ChatGPT for Financial Services: what’s included, delayed and shareable

OpenAI’s September 10 launch brings selected financial datasets into a dedicated ChatGPT plan for institutions, with GPT‑6 Astra handling research, models and presentation materials. The [official launch announcement](https://openai.com/index/introducing‑chatgpt‑financial‑services/) says Morgan Stanley and Evercore were design partners and places the initial work around investment research, financial modelling and presentation materials. The practical change is that some research can begin with licensed data already available inside the product, reducing the work of connecting sources before an analyst can use them. That convenience leaves several distinct decisions for a firm. Buying the right workspace plan establishes product access. Selecting a dataset establishes what the analysis can cover. Its timestamp determines how current the evidence is, while the provider’s terms determine how the resulting material may be used. Those distinctions matter most when an internal answer becomes a spreadsheet, research note or client presentation. The [OpenAI organization profile](https://opentools.ai/organizations/openai) provides the stable provider reference; current availability and reuse limits remain governed by the product documentation and partner terms cited here. ChatGPT for Financial Services is a separate plan built on Enterprise. OpenAI’s [current help documentation](https://help.openai.com/en/articles/20001517‑chatgpt‑for‑financial‑services) says standard Enterprise does not include its listed financial datasets, and standard Enterprise seats cannot be mixed with Financial Services seats in one workspace. Access and pricing go through OpenAI’s sales or account team; eligibility depends on the organization and provider restrictions. A firm evaluating a handful of analysts therefore needs to settle the workspace arrangement before treating this as an individual feature upgrade.

OpenAI describes three routes into financial information: included data hosted and indexed on its infrastructure, connections using a firm’s existing subscriptions, and a broader connector ecosystem. Its [launch announcement](https://openai.com/index/introducing‑chatgpt‑financial‑services/) says included datasets need no separate provider contract or connector setup. It describes shared sign‑in and entitlement integrations with providers including S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva and Moody’s as work in progress, so that announcement does not establish universal availability of every named integration. PitchBook is a useful example of why a provider’s name is insufficient to describe coverage. Its [September 10 announcement](https://pitchbook.com/media/press‑releases/pitchbooks‑private‑market‑intelligence‑now‑accessible‑within‑chatgpt‑for‑financial‑services) identifies an expanded Essential firmographic dataset covering companies, investors and funds. It says this is indexed within the financial‑services product without separate connector activation. OpenAI’s current help page separately labels PitchBook as an Essential dataset, while the partner terms restrict substantial raw exports, CRM ingestion and model development. Included access should therefore be evaluated against the specific fields and permitted uses a research project needs, rather than assumed to reproduce every capability of the full PitchBook platform. [Crunchbase’s partner announcement](https://about.crunchbase.com/press/press‑releases/crunchbase‑private‑market‑data‑is‑available‑in‑chatgpt‑for‑financial‑services) describes structured company, funding, investor and acquisition information. Its examples include screening companies by market, geography, industry or funding profile and comparing investor activity. For a hypothetical analyst assembling a potential‑acquirer list, those fields could support the first screen. The analyst would still need to establish whether the selected sources contain the particular deal history or operating figures needed for the next stage. A company missing from a filtered result is not, by itself, evidence that it fails the investment criteria.

