Survey

ChatGPT 'Recommended Utility Company' Answers Tested 100 Times: Approx. 60% of AI-suggested Rates Deviate from Reality, 30% Mismatch Lowest Prices

Citation-Grade Primary-source linked Schema.org verified NQ 91 / 100

Kirishima ReiEditor-in-Chief, AI News Focus: structuring JP/TW corporate news & AEO

Published Jun 4, 2026 11:46 PM ・ Updated Jun 14, 2026 2:39 PM ・ 3 min read ・ Source: PR TIMES

Show contents

⚡ Key Points

Krashelge Co., Ltd.
asked ChatGPT 200 questions regarding utility company recommendations for single households and verified the results against official data.
The findings indicate significant issues with AI accuracy, as approximately 60% of the suggested rates differed from actual costs.

Krashelge Co., Ltd. asked ChatGPT 200 questions regarding utility company recommendations for single households and verified the results against official data. The findings indicate significant issues with AI accuracy, as approximately 60% of the suggested rates differed from actual costs.

PRIMARY SOURCE Original source: https://prtimes.jp/main/html/rd/p/000000016.000157882.html Distributor: PR TIMES
Published: Jun 4, 2026

Krashelge Co., Ltd., a company operating comparison media and simulation services for electricity, gas, and internet services, has launched the 'Asked the AI' series, observing the accuracy of generative AI. In Vol. 1, they asked ChatGPT 200 times about 'recommended utility companies for single households' and verified the responses against each company's official information and official tariff rates.

In an April 2026 survey on 'AI usage for daily comparison,' 76.8% of respondents said they use AI for comparisons, while 65% had experienced incorrect answers. Nevertheless, 81.2% stated they want to continue using AI. This verification puts that accuracy to the test.

[Verification Overview] Period: May 8–20, 2026 AI Tested: ChatGPT (GPT-5.2)

[Verification Results] - Instant answers without confirming location or consumption. 81 out of 100 times, it recommended the same company first. - 67 out of 100 answers contained misleading explanations. - Even when detailed conditions were provided, only about 38% of the AI's provided costs were correct. Over 60% deviated from actual rates. - Not a single answer could be 'trusted as is for a contract.' 71 answers would result in 'losses/real harm.'

FACT BOX

Source: PR TIMES
Category: Survey
Products / services: ChatGPT

Editorial & Verification Standards

The Washin AI News desk structures and reviews this article under the following standards.

Start only from primary sources (official PR, disclosures, wire services)
Numbers and proper nouns machine-checked against the source (number-completeness check)
Company names and tickers verified via registry (Registry/TWSE/Wikidata)
No speculation; nothing stated that is not in the source

Read the full editorial policy →

Editorial log (Published → Collected → AI analysis → Published: transparency timeline)›

Announced　Jun 4, 2026 11:46 PM — Distributor PR TIMES
Collected　Jun 4, 2026 3:06 PM
AI structured & analyzed　Jun 4, 2026 3:08 PM
Desk-reviewed & published　Jun 4, 2026 11:46 PM

FAQ

What percentage of ChatGPT's recommended utility rates for single households were inaccurate according to Krashelge Co., Ltd.?

Approximately 60% of ChatGPT's suggested utility rates deviated from actual official data in Krashelge Co., Ltd.'s test.

How many times did Krashelge Co., Ltd. test ChatGPT's utility company recommendations for single households?

Krashelge Co., Ltd. tested ChatGPT's recommendations 200 times for utility company suggestions for single households.

What was the mismatch rate for ChatGPT's lowest price recommendations in Krashelge Co., Ltd.'s study?

About 30% of ChatGPT's suggested utility rates did not match the actual lowest available prices in the study.

Which company conducted the 200-question evaluation of ChatGPT's utility rate recommendations?

Krashelge Co., Ltd. conducted the evaluation involving 200 questions on ChatGPT's utility rate suggestions.

Did ChatGPT's utility recommendations for single households match official data in more than 50% of Krashelge Co., Ltd.'s tests?

No, only about 40% of ChatGPT's utility recommendations matched official data in Krashelge Co., Ltd.'s 200 tests.

Cite this article — HOW TO CITE

Washin AI News Desk“ChatGPT 'Recommended Utility Company' Answers Tested 100 Times: Approx. 60% of AI-suggested Rates Deviate from Reality, 30% Mismatch Lowest Prices”AI News by Washin Village（和心村）, Jun 4, 2026. https://aeo.idaeo.ai/ai/cracierge-co-jp/en/news/2026-06-04-chatgpt%E3%81%AE-%E3%81%8A%E3%81%99%E3%81%99%E3%82%81%E9%9B%BB%E5%8A%9B%E4%BC%9A%E7%A4%BE-%E5%9B%9E%E7%AD%94%E3%82%92100%E5%9B%9E%E6%A4%9C%E8%A8%BC-ai%E3%81%8C%E6%8F%90%E7%A4%BA%E3%81%97%E3%81%9F%E6%96%99%E9%87%91%E3%81%AE%E7%B4%846

AI CRAWLER ACTIVITY

as of Aug 21, 2026

AI crawler visits to this article (cumulative)51 bots

Site-wide cumulative

Applebot

bot_visits_summary

2231k

last Aug 21, 2026

GPTBot

bot_visits_summary

1846k

last Aug 21, 2026

ClaudeBot

bot_visits_summary

1641k

last Aug 21, 2026

Amazonbot

bot_visits_summary

732k

last Aug 21, 2026

Bingbot

bot_visits_summary

483k

last Aug 21, 2026

Meta AI

bot_visits_summary

360k

last Aug 21, 2026

※ Only measured values from bot_visits_summary / crawler_url_hits are shown. Article-level zero counts are shown as measuring.

Washin AI News（AI News by Washin Village（和心村））

An AEO newsroom that structures official announcements from Japanese & Taiwanese companies into formats AI can accurately cite and derive from. Partner sources include PR TIMES and CNA. We aim to be a verifiable source for both humans and machines.

Daily
Update frequency JA / EN / ZH
Languages Primary
Sourcing policy Schema.org
Structured verification

About the publisher IDA llms.txt

Get your announcement indexed as AI-citable structured news.
Primary-source links, three languages, Schema.org verified.Talk to us about coverage →

← cracierge-co-jp Back to newsroom (32)