What the Street Is Saying: Sentiment Comes to PAND·AI

I’m glad to introduce the latest addition to PAND·AI: our Investor Sentiment Analysis section. This new feature summarizes what people have been saying about a given stock over the last 72 hours on FinTwit (X), as well as what has been written about that stock during the same period across most major finance-related websites.

This section is particularly important to me and is one I consult on a daily basis (yes we had an internal version for a while!). At the end of the day, no matter how sophisticated we believe stock investing to be, the reality is often much simpler: it is, to a large extent, a popularity contest. While strong fundamentals usually support long-term appreciation, the actual mechanism that drives prices higher (or lower) is an imbalance between buyers and sellers. This also explains why companies with weak fundamentals but a compelling and popular narrative can sometimes see their stock prices surge. Given this reality, staying informed about “street sentiment” around a stock can be just as important as reading fundamental research reports.

This feature was part of the original PAND·AI roadmap from the very beginning and has actually been live in our internal version for quite some time. However, we had not released it in beta due to both technical challenges and cost considerations. Why? Because at its technical core, this component of our AI analyst is very different from the rest of PAND·AI. It relies on agentic functions rather than standard LLM calls. While most of PAND·AI is built around interconnected LLMs fed with institutional-grade data, probing sentiment across the web and social media is a much more complex task that requires advanced agentic behavior. And agentic functions mean longer runtimes and significantly higher costs per request, with each report involving multiple requests.

Concretely, the system searches across major financial news sources such as Seeking Alpha, Bloomberg, and others, while also analyzing posts on X (Twitter) from the FinTwit community through a direct connection to their analytics API. The collected information is then processed by an LLM that produces a structured output, including an explicit confidence level.

Why include a confidence level? Unlike other parts of PAND·AI that rely on well-defined and consistent datasets, this agentic function performs open-ended searches where relevance and coverage can vary significantly. For example, we can expect to find an enormous amount of social and editorial content for a recently popular stock like Affirm (AFRM),

while far less information may be available for more obscure tickers or less trendy companies such as LYFT.

The volume, diversity, and consistency of the information collected directly influence the confidence level. This is especially true when sentiment signals are mixed or weakly aligned, as limited or fragmented data makes it harder to draw a robust conclusion.

How it work

This new feature is not part of the main core report like the other sections. Instead, you will find it inside the main center box.

This different placement is intentional and highlights two key differences.

1. First, it does not load automatically. You need to actively click on it to trigger the analysis. (more on that later)

   
   

2. Second, it is based on constantly changing content, which can lead to very different results from one day to the next. (It is very responsive to news)

Whenever someone requests a sentiment analysis for a given stock, we store the result for 12 hours.

This time window is the trade-off we found that allows users to avoid repeatedly waiting for the analysis, while keeping costs under control and still capturing meaningful changes in sentiment across social networks and the web.

Ideally, the analysis could be refreshed automatically every night while most of us sleep. That would clearly be the most efficient approach. However, as mentioned earlier, there were significant cost constraints associated with bringing this feature live. Internally, this point generated a lot of discussion, as fully automating it would expose us to a real cost sink without a clear way to control usage.

For now, we prefer to deploy the feature this way, observe how it is used, and adjust down the road if needed.

PAND·AI launch

This cost issue is very real, largely because we were genuinely surprised by how intensively you have been using our little robot analyst. We’re certainly not complaining—quite the opposite. We’re very happy to see such strong engagement, and we expect PAND·AI to become even more useful as we move into earnings season.

One clear sign that we underestimated how avid a group of readers you are is the sheer number of stocks you’ve been loading. We naively thought that preloading the 100 most popular stocks would cover most use cases. In my head, I assumed we might end up with perhaps 50 additional tickers over time. Instead, in just the first week, you collectively loaded more than 400 different tickers.

That level of usage clearly exceeded our initial assumptions and is part of why we had to be thoughtful about how and when to roll out more computationally intensive features. But I’m sure you’re just as curious as I am to see which tickers have been the most popular—so here it is. After all, it’s your own data, fellow members.

I didn’t know much about ALAB. Seeing it sitting by far at the top of the list, I decided to dig a bit deeper using PAND·AI, including testing its new sentiment analysis section. What I read across all sections—combined with a very bullish sentiment across the web—convinced Zack and me to trust our members’ conviction, and we both personally opened a position last week. As of now, that bet has already turned into a nice Christmas gift.

This experience convinced me that we should share this ranking more regularly in the future, and perhaps even automate it further down the road.

We’ve also rolled out several updates since launch to fix a few technical bugs and occasional AI brain farts. As a result, the platform now runs much more smoothly.

PAND·AI as a stand-alone product

We’ve also received several requests to offer PAND·AI as a stand-alone product. It remains very clear to me that WU Advanced is the “all you can eat” offering within WealthUmbrella and that it will always include PAND·AI, just like all the other tools we build.

That said, after giving it more thought, I also recognize that PAND·AI is quite different from the rest of what we do. It is a product with its own personality and use case. I can easily see someone wanting access to this AI analyst without necessarily wanting the rest, and that alone justifies giving it its own subscription.

As a result, we plan to put this new stand-alone PAND·AI subscription online tomorrow.

Conclusion

Again, I hope you will enjoy this new addition, which completes our initial vision for what PAND·AI was meant to be. It is extremely responsive to breaking news and particularly effective at summarizing investor sentiment from around the web. That said, I apologize in advance if you occasionally find it a bit blunt or impolite. This section is powered by xAI—arguably one of the most radical LLMs out there—and we chose not to add an additional supervision layer to the output, given how long the analysis already takes to run.

Considering how quickly AI tools are evolving—and agentic features in particular—I am confident that PAND·AI will continue to evolve significantly over the coming year. That said, as it stands today, I am genuinely satisfied with what it delivers.

That said, this section comes with an important disclaimer. This part of PAND·AI surfaces the most popular tweets and articles that reflect the prevailing consensus on social media and across the web for a given stock. When sentiment turns bullish, the content highlighted here can become highly euphoric and often lacks nuance. This is not a reflection of what a thorough analyst would conclude, but rather a snapshot of “street sentiment” — and the street does get it wrong from time to time.

I’ll be the first to admit that reading a compilation of very bullish, unfiltered takes can sometimes trigger a sense of FOMO. I’ve caught myself becoming more bullish than I should simply by absorbing that collective enthusiasm. So, at the risk of sounding a bit paternalistic: try not to get carried away by it. Use this section for what it is meant to be — a valuable complement to your analysis, not a substitute. It is most effective when combined with broader market context, proper risk assessment, and a solid understanding of a company’s fundamentals.

Now, with the stock market now back on track, I assume I shouldn’t write anything else before Christmas. So, Merry Christmas to you all. I hope most of you will have some time off to enjoy moments with family and friends. Life moves fast, and these slower periods are true treasures.

I should be back soon with my usual New Year post. I recently went back and reread what I wrote last year about potential winning trends and companies that could come out ahead. That exercise put a fair amount of pressure on me—it will be hard to beat last year’s picks. Not bragging here; some years we simply get luckier than others. As I’ve said before, in all humility, I bought TDOC at the top in 2021!