unauthenticated bot. - Search snippets can be months old (stale promo). Time: A few minutes of tool calls (not precisely timed) Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0, but expect blocks. Ask the agent to show the date of every price and the page it came from. How to cite: triedagents, "Compare one tire's price across retailers", partly reproduced, checked 2026-10-19, https://triedagents.com/cases/tire-price-comparison/ --- ## Is a stock below my alert price right now? URL: https://triedagents.com/cases/share-price-threshold-thangamayil/ Verdict: Partly reproduced (V1) on 2026-10-19: THANGAMAYL traded at Rs 5,074.00 (two public sources agree), above the Rs 4,770 alert level; the recurring alert itself was not tested. Status: V1 (Partly reproduced) · Access level: L0 (No account access) · Source setup: L2 (scheduled alert to the user) Category: Money Claim (@Thaju_Equity, https://x.com/Thaju_Equity/status/2103538587191149038): Agent set an alert to ping when THANGAMAYL (Thangamayil Jewellery, NSE) trades below Rs 4,770. Agent in source: Grok Bot (per post) Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: Rs 5,074.00 at 15:52 UTC+8 on 2026-10-19 (last 5 daily closes 4,900.5 / 4,849.5 / 4,791.0 / 5,098.0 / 5,074.0). Not below Rs 4,770, so no alert would fire. One-shot check reproduced; the recurring alert was not tested. Ground-truth check: Two independent public sources agreed on Rs 5,074.00 (Yahoo Finance and Google Finance). The official exchange API refused the request (403). Exact prompt: What is THANGAMAYL trading at on NSE right now, and is it below Rs 4,770? Method: One-shot price check from a public quote endpoint, cross-checked with a second public quote page. No recurring alert was created. Steps: - Yahoo Finance chart API THANGAMAYL.NS: 200 - NSE quote API: 403 - Google Finance quote page: price visible Sites and APIs touched: query1.finance.yahoo.com, nseindia.com, google.com/finance Failure modes: - The official exchange (NSE) blocks unauthenticated scripts. - Quotes may be delayed; neither source was checked for delay. Time: About 3 s Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0 for a check. An alert that messages you is L2; start with alerts only to yourself. How to cite: triedagents, "Is a stock below my alert price right now?", partly reproduced, checked 2026-10-19, https://triedagents.com/cases/share-price-threshold-thangamayil/ --- ## Pick a trip window that avoids the rainy season URL: https://triedagents.com/cases/trip-weather-avoid-monsoon/ Verdict: Partly reproduced (V1) on 2026-10-19: 5 seasons of Open-Meteo data show Phu Quoc and Ho Chi Minh City driest for Dec 15-Jan 5 (about 25 mm each) while Ubud stays wet (about 227 mm); flights and budget were not tested. Status: V1 (Partly reproduced) · Access level: L0 (No account access) · Source setup: L0-L1 Category: Travel Claim (@himanshu_v92, https://x.com/himanshu_v92/status/2107980691912986841): Compared Bali, Malaysia, Thailand, Vietnam, Singapore, Japan and Korea for late November vs mid-December to early January from Perth, including avoiding monsoon weather. Agent in source: OpenAI Dots (per post); we did not run Dots Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: Clear signal: Phu Quoc (25 mm, 5.4 wet days) and Ho Chi Minh City (25 mm, 4.8) are driest in mid-Dec-early Jan; Phuket (62 mm) and Krabi (76 mm) dry out after late Nov; Ubud (~227 mm, ~19-20 wet days), Singapore and Kuala Lumpur are wet in both windows. Tokyo and Seoul are dry but cold (10 C and 1 C average highs). Weather part reproduced; budget/flight part not. Ground-truth check: Numbers recomputed from raw Open-Meteo archive JSON (reanalysis data, coarse grid: treat as indicative, not station records). Exact prompt: For a family from Perth: compare rainfall for Nov 20-Dec 10 vs Dec 15-Jan 5 in Ubud, Langkawi, Penang, Kuala Lumpur, Phuket, Krabi, Phu Quoc, Ho Chi Minh City, Singapore, Tokyo and Seoul. Which avoid the wet season? Method: Pulled 5 seasons (2020-2024) of daily rainfall and max temperature from the Open-Meteo historical archive for 11 places and both windows. Flight prices and budgets were not tested. Steps: - 11 Open-Meteo archive calls - Averaged rainfall, wet days (>=1 mm) and max temp per window Sites and APIs touched: archive-api.open-meteo.com Failure modes: - Reanalysis rainfall can overcount drizzle days. - The source mixed weather with flight costs and