How Artificial Intelligence Is Changing The BDC Reporter And What It Means for Our Readers

21 May 2026

All Of A Sudden

Very recently, Artificial Intelligence (AI) has drastically changed how the BDC Reporter conducts its research. No longer do we need to painstakingly and time-consumingly hop from BDC press release to investment presentation to quarterly filing to conference call transcript – reading each in turn, taking notes along the way, to get a full idea of the most recent performance. Now, using the power of Google’s NotebookLM, we load each of those documents, and many others besides, into an eponymous “notebook” that we can interrogate about anything and everything pertaining to a BDC’s performance. The AI can provide a factual answer about anything imaginable, generate dense data tables when asked, and – if properly coached – tease out almost all the issues that we – and our readers – might be interested in.

Ask Anything

How many portfolio companies does BDC XYZ have at the end of the most recent quarter? No need to find the answer in the Investment Presentation or press release. We just ask the AI. What if we want a list of all the companies on non-accrual, the amounts at cost and fair market value, and the totals for those 3 categories? In the past, that data was often hard to find and difficult to extract, as many BDCs don’t like to provide that information in one place, leaving the poor BDC Reporter to read through pages and pages in the filings, trying to identify by the footnotes which companies had loans on non-accrual and copying out the results to patch together a list. This could take an hour or two for a single BDC. (We track 45 BDCs, so you do the math). Now we ask the AI and get a finished table in as little as 30 seconds. Literally. What’s more, the output is highly reliable because all of the data is derived from official sources. There is virtually no risk of the AI “hallucinations” that sometimes come from “chatting” with a ChatGPT or equivalent.

What, us worry?

This might suggest that AI is making what the BDC Reporter does redundant, and that we will soon join untold others on the proverbial pavement as the AI takes over. However, we believe that is not going to happen, based on months of experience with NotebookLM and many other AI tools. There are two main reasons why.

Management

First, there remains a strong need for a human “agent” to manage the process and gather the right information at the right time. NotebookLM can parse only so many different sources before it gets confused and starts making mistakes. Most “notebooks” are limited to a few dozen or, in our case, using the maximum capacity allowed, a few hundred items.

Leadership

Furthermore, the AI continually needs guidance on how to organize, synthesize, and present the nuggets of information within that mountain of source material. That’s why you’ll find us every day downloading the appropriate source content for each BDC into well-organized folders, placing it in their respective “notebook,” and deleting any duplicates or stale data. Moreover, we “teach” the AI what sort of information to look for and what to juxtapose it with. These involve long, proprietary “prompts” written into the DNA of the “notebooks” to ensure appropriate output.

Like With Fine Wine Or Art

Going back to an earlier example, you may want to know how many companies in a BDC are on non-accrual, but also, for a sense of proportion, how many companies there are in the portfolio, and an infinite number of other related metrics.

So we see ourselves now as BDC data “curators”, gathering the “right” information for ourselves and our readers and educating the AI on what is valuable and what is not. Wherever possible, we also serve as a fact checker and validator.

Just Do It

The second reason AI will not be “displacing” the BDC Reporter any time soon is its inability to make good judgments. Research and analysis are all very well, but at the end, some sort of conclusion has to be drawn for anything “actionable” to follow. The AI, if asked, will draw conclusions and outline potential courses of action. As you can imagine, this is a textbook case of “caveat emptor”. NotebookLM itself spells out the risks in this way:

When an LLM shifts from synthesis to judgment, several vulnerabilities emerge:

Contextual Tunnel Vision: NotebookLM only “knows” what is in the uploaded sources. It cannot factor in external market conditions, late-breaking news, or unwritten industry norms unless they are explicitly detailed in the text. A suggested action might make perfect sense based on a single document but be disastrous in the real world.

Missing the Subtext: LLMs process explicit text well, but they struggle to read between the lines. They might miss the subtle, evasive tone of management in a financial filing or the strategic omissions in a legal document, leading to conclusions built on a naive reading of the text.

