Best AI Portfolio Research and Thesis-Generation Tools in 2026

Last updated June 2026

Short answer

AI portfolio research means using AI to gather and reason over company, sector, and theme data; thesis generation is the next step, turning a view into a structured, testable investment thesis with supporting data. There is no single best tool. AlphaSense is the strongest for source-grounded research across filings and transcripts, FinChat (Fiscal.ai) and Perplexity Finance are fast and cited, Kensho is institutional data infrastructure, Bloomberg Terminal AI suits Terminal users, Danelfin gives a quantitative single-stock score, Magnifi is conversational for fund discovery, and Walnut turns a thesis you describe into a weighted basket on your own broker that you approve. Research tools inform the thesis; only a thesis-to-portfolio tool expresses it as holdings.

“Use AI to research my portfolio” covers two jobs that sound like one. The first is research: reading filings, transcripts, fundamentals, and news fast enough to actually form a view. The second is thesis generation: turning that view into a structured, testable investment thesis, then expressing it as positions. Different tools live at different points on that path. This guide covers eight of them (AlphaSense, FinChat (Fiscal.ai), Kensho, Perplexity Finance, Bloomberg Terminal AI, Danelfin, Magnifi, and Walnut), describes each on the same fields, ranks them by use-case, and is honest about where each one, including Walnut, is the wrong fit.

What thesis generation actually means

Thesis generation is turning a view into a structured, testable investment thesis with supporting data. A loose hunch (“AI infrastructure looks strong”) is not a thesis. A thesis names the claim, the evidence, the risks, and how you would act on it. The pieces worth getting right:

The claim. A specific, falsifiable statement about a company, sector, or theme, so you can later check whether it held up rather than rationalize after the fact.
The supporting data. The fundamentals, filings, transcripts, and prices that back the claim. This is where research tools like AlphaSense, FinChat (Fiscal.ai), and Perplexity Finance do the work, ideally with citations you can trace.
The risks. What would prove the thesis wrong, and what you are exposed to if it does. A thesis without an explicit downside is just optimism.
The expression. How the thesis becomes positions: which securities, in what weights. A view that never turns into holdings is an opinion, not a portfolio. This is where Walnut fits, proposing constituents and target weights you approve.

Most tools are strong on one or two of these. Research engines nail the supporting data; a thesis-to-basket tool handles the expression. Almost none do the whole path, which is why people stack two together.

Why research and thesis tools are not the same category

The most common mistake is treating “AI research” and “AI thesis generation” as one purchase. They sit at different points and rarely overlap well:

Research engines (AlphaSense, Perplexity Finance, FinChat, Bloomberg Terminal AI, Kensho) gather and summarize data so you can form a view. They are excellent at the supporting-data step and stop there: none of them connect your brokerage or turn the view into weighted positions.
Stock scorers (Danelfin) output a number on a single ticker. That can seed or test a view, but a score is not a reasoned thesis and not a portfolio.
Thesis-to-portfolio tools (Walnut) start from a view you state and express it as constituents and target weights on your real broker. They are strong on the expression step and lean on you, or on a research engine, for the source documents behind the thesis.

Knowing which step you are stuck on tells you which tool to reach for. The full set is below.

What to look for in an AI research and thesis tool

How it sources answers: grounded in primary documents with citations (AlphaSense, Perplexity Finance), built from structured fundamentals (FinChat), or a model score with no narrative (Danelfin). Citations let you verify; a bare score does not.
Whether it shows its reasoning, not just a verdict. This matters most for anything that shapes a thesis you will act on.
Whether it stops at research or helps you act: most research tools end at the analysis; only a few express the thesis as positions.
Whether it connects your real accounts read-only, so the thesis is grounded in what you actually hold. Most research engines do not; Walnut does, through SnapTrade.
Cost model: a free tier, a flat subscription, or enterprise pricing. AlphaSense, Kensho, and Bloomberg are institutional; FinChat, Perplexity, and Walnut have free tiers.
Honesty of the output: a tool that frames a thesis as testable, with risks and sources, is more useful than one implying a guaranteed market-beating call. No tool can promise a thesis plays out.