budget, which we did not test. Time: About 15 s Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0. Ask for rainfall per window with the data source named. How to cite: triedagents, "Pick a trip window that avoids the rainy season", partly reproduced, checked 2026-10-19, https://triedagents.com/cases/trip-weather-avoid-monsoon/ --- ## Price a Japan round trip (and why agents disagree) URL: https://triedagents.com/cases/japan-flight-price-check/ Verdict: Partly reproduced (V1) on 2026-10-19: Google Flights gave a lowest fare of $1,214 round trip for a stand-in SFO to Haneda itinerary; the two-agent price comparison in the claim could not be tested with Grok Bot alone. Status: V1 (Partly reproduced) · Access level: L0 (No account access) · Source setup: L0-L2 (tracking) Category: Travel Claim (@petergyang, https://x.com/petergyang/status/2103693608729932025): Two agents tracked the same Japan itinerary; Muse suggested a price $1K+ higher and said it searched Duffel instead of Google Flights. Agent in source: Grok Bot and Muse (per post) Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: Got a usable answer: lowest $1,214 round trip; repeated $1,222 Delta 1-stop options via SEA/LAX. Could not reproduce the comparison part because we only ran Grok Bot, and the source's route is unknown. Ground-truth check: Prices read from the Google Flights page itself (same page the agent used). Not checked against booking a fare. Exact prompt: What is the cheapest round trip SFO to HND, Dec 10-20, 2026, 1 adult, economy, on Google Flights? Method: One Google Flights fetch for a stand-in itinerary (the post does not give the route). Did not run Muse; no price tracking over time. Steps: - Fetched Google Flights results page - Extracted all prices and itinerary rows Sites and APIs touched: google.com/travel/flights Failure modes: - The source compares two agents; a single-agent re-run cannot test that. - Which data source an agent uses (Google Flights vs Duffel) changes the price; ask it. Time: Under 1 s per fetch Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0. Ask for the source of the fare and a link to the exact search. How to cite: triedagents, "Price a Japan round trip (and why agents disagree)", partly reproduced, checked 2026-10-19, https://triedagents.com/cases/japan-flight-price-check/ --- ## Restaurants within walking distance of a Miami Beach hotel URL: https://triedagents.com/cases/restaurants-near-south-beach-hotel/ Verdict: Partly reproduced (V1) on 2026-10-19: OpenStreetMap listed 33 restaurants within 500 m of The Betsy South Beach, with distances and cuisines but no opening hours, ratings or prices. Status: V1 (Partly reproduced) · Access level: L0 (No account access) · Source setup: L0 Category: Travel Claim (@Mbrown00l, https://x.com/Mbrown00l/status/2110175707963826349): Asked for restaurants close to a South Beach hotel; got 'incredibly informative information'. Agent in source: Grok Bot (per post) Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: 33 restaurants within 500 m. Nearest: Grillfish (108 m, grill/fish), Havana 1957 (110 m, Cuban), Coyote Taqueria (186 m), Donatella (205 m). Distances and cuisines are solid; opening hours were missing for the top 12 and there are no ratings, prices or reservation info, so it is less 'informative' than the claim. Ground-truth check: Checked against OpenStreetMap data itself; whether each place is still open was NOT confirmed. Havana 1957 appears twice (two OSM entries). Exact prompt: List restaurants within a 5-minute walk of The Betsy South Beach, with cuisine and distance. Method: Geocoded a stand-in hotel (The Betsy, 1440 Ocean Drive) with Nominatim, then queried OpenStreetMap Overpass for restaurants within 500 m and sorted by distance. Steps: - Nominatim geocode - Overpass amenity=restaurant around:500 - Haversine distance sort Sites and APIs touched: nominatim.openstreetmap.org, overpass-api.de Failure modes: - No hours or ratings in open map data. - Duplicate entries; possibly closed venues. Time: About 38 s (mostly Overpass queue) Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0. Ask for distance plus a link per venue, and check hours yourself before going. How to cite: triedagents, "Restaurants within walking distance of a Miami Beach hotel", partly reproduced, checked 2026-10-19, https://triedagents.com/cases/restaurants-near-south-beach-hotel/ --- ## Build a 'who to follow in