Physix FrontierCan AI analyze financial reports? I tested it on three stocks
Using AI tools for stock analysis to help you make money is still a long way off. Over the past two days, I ran earnings news for AMAT, RDDT, and GEMI through three models, and I did gain some clarity on the industrial chain in the AI era.
The story started with Bloomberg's post-market quote on Thursday: Applied Materials (AMAT, a semiconductor equipment leader) beat earnings expectations and raised guidance, yet its stock fell 1.7% after hours. Reddit (RDDT, the publicly listed forum company) rose 6%. Gemini Space Station (GEMI, a space company unrelated to Google's Gemini) reported its fourth consecutive quarterly loss since IPO, and its stock dropped further again. Putting these three together was interesting, so I casually threw the news summaries to Claude, ChatGPT, and Gemini Spark, asking each to write a 200-word investment judgment.
| Tool | Who Beat Expectations | Did It Name Risks | Conclusion Style |
|---|---|---|---|
| Claude | AMAT | Vaguely mentioned | Balanced allocation, avoided direct judgment |
| ChatGPT | None | Not mentioned | Macro narrative, all correct nonsense |
| Gemini Spark | Only said AMAT | Not mentioned | Translationese, like an earnings press release |
First, let's talk about Claude. I gave it the raw text of the three news items. It took it seriously, listing a framework in three points and giving generic phrases like "short-term pressure, long-term optimism." But one detail made me frown: it conflated Gemini Space Station with Google's Gemini model, spouting stuff about "Gemini ecosystem synergy." I stared at the screen for a few seconds. I specifically didn't tell it these were two different companies, and it genuinely failed to distinguish them. It's fine for information organization, but for industry judgment, it talks nonsense with a straight face.
ChatGPT performed more steadily, but also more "correctly." It wrote a paragraph for each of the three stocks, staying on topic, ending each with phrases like "need to monitor subsequent fundamental changes." Which means it said nothing. I pointed out the contradiction of AMAT beating expectations while the stock fell, and it replied with "the market may have already priced this in," which actually hit the nail on the head. But when I pressed further, "To what extent is the expectation priced in?", it got vague and started listing risk warnings. On my end, it seems suitable for information compression, not logical deduction.
Gemini Spark surprised me the most. Using the version configured in my company account, the analysis looked like it was cut directly from an earnings press release. Not only did it fail to distinguish that GEMI is a space company, but it mixed it up with Google Gemini, concluding with "The Gemini ecosystem faces competitive pressure." I saw a post on Reddit saying Gemini 3.5 was a "disaster" for stock analysis, where a guy lost 6.6% doing paper trading with it, with almost all positions stopped out within days. Now I believe it. For fundamental analysis, it's even less reliable than I expected.
However, after running this full circle, AMAT made me want to say a bit more. A good point was made in a Reddit post: The more AI models, the more chips needed; chips need equipment to be made, and AMAT sells the equipment. This logic holds up. AMAT even hit a 52-week high this week, touching $739 per share intraday. But right after hitting a new high on Tuesday, it fell 1.7% on Thursday despite beating earnings—this is a classic case of "good news exhausted." Even more split opinions: In the same forum, someone posted showing they bought high at $739 and got trapped, while another guy said he cut losses at a lower position, losing 28%. Whether the market is optimistic or pessimistic, nobody can say for sure.
So my judgment is this: AI tools are fine as information assistants. Having them read earnings reports, organize announcements, and list key data saves me a lot of effort compared to digging through them myself. But letting them give buy/sell advice is unreliable at this stage. After running three models, none could clearly identify the underlying chip structure issue behind the contradiction of "beating expectations but stock falling." They can tell you what happened, but can't explain why, nor predict what happens next.
By the way, I agree with the "selling shovels" logic for AMAT, but chasing it at this price level feels expensive to me. Have you guys used AI tools to analyze earnings? Would you dare to follow their signals directly?
📌 This article is compiled from Bloomberg Tech. Original: https://www.bloomberg.com/news/videos/2026-08-14/reddit-up-applied-materials-gemini-fall-stock-movers-video
Copyright belongs to the original author. This is a compilation and independent analysis based on public reports.
