AI Financial Research Source
How DealCharts Became a Source for AI Financial Research
DealCharts did not get AI attention by publishing an "AI" landing page. It got attention because the site put structured finance facts where retrieval systems could reach them: stable deal URLs, source-linked pages, facts JSON, dataset exports,
, and chart embeds that preserve attribution.llms.txt
The original answer engine optimization note framed the story around 1,185 structured finance deals with full lineage. By the July 2026 citation-readiness audit, the public dataset surface had grown to roughly 1,230 deal CSVs, 41 BDC CSVs, and an AI-surface sitemap near 16,000 machine-readable entity URLs. The number changed. The pattern did not: make the evidence easy to fetch, quote, and verify.
This is the practical answer to how to get cited by ChatGPT in a high-trust finance niche. Publish original data that answers a real question, keep the URL stable, expose the same fact in human and machine forms, and show the source trail clearly enough that a model can attribute the answer without guessing.
What the February 2026 Pulse Showed
The 2026-02-19 AEO pulse found AI systems already reading DealCharts before the site had a formal AEO article. The useful signal was not one bot name. It was the mix of model crawlers, retrieval user agents, referrers, and page-level concentration.
| Signal | Observed volume | Why it mattered |
|---|---|---|
| ChatGPT-User | 415 hits/week | Retrieval-style visits from ChatGPT sessions, including repeated hits to finance answer pages. |
| ClaudeBot | 3,352 hits/week | The heaviest AI crawler in that pulse, showing broad discovery beyond ordinary search. |
| PerplexityBot | 62 hits/week | Search-answer crawler traffic, plus a Perplexity.ai referrer observed on the adjacent GitHub edgar-change-interpreter repo. |
| OAI-SearchBot | 24 hits/week | OpenAI search-crawl activity separate from ChatGPT-User retrieval sessions. |
| Amazonbot | 81 hits/week | Another non-Google crawler class reaching structured finance pages. |
Two pages were especially revealing. ChatGPT-User repeatedly hit
about eight times per day and
about five times per day. Those are not broad brand pages. They are narrow answers: one entity credit question and one identifier-mapping workflow. That is exactly where AI search tends to need an attributable source.
What We Can and Cannot Infer
Crawler traffic is not identical to a citation. A bot hit proves a page was fetched. A ChatGPT-User visit is closer to answer retrieval, but it still does not reveal the user prompt or whether the final answer displayed DealCharts as a source. A Perplexity referrer is stronger, but still page-level rather than prompt-level evidence.
The reliable conclusion is narrower and more useful: DealCharts had already become retrievable by AI systems for structured finance questions. That made the next job clear. Tighten the pages so retrieval can turn into clean attribution.
Why Answer Engine Optimization Worked Here
Most finance sites try to win answer engines with prose. That is weak. The stronger pattern is AI citation structured data: the same claim appears as readable text, JSON-LD, a machine endpoint, and a source-linked citation path.
DealCharts had several advantages:
- Stable entity URLs. CMBS and Auto ABS pages use durable paths such as
instead of query strings or session URLs./capitalmarkets/abs/cmbs/ - Machine-readable facts. Deal and entity pages point toward
JSON, such as/llm/facts/
, so a model can retrieve compact context without scraping every table./llm/facts/jpmcc2017-jp6.json - Dataset exports.
exposes CSV surfaces and metadata for analysts who need rows, not just article text./datasets/ - Discovery files.
,/llms.txt
, andsitemap-llm.xml
make the AI-facing surface explicit instead of hoping crawlers infer it.sitemap-datasets.xml - Source lineage. Pages tie answers back to SEC EDGAR filings, CIKs, accessions, reporting periods, and as-of dates.
- Embeddable attribution. Chart embed paths and oEmbed support keep a DealCharts source link attached when a chart travels into another article or workflow.
For structured finance professionals, this matters because a cited answer is only valuable if it is checkable. "The model said so" is not a research artifact. "The model cited a deal page, the deal page links to facts JSON, and the facts JSON carries an as-of date and source trail" is closer to how an analyst actually works.
The Structured Data Pattern
The pattern that worked is simple enough to copy:
human answer page-> JSON-LD schema-> facts JSON or dataset CSV-> source filing or documented provenance-> stable citation snippet
The human page gives an answer that can be quoted. JSON-LD tells search and answer engines what the page is about. Facts JSON gives retrieval systems a smaller, less ambiguous payload. Dataset CSV gives analysts and models a tabular artifact. Source links keep the whole chain auditable.
That pattern now shows up across the site. The CMBS delinquency tracker gives a market-level surveillance view. The CMBS maturity wall tracker translates disclosed maturity buckets into refinance-risk context. The EDGAR API guide and SEC EDGAR API article explain the upstream source layer. The older LLM facts endpoint guide and AI provenance citations article describe the same architecture from the developer side.
What Finance Sites Should Copy
If you publish financial data and want answer engines to cite it, do not start with a "please cite us" page. Start with the data contract.
- Put one exact answer near the top of each page.
- Include the entity name, metric, source, and as-of date in that answer.
- Use stable canonical URLs and avoid changing slugs after publication.
- Add JSON-LD that matches the visible page content.
- Expose a compact facts endpoint or dataset file for high-value pages.
- Link to the underlying source document, not just to your own summary.
- Maintain
,robots.txt
, and sitemaps so crawlers can discover the surface.llms.txt - Track crawler classes separately from human search traffic, because GA and GSC do not show the full AI retrieval surface.
This is the boring part of answer engine optimization, and it is the part that compounds. A model can ignore marketing language. It has a harder time ignoring a stable, source-backed answer that is easier to cite than every alternative.
What We Still Had to Fix
The July 2026 citation-readiness audit scored DealCharts at 76/100 overall and 98/100 for robots.txt and AI crawler access. That was good enough to prove crawlability, but it also exposed gaps: stale machine promises, some dead API examples, thin metadata on a few pages, and places where human pages and facts JSON did not yet agree perfectly.
That is the right kind of audit finding. The goal is not to declare victory because ClaudeBot or ChatGPT-User found the site once. The goal is to make every public claim easier to quote, easier to verify, and harder to misread.
Frequently Asked Questions
What is answer engine optimization for financial data?
Answer engine optimization is the practice of making a page easy for AI answer systems to retrieve, quote, and attribute. For financial data, that means stable URLs, source-backed facts, explicit as-of dates, JSON-LD, machine-readable facts files, and clear citation language.
How do you get cited by ChatGPT for finance research?
You cannot force a citation, but you can make citation likely by publishing original, source-linked answers at stable URLs, exposing machine-readable data, allowing AI crawlers, keeping pages fresh, and writing extractable answer blocks that include the exact filing, date, entity, and metric behind the claim.
What made DealCharts useful to AI crawlers?
DealCharts exposed structured finance entities as crawlable pages, paired them with facts JSON, dataset CSVs, sitemaps,
, embed paths, and source links back to SEC filings. That combination gave AI crawlers both readable context and machine-checkable data.llms.txt
Is AI crawler traffic the same as being cited?
No. Crawler traffic proves that AI systems or their retrieval agents reached the site. ChatGPT-User and Perplexity referrers are stronger citation or retrieval signals, but prompt-level citation coverage still needs separate monitoring.