AI' list from X URL: https://triedagents.com/cases/who-to-follow-in-ai/ Verdict: Failed (F) on 2026-10-19: a 25-post X search produced a 'who to follow in AI' list with accounts of 5-21 followers in its top 20. Status: F (Failed) · Access level: L0 (No account access) · Source setup: L0 Category: Research Claim (@FundamentEdge, https://x.com/FundamentEdge/status/2109412745045242352): A curated list of the best AI accounts to follow (post had 34,147 impressions and 718 bookmarks when we read it). Agent in source: Not stated in our record Run by: Grok Bot (this assistant) on its own Linux computer with web search/fetch/shell/browser tools Run date: 2026-10-19 (UTC+8) Result vs claim: Low-quality list: accounts with 5-21 followers made the top 20. Failed to produce anything a reader should trust. Ground-truth check: Checked each listed account's follower count from the same API response; obvious mismatch with 'best accounts'. Exact prompt: From recent popular X posts about AI agents, list 20 accounts worth following and why. Method: One public X search (25 posts), ranked authors by bookmarks x 3 + likes. Steps: - X recent search, 25 results (API cap per call) - Scored posts - Ranked authors Sites and APIs touched: X API (read-only) Failure modes: - A 25-post sample is far too small. - Engagement on one post is not account quality. Time: About 8 s (15:41:48-15:41:56 UTC+8) Cost: Not metered (subscription agent); no paid APIs, no purchases Re-verify by: 2026-11-18 Lowest-access way to try it: L0. Needs many searches and a human-curated seed list; do not trust one-shot rankings. How to cite: triedagents, "Build a 'who to follow in AI' list from X", failed, checked 2026-10-19, https://triedagents.com/cases/who-to-follow-in-ai/ --- ## 'Find my missed money' from email and card perks URL: https://triedagents.com/cases/find-my-missed-money-card-perks/ Verdict: Sourced only (V0), listed 2026-10-19: “'Find my missed money' from email and card perks” is not yet verified by triedagents because it needs L1 access (reads private data). Status: V0 (Sourced only, not yet verified) · Access level: L1 (Reads your private data) · Source setup: L1 (email + card account read) Category: Money Claim (@Texas_Twins_Dad, https://x.com/Texas_Twins_Dad/status/2111875644904943782): The agent checked email and Chase Sapphire Preferred perks and found a $100 hotel credit and a free Apple TV year to activate by Dec 31. Agent in source: Grok Bot (per post) Run by: Not run yet Run date: not run What we would need to verify it: A test email inbox and a test card account (or a volunteer who shares redacted results); the card issuer's benefits page to confirm each perk. Lowest-access way to try it: Start read-only: forward one statement or benefits page instead of connecting your inbox. How to cite: triedagents, "'Find my missed money' from email and card perks", sourced only, not yet verified, checked 2026-10-19, https://triedagents.com/cases/find-my-missed-money-card-perks/ --- ## 'Organize me a trip based off today's e-mail' URL: https://triedagents.com/cases/organize-trip-from-todays-email/ Verdict: Sourced only (V0), listed 2026-10-19: “'Organize me a trip based off today's e-mail'” is not yet verified by triedagents because it needs L1 access (reads private data). Status: V0 (Sourced only, not yet verified) · Access level: L1 (Reads your private data) · Source setup: L1-L3 Category: Travel Claim (@giffmana, https://x.com/giffmana/status/2099911902360907977): Compared three agents; two found trains and hotels but failed to check hotel room availability; Gemini got prices via Google Travel but added unrelated flight suggestions. Agent in source: ChatGPT Work, Claude CoWork, Gemini (per post) Run by: Not run yet Run date: not run What we would need to verify it: A test inbox with a planted trip email and accounts on each agent for a side-by-side run. Lowest-access way to try it: Paste the email text instead of connecting the inbox. How to cite: triedagents, "'Organize me a trip based off today's e-mail'", sourced only, not yet verified, checked 2026-10-19, https://triedagents.com/cases/organize-trip-from-todays-email/ --- ## Agent plans a work trip and buys the flight URL: https://triedagents.com/cases/trip-planned-and-tickets-bought/ Verdict: Sourced only (V0), listed 2026-10-19: “Agent plans a work trip and buys the flight” is not yet verified by triedagents because it needs L4 access (spends money). Status: V0 (Sourced only, not yet verified) · Access level: L4 (Spends, deletes or escalates) · Source setup: L4 (spends money) Category: Travel Claim (@vinceisbullish, https://x.com/vinceisbullish/status/2105352089689858221): Researched flights to Munich, booked the chosen one (the user had to take over to fill in data), and picked a hotel on the same train line as the destination. Agent in source: Grok Bot (per post) Run by: Not run yet Run date: not run What we would need to verify it: A refundable fare, a prepaid test card with a hard limit, and founder approval before any purchase. Lowest-access way to try it: Let the agent research and shortlist; do the purchase yourself. How to cite: triedagents, "Agent plans a work trip and buys the flight", sourced only, not yet verified, checked 2026-10-19, https://triedagents.com/cases/trip-planned-and-tickets-bought/ --- ## Amazon replenishment: find high-refund, low-profit ASINs URL: https://triedagents.com/cases/amazon-replenishment-sellerboard/ Verdict: Sourced only (V0), listed 2026-10-19: “Amazon replenishment: find high-refund, low-profit ASINs” is not yet verified by triedagents because it needs L1 access (reads private data). Status: V0 (Sourced only, not yet verified) · Access level: L1 (Reads your private data) · Source setup: L1 (seller account data; third-party API) Category: Research Claim (@TonyGainzzz, https://x.com/TonyGainzzz/status/2110288321788129343): Connected to Sellerboard and Amazon's API, analysed 6 months of data and returned about 35 ASINs with high sales but high refunds and little profit. Agent in source: Grok Bot (per post) Run by: Not run yet Run date: not run What we would need to verify it: A seller account sandbox or anonymised export with known answers to score against. Lowest-access way to try it: Export a report and give the agent the file, not your API keys. How to cite: triedagents, "Amazon replenishment: find high-refund, low-profit ASINs", sourced only, not yet verified, checked 2026-10-19, https://triedagents.com/cases/amazon-replenishment-sellerboard/ --- ## Analyse last month's bills and recurring subscriptions URL: https://triedagents.com/cases/bills-and-subscriptions-review/ Verdict: Sourced only (V0), listed 2026-10-19: “Analyse last month's bills and recurring subscriptions” is not yet verified by triedagents because it needs L1 access (reads private data). Status: V0 (Sourced only, not yet verified) · Access level: L1 (Reads your private data) · Source setup: L1 Category: Money Claim (@davorb89, https://x.com/davorb89/status/2105301759086514367): Set the agent to analyse the previous month's bills and recurring subscriptions. Agent in source: Grok Bot and OpenClaw (per post) Run by: Not run yet Run date: not run What we would need to verify it: A synthetic bank/email dataset with known subscriptions, so we can score what the agent finds and misses. Lowest-access way to try it: Paste a CSV export with account numbers removed, rather than connecting a bank. How to cite: triedagents, "Analyse last month's bills and recurring subscriptions", sourced only, not yet verified, checked 2026-10-19, https://triedagents.com/cases/bills-and-subscriptions-review/ --- ## Cancel a pickup order by phone on your behalf URL: https://triedagents.com/cases/cancel-wrong-store-order-by-phone/ Verdict: Sourced only (V0), listed 2026-10-19: “Cancel a pickup order by phone on your behalf” is not yet verified by triedagents because it needs L3 access (acts as you). Status: V0 (Sourced only, not yet verified) · Access level: L3 (Acts as you, visible to others) · Source setup: L3 (speaks as you to a business; reads email) Category: Life admin Claim (@xtophermiller, https://x.com/xtophermiller/status/2100762896867217804): The agent found the order number in email, phoned the store, navigated the phone menu and got the wrong-location order refunded. Agent in source: Grok Bot (per post) Run by: Not run yet Run date: not run What we would need to verify it: A test order we own, a phone-capable agent account, and founder approval because it speaks to a real business as us. Lowest-access way to try it: Have the agent draft the call script and find the number; place the call yourself. How to cite: triedagents, "Cancel a pickup order by phone on your behalf", sourced only, not yet verified, checked 2026-10-19, https://triedagents.com/cases/cancel-wrong-store-order-by-phone/