Sample report · redacted

The free baseline report, with the names taken out.

This is a real report from September 28, 2026, for a seed-stage financial-data company we approached cold. Its name, domain, founder and product names are replaced. Every question, number, competitor and quotation is as measured.

You get one of these for your own category, free, whether or not we end up working together. It is five pages as a PDF; this page is the whole of it. What it is not: a score, a forecast, or a promise that anything will move.

how the numbers are made → Methodology · the same test on us → Lab

Methodology note

We ran all 20 questions on September 28, 2026, once with Claude Opus 5 and once with OpenAI GPT-6 Sol (via Codex), both with web search on. That produced 40 live answers and 292 citations. The site and third-party checks were done the same day. Nothing in this report is modeled.

This is a one-day snapshot; Gemini, ChatGPT and Perplexity were not tested. A † marks a question where at least one answer did not search the web. GPT-6 Sol searched on every question except the bare brand name; Claude searched on 12 of the 17 questions that leave out the company’s name. Claude’s citations come from the source list at the end of each answer, GPT-6 Sol’s from links inside its answers.

1. Headline

Company A was recommended on 5 of 20 questions. On five questions that each describe a product it sells (SEC filings with standardized statements, delisted fundamentals, operating KPIs, pay-as-you-go credits and real-time filing alerts), none of the 10 answers named it.

The most frequent recommendation was not a vendor. On 12 of the 17 questions without the company’s name, the answers sent buyers to the SEC’s free EDGAR API.

Company AFinancial Modeling Prepsec-api.ioSEC EDGAR (free)
Recommendation rate25% (5 / 20)40% (8 / 20)30% (6 / 20)60% (12 / 20)
Mention rate45% (9 / 20)65% (13 / 20)30% (6 / 20)75% (15 / 20)
Non-brand questions named6 / 1712 / 176 / 1713 / 17
Own domain cited, non-brand8 of 269 citations241326

Company A starts with real assets. Asked its own homepage line, how to connect an agent to the stock market, Claude recommended it first, from memory. GPT-6 Sol picked it first for clean JSON with a link to the source filing, and quoted its pricing page exactly. Both engines tied the founder’s two repositories to the company, and both called it legitimate. The site lets every major AI crawler in and serves llms.txt, markdown copies of each page and an OpenAPI spec.

2. The 20 questions

Each question received two answers. The questions follow the homepage, pricing page, docs, blog and Y Combinator launch post.

Category discovery (7 questions)

#QuestionCompany AWho gets recommended
1Best stock market data API for AI agents?AbsentPolygon / Massive (both), FMP, Alpha Vantage
2Which providers have an official MCP server for Claude or ChatGPT?Claude: recommended. GPT-6 Sol: listed, 4th of 7Alpha Vantage, FMP, Massive, EODHD; FactSet, LSEG for funds
3Best API for SEC filings and standardized statements in an LLM app?AbsentSEC EDGAR, sec-api.io, Intrinio, FMP
4 †Stock API with fundamentals for delisted companies, for backtests?AbsentSharadar, first in both
5APIs for operating KPIs and management guidance?AbsentDaloopa, LSEG, FactSet, Visible Alpha, Bloomberg
6Pay-as-you-go stock data API for a solo developer?AbsentDatabento, first in both
7Alerts within seconds of an 8-K, Form 4 or 13D filing?Absentsec-api.io, in both; SEC feeds

Questions 3 to 7 each match a product Company A sells; none of the 10 answers named it. Its site says its data is “survivorship-bias-free for backtesting,” lists KPI, guidance and non-GAAP endpoints, processes filings “within seconds of publication,” and sells $20 of “Pay-as-you-go” credits.

Question 6 went to its own supplier. Both engines picked Databento; the company’s markdown homepage says its stock prices come from Databento.

