Cooley GO Public: ChatGPT Work for IPO Preparation
Cooley's proprietary GO Public workflow uses ChatGPT Work to prepare a client-specific starting point for IPO lawyers; public access and measured gains are not disclosed.
Cooley GO Public uses ChatGPT Work to help prepare initial public offerings, with lawyers reviewing the resulting work. OpenAI’s September 17, 2026 customer story describes Cooley’s proprietary system combining client information, public sources and selected precedents into a tailored starting point.
What changed with Cooley GO Public?
Previously, teams often adapted a comparable company’s precedent. The reported shift starts with the client’s circumstances. Cooley’s legal engineers, innovation counsel and practitioners translated their expertise into a controlled agent workflow, specifying automated steps and points requiring lawyer validation.
The firm’s stated aim is to free lawyers and management for judgment, disclosure review and strategic decisions. That is a reported benefit, not an independently measured outcome. The announcement provides no numerical time saving, accuracy benchmark, price, public signup procedure or supported-platform list. Its audience is IPO clients and their advisers; it does not establish that an ordinary ChatGPT account includes GO Public.
For readers evaluating ChatGPT Work for IPO preparation, the practical question is therefore what evidence reaches the reviewer. A fluent draft is easy to read, but a reviewable draft should make its assumptions visible. The sections below offer our own evaluation suggestions, not a description of undisclosed Cooley features.
What should an IPO team ask before using it?
Ask the provider to demonstrate a single proposed disclosure alongside the materials supporting it. Request separate labels for company-provided facts, precedent language and unanswered questions. This gives a reviewer a concrete basis for deciding whether the wording fits the company.
An IPO team could use these questions in a product demonstration:
- Can a reviewer identify the source and version behind each material assertion?
- When two documents disagree, does the output preserve the disagreement?
- Can the reviewer distinguish an actual company fact from a statement borrowed from a precedent?
- Who accepts changes, and how is that decision recorded?
- What happens when a required document is missing or unreadable?
These are suggested purchasing and workflow questions. We have not tested GO Public or confirmed how its interface answers them. Readers wanting a small, inspectable comparison exercise can use our document reconciliation example, which has a different purpose from this dated customer announcement.
An illustrative document-review exercise
A fictional company, Harbor Metrics, could prepare a synthetic packet containing a business overview, a management questionnaire and an earlier draft. Suppose the overview says one customer represents 24% of revenue, while the questionnaire says 19%, and neither explains the reporting period.
Our suggested exercise is to ask the system to produce an issue entry rather than choose a number. The entry should identify both statements, point to their locations, state that the periods are unknown and leave the resolution with a named reviewer. A successful result would preserve the ambiguity even if one number appears more plausible.
Next, add a precedent that describes subscription revenue while the fictional company’s materials describe transaction fees. Ask for a draft paragraph grounded only in the supplied company facts, with missing information listed separately. The reviewer should inspect whether subscription language slipped into the answer. That test targets a specific failure that a polished writing sample could conceal.
Finally, change one source document and repeat the exercise. Check whether the output follows the current version while retaining enough information to explain the revision. This is an untested editorial scenario, not an IPO disclosure template or a demonstration of Cooley’s implementation. Our document consistency check workflow provides a general structure for recording located discrepancies and matching controls.
How to measure a useful first draft
OpenAI’s evaluation guidance recommends task-specific tests, explicit success criteria, representative datasets and human feedback. It also recommends evaluating continuously as systems change, including typical, difficult and adversarial cases. Those principles can inform a buyer’s assessment, but they are not evidence of GO Public’s results.
For the fictional exercise above, our suggested scorecard would record unsupported assertions, missed conflicts, incorrect source locations and reviewer correction time. Set acceptance criteria before seeing the output. A reviewer might require every planted conflict to be surfaced and every unsupported sentence to be removed before a draft enters the next review stage.
Compare outputs on the same packet, with reviewers using the same instructions. Record both drafting time and correction time; otherwise a quick first response could simply transfer effort into checking. Keep examples of failures as future regression cases. Any claimed improvement should identify the sample, baseline and review method, making clear whether it measures drafting speed, factual consistency or the complete process.
What privacy checks belong before a pilot?
OpenAI’s business data commitments say organizational inputs and outputs from its listed business products and API are not used for model training by default. The page also describes encryption, access controls and retention options for qualifying organizations. These are general platform statements, not documentation of Cooley’s particular configuration.
Our recommendation is to begin demonstrations with synthetic materials, then establish the approved data boundary before introducing real client files. Ask which workspace processes the documents, who can access the output, what connected tools can receive information and how retention is configured. Record those answers separately from the model-quality scorecard.
A buyer should ask for evidence about the configured service rather than infer settings from a general privacy page. For broader workflow context, our practical guides explain the site’s approach to keeping sources and reviewer decisions visible. The next useful step for an interested IPO team is a focused demonstration using an agreed packet and explicit review criteria.