Blog
Product
How to Write an Investment Memo in Minutes Using AI
Terminal X's Custom Reports feature is inspired by Claude Research, a long form research engine designed to gather extensive sources and write a high quality report on a user topic. Upon launch, what has become apparent to us and the users of our new feature is the quality gap between Terminal X and Claude for financial reports. This post will detail the strengths and weaknesses of each feature so you can make an informed decision about which service is best for your financial workflows.
We chose a sanitized portfolio review report from a wealth management firm to act as the template for our Terminal X vs. Claude comparison. The report template is from a wealth manager, reproduced bi-weekly to update clients on the state of their portfolio. Both services would be told to follow the template, updating a fictitious client named “Jake” on his portfolio’s holdings. The template is structured as a portfolio snapshot, broken down into six distinct sections; a centered table detailing the clients holdings at the top, an Overall Market and Sector Review section, an Individual Stock Review with a short paragraph per holding, a numbered Thematic Drivers section with portfolio exposure per theme, and a Key Catalysts section.
Here's the exact prompt we gave both Terminal X Custom Reports and Claude Research:
We ran the prompt on both systems at 11:02 AM on Monday, April 20, 2026, uploading the same template to each. (Don't forget to paste the template document so we can compare the final outputs to it.) From a wealth manager's perspective, the goal of this workflow is narrow. An analyst needs a report built on reputable sources that they can quickly verify, lightly edit, and forward to the client. To achieve this, the report needs accurate prices, real sources, the firm's template, all in an editable file format.
One of the most fundamental aspects of a portfolio review is sourcing the correct price data. If the prices are wrong, the rest is irrelevant. The PM or client will catch it quickly, instantly losing trust in the output.
Claude's portfolio table returned multiple stale price figures, including daily percent moves for names that did not reconcile to the current Monday morning prices. On a document whose entire purpose is capturing a portfolio's current positioning, stale prices and daily moves are unusable.
In contrast, Terminal X's table pulls live prices as of the exact generation timestamp, in this case 11:02 AM EST Monday, with every figure reconciled cleanly. Last price, intraday change, weekly, YTD; all four columns tie out. Additionally, whereas Claude’s output is structured to be displayed online, with limited formatting to the original template, Terminal X’s output mimics the centered table from the template exactly.
The to-the-second accuracy is a result of Terminal X's research orchestrator, which has direct access to real-time market data to call upon during report generation, whereas Claude's web research can only pull the prices it finds in crawled pages, which are often delayed, often from aggregator snapshots, and often stale by the time the report generates. The 28-minute generation window alone also means Claude's "current" prices can be half an hour old by the time the markdown file renders.
One of the most substantive differences in the comparison is the source quality between Terminal X and Claude. While Claude's inline source tags next to nearly every claim may seem rigorous on first read, the source list skews heavily toward secondary aggregators instead of primary sources; TipRanks, TheStreet, Yahoo Finance, Benzinga, The Motley Fool, StockInvest.us, InsiderFinance, TECHi, FinancialContent, 24/7 Wall St., MEXC, Stocktwits, and Intellectia.AI, among others.
These are not sources a buy-side analyst can cite in an investment committee memo. They are downstream of the actual Tier 1 sources, which are SEC filings, sell-side research, Bloomberg, company earnings calls, FactSet, and reputable news sources trusted by Wall Street. Terminal X's footnote list pulls directly from the primary sources an institutional analyst would cite themselves. Specifically, the Jake portfolio report cites SEC filings by company and form type (PLTR 8-K, COST 10-Q), company earnings calls by quarter (Alphabet Q4 2025, Palantir Q4 2025), sell-side research by firm and report title (BofA Cloud Capex Preview, Citi Software's $100B+ AI Shock), Bloomberg news content, and Seeking Alpha and CNBC for macro color.
Architecturally, the gap exists because Claude Research is a general purpose web research tool. Its data corpus is what Google indexes, which heavily overweights aggregator content as it is highly SEO-optimized. Claude cannot access paywalled sell-side research, your firm's private data room, premium news sources, or proprietary transcript databases, which is where an institutional analyst's primary sources actually live.
Terminal X's research agent has access to the full SEC and DART filing corpus with structured retrieval, earnings call transcripts that are searchable, timestamped, and speaker-indexed, global and Korean and Japanese sell-side research across hundreds of firms, Bloomberg content through direct integration, and the user's own private data room documents. For a Korean pharma name, a Japan-listed industrial, or a European bank, the primary research will not be found through generic web crawling. The information your firm needs is in language specific sell-side notes that the open web simply does not see.