The following comparison brings together the [included‑source limits](https://help.openai.com/en/articles/20001517‑chatgpt‑for‑financial‑services) and [partner‑specific terms](https://openai.com/policies/financial‑services‑terms/). It describes selected documented boundaries, not every field or every contractual provision. | Source | Included coverage | Timing or allowance | Access or reuse boundary | |---|---|---|---| | Daloopa | Financial statements and selected metrics | 24‑hour delay; 3,000 datapoints per user per month | Included allowance is not an unrestricted data feed | | Nasdaq through FMP | US equity pricing and market data | At least 15 minutes delayed | Further dissemination requires Nasdaq authorization | | PitchBook | Selected private‑company and financing information | Coverage differs from the full product | Internal research; substantial standalone raw exports and CRM ingestion are restricted | | LSEG / Reuters Ready News | Financial and business news | Check the source timestamp | Eligible professionals at qualifying US‑domiciled or incorporated financial‑firm entities; internal‑use terms apply | The important distinction is between the time an answer was generated and the time of the evidence inside it. Consider a hypothetical Monday‑morning company comparison. A newly generated response could combine a prior reporting period’s financial statement, delayed market pricing and a recent news item. Calling the whole response “current” would conceal those different clocks. A useful worksheet would record the period covered, source timestamp and retrieval time alongside each material figure, so the next reader can judge whether the inputs belong in the same comparison. The Daloopa allowance also makes the size of a research request relevant. A request for one company and a few measures differs from a broad screen across many companies and periods. The published documentation names a datapoint cap; it does not define a conversion between prompts and datapoints. Treating 3,000 datapoints as 3,000 questions, or forecasting the exact number of screens it permits, would add a precision the stated allowance does not provide.

Under OpenAI’s [Financial Services Terms](https://openai.com/policies/financial‑services‑terms/), export and sharing features do not enlarge the rights granted over partner data. Ownership of generated output as between a customer and OpenAI does not transfer the underlying partner rights. PitchBook permits internal business research subject to restrictions including CRM ingestion and substantial standalone raw‑data exports. Reuters downloads, where available, are for internal business use; its terms also limit internal distribution and prohibit publishing the information on the internet. Those are source‑specific constraints, rather than a single rule attached to the finished file format. For a hypothetical client presentation, the useful question is therefore what material each page contains and who will receive it. A firm might have an internal working sheet with licensed source extracts, calculations derived from those inputs and an analyst’s narrative. Those components should not be assumed to carry identical reuse rights merely because they sit in the same workbook. The intended recipient and the exact material proposed for distribution need to be checked against the applicable source terms and the firm’s own approved workflow before delivery. Keeping citations with the work helps that review, but attribution is only one part of it. A source link lets a colleague trace a number back to its origin; it does not by itself answer whether the source material may be sent to someone else. The access decision belongs at the beginning of the assignment, while the distribution decision needs the actual proposed deliverable and audience.

OpenAI’s [launch description](https://openai.com/index/introducing‑chatgpt‑financial‑services/) says administrators can publish Excel, Word and PowerPoint templates and apply firm style guides. It also describes role‑based controls, configurable retention, supported compliance‑log exports and business data excluded from model training by default. These facilities can help a firm establish a consistent workflow, but a polished template does not demonstrate that its figures, assumptions or source permissions have been checked. Workspace owners can enable financial‑data permissions through defaults or role overrides after account setup, according to the [setup instructions](https://help.openai.com/en/articles/20001517‑chatgpt‑for‑financial‑services). A useful first assignment would have a defined question, an approved set of sources and a named intended audience. For the hypothetical company screen, the handoff could include the screening criteria, why each company qualified, the dates of the financial inputs and a list of fields still missing. A reviewer can then assess the actual reasoning instead of guessing how a confident‑looking result was assembled. OpenAI’s treatment of firm data is also separate from what the customer may do with licensed inputs. Under the [partner terms](https://openai.com/policies/financial‑services‑terms/), PitchBook forbids using its data for training, fine‑tuning, grounding or model development. Reuters restricts using its content, or associated data, signals or output, to train or develop AI models that persistently encode training data. A firm evaluating an internal model project therefore needs a separate permission assessment for its intended sources; the product’s own privacy setting does not settle that reuse question. *Photo: The Pioneer Building in San Francisco, photographed on July 27, 2019, when it housed OpenAI offices. Source: [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Pioneer_Building,_San_Francisco_%282019%29_-1.jpg). HaeB, [CC BY‑SA 4.0](https://creativecommons.org/licenses/by‑sa/4.0/). Previously cropped by the source; resized and cropped for display under the same licence.*