The eight AI research and thesis tools worth knowing

Each tool below is described on the same six fields, so you can scan across them: what it is, what the AI does in research, how it sources its answers, the pricing model, who it suits, and one honest limitation.

AlphaSense

A market-intelligence search engine that indexes filings, earnings-call transcripts, broker research, news, and expert-call notes, then layers a generative AI assistant that summarizes and answers questions across that corpus.

What the AI does: Searches a deep document library and generates sourced summaries, sentiment reads, and answers grounded in primary documents.
How it sources answers: Grounded in indexed primary documents (filings, transcripts, broker research, expert calls).
Pricing model: Enterprise subscription (priced for funds and corporates; verify on their site).
Best for: Deep, source-grounded research across filings, transcripts, and broker reports.
One honest limitation: Built and priced for institutions; it informs a thesis but does not connect your brokerage or place trades.

FinChat (Fiscal.ai)

A conversational research tool, now branded Fiscal.ai, that answers natural-language questions about public companies using structured fundamentals, segment data, and KPIs, with charts and tables generated from your prompt.

What the AI does: Answers plain-English questions about a company's fundamentals and builds charts and tables from structured financial data.
How it sources answers: Grounded in structured fundamentals, segments, and KPIs (cites the underlying figures).
Pricing model: Free tier plus paid plans (flat subscription).
Best for: Fast, conversational fundamental research and KPI comparisons on individual companies.
One honest limitation: Centered on company fundamentals, not your own portfolio; it researches the thesis but does not act on it.

Kensho (S&P Global)

An AI and analytics arm of S&P Global that builds machine-learning tools for financial data: entity linking, document extraction, transcription, and analytics over S&P Global's datasets, used mostly inside institutional workflows.

What the AI does: Extracts, links, and analyzes financial data at scale (NLP over documents, entity resolution, analytics) feeding institutional research.
How it sources answers: Grounded in S&P Global datasets and extracted document data.
Pricing model: Enterprise (embedded in S&P Global products; not a consumer subscription).
Best for: Institutional data extraction and analytics that feed a research process.
One honest limitation: Infrastructure and data tooling rather than a tool an individual investor sits in front of to draft a thesis.

Perplexity Finance

The finance mode of the Perplexity AI answer engine: ask a question about a company, a sector, or a macro theme and get a cited answer with live pricing, fundamentals, and links to the sources it used.

What the AI does: Answers research questions with live market data and inline citations, so you can trace each claim to a source.
How it sources answers: Cites live web and market-data sources inline with each answer.
Pricing model: Free tier plus a Pro subscription (flat).
Best for: Quick, cited answers and a fast first pass when forming a view on a company or theme.
One honest limitation: A general answer engine, not a portfolio tool; it can ground a thesis but cannot see or act on your holdings.

Bloomberg Terminal AI

The AI features layered onto the Bloomberg Terminal, including an AI-powered document search and a generative summarization of earnings calls and filings, sitting on top of Bloomberg's market data, news, and analytics.

What the AI does: Summarizes earnings calls and documents and answers questions across Bloomberg's data and news inside the Terminal.
How it sources answers: Grounded in Bloomberg's licensed market data, news, and filings.
Pricing model: Bundled into the Bloomberg Terminal subscription (institutional, several thousand dollars per user per month range; verify).
Best for: Professionals who already live in the Terminal and want AI summarization on top of it.
One honest limitation: Priced for institutions and tied to the Terminal; overkill for an individual and not connected to your retail brokerage.

Danelfin

An AI stock-scoring service that assigns each stock an AI Score from 1 to 10 estimating its probability of beating the market over the coming months, built from a large set of technical, fundamental, and sentiment features.

What the AI does: Scores individual stocks (AI Score 1-10) and ranks them on the probability of beating the market.
How it sources answers: A quantitative model over technical, fundamental, and sentiment features (a score, not a sourced narrative).
Pricing model: Flat subscription.
Best for: A quantitative signal to test or seed a single-stock view.
One honest limitation: It outputs a score, not a reasoned thesis or supporting narrative, and it does not connect your broker or build a portfolio.

Magnifi

A conversational AI investing assistant you can ask natural-language questions about funds, stocks, and holdings, with fund-discovery and account-connection features.