This gap gets even wider when working in CJK languages. Clients with Asian exposure rely on Korean and Japanese broker notes, DART filings, and Korean-language earnings coverage that simply do not live on Google and the open web. Claude will do its best to translate what it can find, but coverage is often minimal while translations lose analyst shorthand and jargon. Terminal X operates natively in both languages across the broker and news corpus, and our prompting itself is designed to handle all languages, ensuring that the outputs read like a native analyst wrote them.
On the subject of speed, Terminal X’s five minutes versus Claude’s twenty-nine minutes to final generation is a difference that will fundamentally change how an analyst approaches AI usage. Five minutes from prompt to final report is short enough for a user to check through their inbox and have the draft waiting for them before they’re finished.
On the contrary, if an analyst attempts to use Claude and waits twenty-nine minutes for a markdown report, they are less likely to use AI for their report tomorrow, and will likely go back to doing it the old way. Speed also determines whether the output receives thoughtful edits post generation. With Terminal X producing a draft in five minutes, an analyst can download the report, verify sources and figures, make changes, or regenerate again, quickly converging on a usable report in one sitting. With a twenty-nine minute generation time, iterating with Claude Research can quickly consume an analyst or PMs entire morning.
Terminal X’s speed can be attributed to two places. The first is source quality, as Terminal X pulls from a corpus of primary sources ranging from SEC filings, broker research, premium news like Bloomberg and the Wall Street Journal, FactSet, earnings calls, and more. This eliminates the need to spend countless long rounds of research aimlessly scraping irrelevant websites for information (which bloated Claude’s source list to over 1000 sites). The second is the prompting and retrieval pipeline itself, built by finance professionals who know which sources matter for a given query and how to weight them. In this way, our agents are able to pinpoint the proper source for the job instantly without the need to constantly course correct.
Claude's output hits the top-level section headings and is dense with data, but the format on the page is a generic long form research report with little regard for the template style it was based upon. Inline source tags bleed directly into the prose; phrases like "Yahoo Finance +4" and "TRADING ECONOMICS" appear mid-sentence, interrupting the reading experience. The Individual Stock Review paragraphs run substantially longer than the template's short paragraph convention, and because the final output is a PDF, the analyst cannot edit it directly.
Terminal X's output runs through the template-aware pipeline we built to fix this exact problem. The final Word document generation stage replicates the template's visual style, fonts, spacing, headers, and table formatting. All bold header conventions are preserved and the individual stock paragraphs are short, front loading the thesis in the same manner as the template, and citations are footnoted numerically instead of being inlined in the prose. Terminal X recognizes the importance of capturing the details that are significant to match the standards of our institutional clients.
Claude Research outputs a markdown by default that the user has to copy out manually, while giving the option to download to PDF. Recognizing the frustration of transferring over obscurely formatted research, Terminal X generates Custom Reports as a Word document output. All edits are done in Word. This is where analysts add firm specific language, flag positions to watch, and make any additional formatting tweaks. A PDF output forces a copy-paste step where the analyst extracts the text, reformats in Word, spends extensive time retracking and re-citing sources, rebuilds the tables that didn’t copy over, and reapplies firm branding. Markdown forces an even larger copy-paste effort where the analyst essentially rebuilds the document from scratch using the firm template. While it is possible for Claude to generate outputs in both Word and PDF, this has to be made upon request, and it often struggles to replicate the intricacies of the user’s original report style, including fonts, logos and charts. For a weekly portfolio report generated across many equities, the time to make edits on these inaccurate documents can quickly add up.
Claude Research has certain strengths that are superior to Terminal X Custom Reports, especially in places where Reports wasn’t built to compete. As a general tool, Claude pulls from a much broader web corpus, which matters if you are researching topics outside of institutional finance. Claude is also built for conversational iteration, where a user can go back and forth inside a single session and refine the output over multiple turns. Terminal X's Custom Reports currently functions best as a single-shot workflow, built to produce a near-finished document on the first pass rather than through a long conversation. If your main workflows are less finance focused and do not require institutional sources, real-time prices, template matching, or editable Word output, Claude is a reasonable choice.
The final comparison shows that Claude Research is a capable general research tool, while Terminal X functions better as a research product for financial institutions.
Notably, this gap compounds on recurring workflows such as the portfolio review from the head-to-head. An analyst running this report every Monday on Claude waits nearly half an hour for their output, only to spend another hour reformatting it into their firm's template and catching the price errors and hallucinations that come from pulling sources off the open web. The same report on Terminal X takes five minutes, comes out already matching the template, and gets sent along after light edits. These time differences add up to weeks of analyst capacity, making Terminal X an attractive choice for firms looking to take the next step in their AI utilization.
Full outputs from both systems are available upon request, and we are happy to run a live head-to-head on a query or report of your firms choosing. Please reach out at [email protected] to set up a demo and try Custom Reports yourself.
.png&w=3840&q=80)