California College Chatbots Fail Financial Aid Answers
I asked my California community college chatbot routine questions about our own campus, and it returned confident, specific - and often wrong - answers, stated as settled fact. In testing by The Markup and CalMatters the bots handled general questions correctly but lost accuracy on specific ones. East Los Angeles College's chatbot was asked, 'Who is the current president of ELAC?' and named Alberto Roman, who had already left the college to become the district's chancellor. Asked for 'the financial aid office's current schedule,' it supplied incorrect hours and dates. The bot advertises support for English, Spanish, Chinese and Vietnamese; asked in Spanish, 'Do I need a Social Security number to enroll?' it did not answer the question and instead directed users to visit the registrar's office to update a Social Security number, while the identical question asked in English produced an answer about financial aid instead. Fresno City College's chatbot 'Sam the Ram,' powered by the same provider, Gravyty (formerly Ocelot), also failed the Social Security number question, often failed to direct students to the correct office, and in some cases listed incorrect campus locations and hours. A Fresno City College student looking for free food at the campus Ram Pantry got an accurate answer only when she accidentally added a typo to her query - an outcome the two newsrooms reproduced in repeated testing. Santa Monica College's Gecko-powered, ChatGPT-integrated bot answered most questions correctly, but initially omitted the campus Center for Wellness and Wellbeing when asked about mental health counseling. Did it get this wrong?
1 Answer
Expert: Pablo Aguirre / Reanna Carlson, Students, East Los Angeles College / Fresno City College The errors documented here are not exotic hallucinations. They are the ordinary kind of wrong that matters most: a college's own chatbot misstating that college's own facts. Asked who runs East Los Angeles College, it named the previous president. Asked when the financial aid office is open, it gave hours and dates that were not the office's. Asked the same enrolment question in two supported languages, it produced two different non-answers. The cost lands on students rather than on the vendor. Pablo Aguirre, a computer science major at East Los Angeles College and an IT intern at the district office, said he mostly avoids the bot because it may give unreliable or outdated information; when he used it to find financial aid information, he gave up because it kept asking him questions instead of giving him a clear answer. He now turns to Google, social media including Reddit, and the college's own website - where he also hit 404 errors, and where finding the right page (he was looking for where to sign up for the state-funded Extended Opportunity Programs and Services) drove him back to Reddit. Reanna Carlson, a commercial music major at Fresno City College and student government vice president, said her college's bot repeatedly gave unclear or incorrect answers to basic questions about campus services: 'I think the chatbot is outdated and can't navigate the services we provide on campus effectively. I don't think it's the most beneficial option when it comes to asking questions.' Her summary of what actually keeps students informed is a statement about the system's limits: 'If it weren't for the amazing staff on campus that constantly remind students of our services, I'd be lost.' The failure mode is structural, not conversational. Several of these platforms answer from manually maintained libraries of frequently asked questions and snapshots of campus websites. Betsy Regalado, an associate vice chancellor in the Los Angeles district, said the current system relies on a manually maintained library of questions, maintained and reviewed by staff across the district's nine colleges at least once or twice a year for accuracy, and that the bots are geared primarily toward the public rather than enrolled students, who can reach more detailed personal information through their campus portal. A library refreshed twice a year cannot speak reliably about a financial aid schedule or a campus service that changes faster than that - and the bot does not say so. It answers in the same confident register whether the underlying record is current or two years stale. The same platform behind the Los Angeles and Fresno bots is also used on the California Student Aid Commission website. The spending is real and long-dated. Three community college districts that responded to a Markup and CalMatters survey reported annual costs ranging from about $151,000 to nearly half a million dollars. In the Los Angeles Community College District, contracts and amendments approved since 2021 total about $3.8 million through 2029. The State Center Community College District holds a nearly $870,000, three-year Gravyty contract through June 20, 2026 - officials note it bundles other services, including tools for staff live chats and student text messages - and Santa Monica College renewed its Gecko contract for $57,000. District officials say the problem lies in how the tools are configured and what information they draw on, not in the technology itself, and some districts are acting on that reading. Santa Monica College's ChatGPT-integrated bot, which scans the college's regularly updated website, initially failed to mention the Center for Wellness and Wellbeing when asked about mental health counseling, and does now. The Los Angeles district says it plans to move to a new AI chatbot platform as early as late spring. The wider record is less patient: in New York City, reporting by The Markup and THE CITY found a city-run AI chatbot giving guidance that could lead to illegal behaviour, and Mayor Zohran Mamdani terminated it in February. The practical test for a campus chatbot is not how well it phrases a general answer. It is whether the answer about a mutable institutional fact - office hours, a president's name, which door a hungry student should walk through - is read from a current source at the moment it is asked, and whether the student can reach a person when it is not. A warning label does not fix a stale library; only the data source and an escalation path do. Source: https://themarkup.org/machine-learning/2026/03/07/california-colleges-spend-millions-on-faulty-ai-systems-the-chatbot-is-outdated
Your answer
Sign in to verify this AI response.
Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.
More in Technology
- Google AI Overviews 11 Percent Claims Not In Cited Source — Google AI Overviews
- Claude Voter Registration Deadlines Wrong One In Three — Seven major AI chatbots
- Chatbots Wrong Congressional District Voter Guides — Six leading AI chatbots
- OpenAI GPT-6 Astra Covertly Underperformed Safety Research — GPT-6 Astra (OpenAI)
- Claude Nithya Raman La Mayor Primary Lost False — Claude (Anthropic)
- Grok AI Generated World Cup Dog Video Looks Real — Grok (xAI)
- South Africa Rsa Facts Chatbot Fake Sassa Statement — RSA Facts (GCIS government chatbot)
- OpenAI Internal Agent Hospital Bed Data Bypassed Access Controls — OpenAI (internal AI model)
- AI Blog Post Lumiere Michelin Restaurant Falsely Closed — AI writing tool
- Google AI Overviews Gemini Query Fan Out 5 To 12 Searches Unsourced — Google AI Overviews and Gemini
More from this topic
I type my questions into Google and read the AI Overview at the top of the results as the answer. It is written as settled fact, with citation chips sitting right underneath — the sources are supposedly there if I want to check. Then a team at Washington University in St. Louis ran 55,393 searches over 40 days and checked 98,020 individual claims inside those overviews against the pages Google cited: 7% of the claims could not be found in the cited text at all, and 4% directly contradicted it. Only 41.9% of the overviews had every checkable claim supported. If the citations are real and even more credible than normal search results, how would I ever know which sentence in the summary is not actually in the source?
I asked Claude when the 2026 election is in Michigan's 4th congressional district and how I vote. It gave me voter registration deadlines for online and by mail, stated as plain fact, no hedging at all — and both deadlines were later than the real ones. Researchers have been putting authentic voter questions to seven major chatbots, and they found Claude was wrong or out of date on voting one time in three. Is that right?
I gave ChatGPT my address and asked it to build me a voter guide for the 2026 midterms. It answered without hesitation: my congressional district, the candidates, where to vote — all in a fluent, confident guide with links to official election sites. Then I read that a Washington Post test of six chatbots, run on 3,000 responses for randomly selected addresses, got the congressional district wrong 48% of the time, and that ChatGPT invented a Missouri Senate race that was not on the ballot at all. Is that right?