False Certainty (Automation Bias): LLMs generate responses with a highly authoritative tone, even when the underlying logic is fragile or probabilistic. This can lead to “automation bias,” where users begin to trust the AI’s conclusions over their own critical thinking, potentially skipping the vital step of verifying the underlying data points.

Amplification of Flawed Data: If the source documents contain errors, biases, or non-standard accounting (like heavily adjusted non-GAAP metrics), the LLM will treat them as ground truth. It will logically derive flawed conclusions from those flawed premises.

Better

The BDC Reporter may not be perfect, but we have a much, much wider lens than the AI and a sense of context and historical precedent. We like to think we know when we’re guessing at an answer and when we are not. There’s no “false certainty” here. The BDC Reporter’s corporate slogan is “News-Analysis-Views”. The last leg of that stool was always the most important and has become even more critical in this new AI-dominated world. We fear for those out there who will turn to their computers to draw actionable conclusions. Contrary to what some claim, this is not a “second brain”. Instead, we regard this fantastic technology as an able, hard-working assistant, but one who requires constant monitoring and direction. (You know the type).

Implications

What does this all mean for the BDC Reporter and its readers going forward?

Ch-anges

For us, AI – as suggested above – has already changed the way we work. As mentioned above, we have built deep, “curated” depositories of knowledge about every public BDC and much else besides. We are adding – and deleting – new sources daily. Although the setup time for our research has greatly increased – those sources don’t magically find their way into our 50+ “notebooks” – we can gather all the research we need to be fully informed far faster than ever before. (At the moment, our many “notebooks” are available only to ourselves, but we hope to find a way to make those data mountains available to our readers as well. This would allow 24/7 access to everything you wanted to know about your favorite BDC(s) but didn’t know who to ask.)

Jumping Forward

This accelerates the pace at which we can get around to generating BDC “analysis” of one kind or another. We consider that the hard-to-define process of synthesizing all the research we’ve done to determine what might matter most – or matter less – in the future to ourselves and our readers. For example, we recently analyzed a BDC’s latest unsecured note offering to demonstrate how profits are being squeezed across the sector and how that might translate into the stock price outlook.

Best For Last

Most of all, expect the BDC Reporter to become more forthcoming about our “Views” going forward. This will include commentary on major developments across BDC-land, on both the public and private sides. However, we can’t claim that’s anything particularly new, as we offer up those sorts of “thought pieces” on occasion. What is new is that we’re planning to offer more of our thoughts on which BDCs to invest in and which to avoid. Until recently, we had a separate publication called BDC Best Ideas, dedicated to BDC stock selection, but we closed it down due to the pressures of running three publications simultaneously. We’re going to bring back some of the content and ethos of BDC Best Ideas and incorporate it into the BDC Reporter. However, we will always make clear the distinction between our objective BDC analysis and our subjective views on what constitutes a worthy investment.

Model Portfolio

We intend to bring back our BDC Best Ideas portfolio, whose main goal, when set up at the beginning of 2025, was to build a portfolio of BDC stocks that could outdo the total return of the only BDC ETF – Van Eck’s BIZD. Nearly a year and a half later, the Best Ideas portfolio remains ahead of its ETF counterpart, but we’re the first to concede that investing conditions have been horrendous – the worst we’ve experienced since 2020 and increasingly reminiscent of the GFC.

New Look

In anticipation of these changes, we’re planning to move the BDC Reporter to a new platform – moving from WordPress to Ghost.org. The latter, on which the BDC Credit Reporter is built, as was BDC Best Ideas, has a more elegant design and is less conducive to crashes than our nearly decade-old WordPress site. For our readers, the most significant difference will be a new way to sign in to your account: a link will be sent to your email. This provides a higher level of security and eliminates the need to remember passwords! Otherwise, nothing should change, with new articles continuing to be sent directly to the email addresses of Premium subscribers who’ve opted in.