What the AI does: Answers plain-English research questions and helps discover funds and securities that fit a view.
How it sources answers: Conversational over fund and security data (more discovery than document-grounded research).
Pricing model: Flat subscription.
Best for: Research and fund discovery in plain English when shaping a view.
One honest limitation: Skews toward fund discovery rather than building a structured, weighted thesis on your own broker.

Walnut

Connects your real brokerage through SnapTrade and turns a thesis you describe in plain language into a thematic basket: you state a view, it proposes constituents and target weights, and you approve before anything trades.

What the AI does: Turns a stated thesis into a structured basket (proposed constituents and target weights) you can approve, then tracks it against the S&P 500.
How it sources answers: Conversational, working from the view you describe plus live prices and your real connected holdings.
Pricing model: Free tier.
Best for: Going from a thesis to a structured, weighted basket on the broker you already use.
One honest limitation: It sits on top of your broker rather than being a deep research terminal, so it leans on you (or another tool) for the source documents behind the thesis.

At a glance

Tool	Best for	How it sources answers	Pricing model
AlphaSense	Deep, source-grounded research across filings, transcripts, and broker reports	Grounded in indexed primary documents (filings, transcripts, broker research, expert calls)	Enterprise subscription (priced for funds and corporates; verify on their site)
FinChat (Fiscal.ai)	Fast, conversational fundamental research and KPI comparisons on individual companies	Grounded in structured fundamentals, segments, and KPIs (cites the underlying figures)	Free tier plus paid plans (flat subscription)
Kensho (S&P Global)	Institutional data extraction and analytics that feed a research process	Grounded in S&P Global datasets and extracted document data	Enterprise (embedded in S&P Global products; not a consumer subscription)
Perplexity Finance	Quick, cited answers and a fast first pass when forming a view on a company or theme	Cites live web and market-data sources inline with each answer	Free tier plus a Pro subscription (flat)
Bloomberg Terminal AI	Professionals who already live in the Terminal and want AI summarization on top of it	Grounded in Bloomberg's licensed market data, news, and filings	Bundled into the Bloomberg Terminal subscription (institutional, several thousand dollars per user per month range; verify)
Danelfin	A quantitative signal to test or seed a single-stock view	A quantitative model over technical, fundamental, and sentiment features (a score, not a sourced narrative)	Flat subscription
Magnifi	Research and fund discovery in plain English when shaping a view	Conversational over fund and security data (more discovery than document-grounded research)	Flat subscription
Walnut	Going from a thesis to a structured, weighted basket on the broker you already use	Conversational, working from the view you describe plus live prices and your real connected holdings	Free tier

Ranked by where you are stuck

There is no overall number one, because the right tool depends on which step of the path you are on: gathering data, scoring a ticker, or expressing a finished thesis. Below the field is ranked inside each use-case, with the stronger fit first. Walnut leads only in its own category (turning a thesis into a structured basket on your own broker), not across the board.

Best for deep, source-grounded research

If you want a thesis built on primary documents (filings, transcripts, broker research) with citations you can trace, the document-grounded engines lead.

1. AlphaSense. Indexes filings, transcripts, broker research, and expert calls, then summarizes and answers questions grounded in those primary sources.
2. Bloomberg Terminal AI. Adds AI summarization on top of Bloomberg's licensed data and news for professionals already in the Terminal.

Best for fast, cited answers when forming a view

If you want to interrogate a company or theme quickly and check the sources behind each claim, the conversational answer engines fit.

1. Perplexity Finance. Answers research questions with live market data and inline citations, so you can trace each claim to its source.
2. FinChat (Fiscal.ai). Conversational fundamentals: ask about a company's segments and KPIs and get charts and tables built from structured data.

Best for a quantitative single-stock signal

If you want a number to test or seed a view on one ticker rather than a narrative, the scoring tools fit.

1. Danelfin. Assigns each stock an AI Score (1-10) on the probability of beating the market over the coming months.
2. Magnifi. Conversational research and fund discovery to find securities that fit a view in plain English.

Best for turning a thesis into a structured basket

If you have formed a view and want it expressed as constituents and weights on the broker you already use, the thesis-to-basket tools fit.