Scenario and problem (5 questions)

#QuestionCompany AWho gets recommended
8 †My agent needs prices, statements and SEC filings. Easiest way to connect it to the stock market?Claude: recommended first. GPT-6 Sol: absentGPT-6 Sol: FMP, plus SEC EDGAR
9Bloomberg and FactSet cost too much. Clean JSON with a link to the source filing?GPT-6 Sol: recommended first. Claude: absentClaude: SEC EDGAR, sec-api.io, DataCedar, EvidInvest, FMP
10 †My agent gets revenue and EPS wrong. Fundamentals checked against SEC filings?AbsentSEC EDGAR Company Facts, in both
11[The founder’s two open-source projects] ask for a data key. Which provider, and a cheaper option?Recommended in bothCompany A; forks using Finnhub and Alpaca
12Consumer investing app: which APIs allow redistribution without an enterprise contract?GPT-6 Sol: recommended (Scale, $2,000/month). Claude: absentDatabento, SEC EDGAR, Intrinio, Twelve Data

Questions 8, 9 and 10 restate the company’s own pitch. Each engine named it once, never on the same question. On question 8, Claude answered from memory and called it “purpose-built for this exact trio.” GPT-6 Sol searched and chose FMP. On question 9, GPT-6 Sol said “For a startup agent, I’d start with [Company A]”; Claude named DataCedar and EvidInvest, which “explicitly store filing_date + accession + source_url per normalized row.” On question 10, both pointed to the SEC’s free Company Facts API.

Question 11 names the founder’s repositories, and both engines tied them to the company.

Competitive comparison (4 questions)

#QuestionCompany AWho gets recommended
13FMP vs Polygon vs Alpha Vantage for fundamentals and SEC filings?AbsentFMP, first in both; Polygon / Massive
14 †Parse EDGAR with edgartools, or pay for a data API?Absent"Start with edgartools and SEC EDGAR" (GPT-6 Sol); Claude the same
15 †Cheaper alternatives to Bloomberg, FactSet or Capital IQ for a small fund?AbsentPolygon, Sharadar, FMP, sec-api.io, EDGAR, Koyfin
16Equibles vs EODHD vs FMP: which MCP server works best with Claude?AbsentClaude: EODHD. GPT-6 Sol: Equibles

Question 15 is the launch post’s own framing: “Bloomberg, FactSet, and S&P spent 30 years building market data for humans staring at terminals.” Claude named more than 20 alternatives, GPT-6 Sol six. Neither included Company A.

Long-tail (1 question)

#QuestionCompany AWho gets recommended
17Most accurate LLM for extracting KPIs from earnings releases? Is there a benchmark?Claude: [its own benchmark] cited. GPT-6 Sol: absentGPT-6 Sol: RedCrown, HiFi-KPI, FinanceBench

The two engines gave opposite answers to question 17. Claude: “there is one benchmark that targets exactly this task,” and named the company’s own. GPT-6 Sol: “I found no public benchmark that compares Claude Opus, Claude Sonnet, and GPT head to head on that exact task.”

Brand-direct (3 questions)

#QuestionCompany AWho gets recommended
18 †"Company A" (the bare name)Claude: correct company, listed first. GPT-6 Sol: read as a generic phraseGPT-6 Sol asked what the user wanted to do with the phrase
19Company A reviews: is it legit?Legit, 2 of 2Both: few independent reviews
20How much does the Company A API cost?Priced, 2 of 2GPT-6 Sol matches the pricing page; Claude adds a free tier and old per-request rates

With “reviews” or “API” added, both engines found the right company. Claude called the founder’s repositories “the strongest single signal here.” No answer confused it with the founder’s 2024 Python library of the same name.

3. Competitive set

Who wins the answers

BrandRecommended onWhat the answers drew on
SEC EDGAR APIs (free)12 / 20sec.gov documentation, 26 citations
Financial Modeling Prep8 / 20Its developer docs, 24 citations; the MCP page alone 6
Polygon.io / Massive8 / 20massive.com docs, 10 citations
sec-api.io6 / 20Its docs, 13 citations; won both filing-alert answers
Company A5 / 20Its pricing page; the founder’s repositories on question 11

FMP, Polygon / Massive, Alpha Vantage and EODHD are Tier 1. The SEC’s free API, which Company A itself parses, was recommended more often than any of them.