1. Walnut. Describe a thesis and Walnut proposes constituents and target weights as a basket on your connected broker; you approve before anything trades, and it tracks the basket against the S&P 500.
2. Magnifi. Helps surface the funds and securities that express a view, though it stops short of a weighted, approved basket on your own broker.

How we evaluated these

We limited the field to tools that actually help you research a view or turn it into a thesis and positions, which is why pure robo-advisors and bare price trackers are not on the list. Within that set we weighed five things specific to research and thesis work:

Sourcing: whether answers are grounded in primary documents with citations, in structured fundamentals, or in an opaque model score.
Reasoning transparency: whether the tool shows the why behind an answer or hands you a verdict to trust blindly.
Where on the path it sits: research, scoring, or thesis-to-positions, and how honest it is about where it stops.
Grounding in your real holdings: whether it can read what you actually own, so the thesis is about your portfolio rather than a generic example.
Honesty of the marketing: we marked down anything implying a guaranteed market-beating thesis, because no tool can promise that.

We did not crown a single overall winner. The best tool depends on which step you are stuck on and how hands-on you want to be. Figures and features change; treat the specifics here as a starting point and verify on each provider's site.

Which one should you pick?

The quickest way to narrow it down is to match the tool to the step you are on.

You need deep, source-grounded research. AlphaSense searches filings, transcripts, and broker research with citations; Bloomberg Terminal AI adds summarization if you already use the Terminal.
You want fast, cited answers to form a view. Perplexity Finance answers with live data and inline citations; FinChat (Fiscal.ai) is conversational over company fundamentals.
You want a quantitative read on one stock. Danelfin scores each ticker (1-10) on the probability of beating the market; Magnifi helps discover funds and securities that fit the view.
You need institutional data infrastructure. Kensho extracts and analyzes financial data at scale across S&P Global datasets, feeding a larger research process.
You have a thesis and want it as a portfolio. Walnut turns a thesis you describe into a basket of constituents and target weights on your connected broker, which you approve, then tracks against the S&P 500.

Where Walnut fits

To be upfront, since this is our site: Walnut is a thesis-to-portfolio tool, and it leads in that category rather than overall. It is not a research terminal. Where AlphaSense, Perplexity Finance, and FinChat (Fiscal.ai) help you gather the data and form a view, Walnut picks up at the expression step. You describe a thesis in plain language, and it proposes constituents and target weights as a thematic basket on the brokerage you already connected through SnapTrade. You approve before anything trades, every trade needs your approval, and Walnut then tracks the basket against the S&P 500. It is read-only by default and you keep the broker you already use. Walnut is not an investment adviser.

Where Walnut is the wrong choice

Just as importantly, here is when another tool fits the research-and-thesis job better:

You want deep, document-grounded research. AlphaSense and Bloomberg Terminal AI search filings, transcripts, and broker reports with citations; Walnut is not a research terminal and leans on you or another tool for the source documents.
You want fast, cited answers before you have a view. Perplexity Finance and FinChat (Fiscal.ai) are built for that first research pass; Walnut starts once you already have a thesis to express.
You want a quantitative single-stock signal. Danelfin outputs an AI Score per ticker; Walnut builds and tracks a weighted basket rather than scoring one name.
You want institutional-grade data tooling. Kensho and the enterprise platforms are built for that scale; Walnut is a consumer app on top of your retail broker.
You do not want to connect a brokerage at all. Walnut sits on top of your real account, so it needs one. A pure research engine that never touches your holdings would suit better.

From a connected account you can dig into a specific stock, an ETF you hold, or a theme you want exposure to. For the wider field, see the best AI portfolio analyzers roundup, or how to connect your brokerage to an AI assistant.

Try Walnut on top of your broker

Walnut connects any major US broker in a few clicks, then turns a thesis you describe into a basket of constituents and target weights you approve, and tracks it against the S&P 500. Read-only by default; you approve every trade.

Create a free Walnut account

FAQ

What is AI portfolio research and thesis generation?

Portfolio research is using AI to gather and reason over data about companies, sectors, and themes. Thesis generation is the next step: turning a view into a structured, testable investment thesis with supporting data, such as a claim about a company or theme, the evidence behind it, and how you would express it as positions. Research tools inform the thesis; a thesis-to-portfolio tool expresses it as holdings.

What does thesis generation mean in investing?

Thesis generation means turning a loose view into a structured, testable investment thesis: a clear claim (for example, that a theme or company will do X), the supporting data and reasoning, the risks, and how you would act on it as specific positions and weights. A good thesis is falsifiable, so you can later check whether it held up. AI tools speed up the research and structuring; you still own the judgment.

What is the best AI tool for investment research in 2026?

There is no single best one; it depends on the depth you need. AlphaSense is strongest for source-grounded research across filings, transcripts, and broker reports. Perplexity Finance and FinChat (Fiscal.ai) are fast and conversational with citations. Bloomberg Terminal AI suits professionals already in the Terminal. Danelfin gives a quantitative single-stock score. Walnut turns a finished thesis into a weighted basket on your own broker.

Is AlphaSense good for portfolio research?

AlphaSense is one of the strongest source-grounded research tools. It indexes filings, earnings-call transcripts, broker research, and expert-call notes, then uses generative AI to summarize and answer questions across that corpus with citations. It is built and priced for institutions, so it informs a thesis well but does not connect your brokerage or place trades.

What does FinChat (Fiscal.ai) do?

FinChat, now branded Fiscal.ai, answers natural-language questions about public companies using structured fundamentals, segment data, and KPIs, and builds charts and tables from your prompt. It has a free tier plus paid plans. It is centered on company fundamentals rather than your own portfolio, so it researches a thesis but does not act on it.

Can ChatGPT or Claude generate an investment thesis?

They can help structure a thesis and stress-test the reasoning, but on their own they cannot see live prices or your real holdings, so they reason about generic examples and can get figures wrong. Pairing them with a research tool that cites sources, or with a tool like Walnut that connects your broker, grounds the thesis in real data and lets you express it as positions you approve.

Is Danelfin a research or a thesis tool?

Danelfin is a quantitative stock-scoring tool. It assigns each stock an AI Score from 1 to 10 estimating its probability of beating the market over the coming months. That is a signal, not a reasoned thesis with supporting narrative, and it does not connect your broker or build a portfolio. You would use it to test or seed a single-stock view, not to write the full thesis.

What is Kensho?

Kensho is the AI and analytics arm of S&P Global. It builds machine-learning tools for financial data, such as document extraction, entity linking, transcription, and analytics over S&P Global datasets, used mostly inside institutional workflows. It is research infrastructure rather than a tool an individual sits in front of to draft a thesis.

Does Perplexity Finance show its sources?

Yes. Perplexity Finance answers research questions with live market data and inline citations, so you can trace each claim back to the source it used. That makes it a fast first pass for forming a view. It is a general answer engine, though, not a portfolio tool, so it cannot see your holdings or act on the thesis.

How do I turn an investment thesis into actual positions?

Once you have a thesis, you express it as specific holdings and weights, then place the trades at your broker. Tools differ on how much they help. Research engines like AlphaSense and Perplexity Finance stop at the analysis. Walnut takes a thesis you describe and proposes constituents and target weights as a basket on your connected broker, which you approve before anything trades, then tracks it against the S&P 500.

How much do AI investment-research tools cost?

It varies widely. FinChat (Fiscal.ai), Perplexity Finance, and Walnut have free tiers. Danelfin and Magnifi are flat-subscription. AlphaSense, Kensho, and Bloomberg Terminal AI are enterprise-priced and built for institutions, with the Terminal running into thousands of dollars per user per month. Verify current pricing on each provider's site, because it changes.

Are AI research and thesis tools investment advice?

No. These tools surface data, summarize documents, and help you structure a view, but the judgment and the decision are yours. None can promise a thesis will play out. Walnut, for instance, is informational and is not an investment adviser: it proposes a basket from a thesis you describe, and you approve any trade. Treat every output as input to your own decision.

Walnut is informational and is not an investment adviser. App features, pricing, and availability change; verify current details on each provider's site before deciding. Nothing on this page is a recommendation to buy, sell, or hold any security or to use any particular product.