The size inversion: MCP servers

MCP server on GitHubStars, Sep 28Answers recommending it, questions 1, 2 and 16 (of 6)
Company A’s MCP server2,3001 (Claude, question 2)
massive-com/mcp_massive3922
daniel3303/Equibles2312
alphavantage/alpha_vantage_mcp2123
EodHistoricalData/EODHD-MCP-Server213

The sharpest case is question 1, which nearly repeats the homepage title. Both answers recommended Massive, FMP and Alpha Vantage, not Company A, whose repository has more stars than the Massive, Alpha Vantage and EODHD servers combined. That repository was last updated June 5, 2025.

Equibles is the closest Tier 2 company: an MCP-first API for filings, KPIs and ownership. It publishes a “[Company A] Alternative: Equibles Compared” page, which Claude cited on question 19.

The pages AI found

We could read 19 of the 23 “best of” and comparison pages cited on non-brand questions. 16 were published by vendors; 8 rank the publisher first. Three mention Company A, all written by competitors: one ranks it first by GitHub stars, one seventh, and one lists it at “Free–$49/mo,” a price that is not on its site. Company A has two blog posts and no comparison page.

4. Source map: where AI gets its answers

The 34 answers to the 17 non-brand questions carried 269 citations, 185 from Claude and 84 from GPT-6 Sol.

Source typeShareExamples in the answers
Vendor sites, docs and vendor-written lists62.1%FMP, sec-api.io, EODHD, Lambda Finance
SEC, exchanges, government11.9%sec.gov
Other GitHub repos, press, blogs, directories, research19.3%Community MCP servers, Medium, arXiv
Company A and the founder’s repositories6.7%Pricing page, the two repositories

Sources Company A owns or can claim

SourceControllabilityNon-brandBrandStatus
companya.example and its docsHighest87Pricing page cited 5 times; none of the 57 docs pages on non-brand questions
The founder’s two repositoriesHighest7, all on question 112About 90,000 stars between them; one README line each
Company A’s MCP serverHighest10Last updated June 5, 2025
Directory listings, YC, founder’s LinkedIn, partner docsHigh25PulseMCP and ChatGPT listings on question 2; the rest only on question 19
G2, Trustpilot, Capterra, Product HuntHigh00All blocked direct checks

Owned source share: 6.7% (18 of 269 non-brand citations). Without question 11, which names the founder’s repositories, it is 3.5% (9 of 255). On the three brand questions it was 61% (14 of 23).

5. What AI currently says about Company A

Accurate

GPT-6 Sol’s price answer matched the pricing page line for line, and it read the terms correctly: redistribution only on the Scale plan. Claude gave the coverage as “27,000+ active and delisted US tickers”; no answer repeated the “global” claim on the pricing FAQ.

Claude repeated four claims the site’s own files contradict

ClaimWhereWhat Company A’s own files say
A "$0" free tierQuestion 20Its skill.md: "There is no free tier"; every page’s JSON-LD gives a price of 0
Per-endpoint prices, $0.01 to $0.10Question 20Match the rate card in its llms-full.txt; the pricing page shows only plans
Five tickers "with no key at all"Question 11Keyless requests return 401
"options, crypto" among its datasetsQuestion 18Docs: options "are not yet available"; the API index still lists crypto

The site’s machine-readable files disagree with each other. The pricing page lists Scale at $2,000 a month; its markdown copy calls it a “custom price”; llms-full.txt still lists two plans that no longer exist.

The strongest proof is not on the site

“Trusted by 1,000+ customers” appears in the YC launch post, not on the company’s own site. The review pages the answers found were an aggregator, a competitor’s comparison and a blog post by a friend of the founder. Claude: “There is no real independent review corpus.”

Buyers in this test asked for delisted fundamentals, operating KPIs, pay-as-you-go access and filing alerts. Company A sells each of these.

The same report for your category

Leave your work email and website and we run your buyer questions the same way. It takes us a day, it is yours to keep, and if the answer is that you are named everywhere already, the report says that.

Get your free baseline report