In a recent Debevoise Data Blog post, we provided a quick guide on selecting the appropriate OpenAI models offered through GPT Enterprise for various legal tasks (e.g., o3 for research, GPT-4.5 for writing, and GPT-4o for image generation). In this follow-up post, we provide a broad overview of the various tools and features that are also available in ChatGPT Enterprise and how we have found them useful for our legal work. All prompts, files, and shared links would be stored according to your firm’s Enterprise retention policy.
| Feature | Description | |
|
File Upload |
What |
This is the way to get an AI model to review a document file (including PDFs, Word docs, Excel sheets, CSVs, PowerPoints, etc.). Attach a file and then use the chat window to instruct the model as to what task to complete involving the uploaded document. You can upload multiple documents, and you can ask for the output to include text, tables, charts, images, redlines, etc. Note that redlines will appear inline (e.g., with strike‑/underline markup) rather than as native track changes. |
|
How |
Click the + button in the chat window and then select which file or files to upload. |
|
|
Example |
Upload a 120-page district court decision along with the underlying briefs. Then submit a prompt: “Generate a table summarizing each of the issues discussed in the decision and the court’s ruling on each issue. Identify any issue discussed in the briefing papers that the court did not address in its decision.” |
|
|
Voice Mode |
What |
This is the way you can interact with ChatGPT Enterprise by voice instead of typing. The inputs and outputs are converted to text, so they are available as text in the chat history. |
|
How |
Tap the vertical lines inside a small circle to the right of the microphone icon at the bottom right of the chat window. You can select among several different voices for the models by clicking on the two horizontal lines with small circles at the top right of the screen when in voice mode. |
|
|
Example |
Anytime it is easier to talk and listen than type and read. This is especially useful on the ChatGPT Enterprise App when outside of the office (e.g., on your way to a meeting using your phone and headphones) or when you are preparing for an interview and want to role-play. |
|
|
Image Input |
What |
You can take a picture of something and have ChatGPT analyze it or convert it (such as whiteboard diagrams or handwritten text or a sign in a foreign language). |
|
How |
Click the + button in the chat window and then select an image to upload. On the mobile phone app, you can select camera and take a picture that will load into the Chat window. |
|
|
Example |
During a meeting, use a whiteboard to diagram a high-level plan; then take a picture and have ChatGPT turn that into a detailed action plan or timeline. Also, take a picture of handwritten notes in a foreign language and have ChatGPT identify the language, read the notes, and convert them into English-language text. |
|
|
Deep Research |
What |
Run complex queries that require multi-step research and analysis. The feature will ask for key clarifications that may be relevant to the query and then run iterative searches and analysis, eventually returning answers in the form of a detailed memo, with citations to support the results. |
|
How |
Within any chat, select Tools → Deep Research. |
|
|
Example |
Enter: “Collect all U.S. federal and state court opinions from the past two years that discuss generative AI copyright or authorship issues, extract each court’s key holdings and reasoning, and return a three-page briefing book, organized by jurisdiction, with hyperlinks to the full opinions.” |
|
|
Canvas |
What |
Draft content in a co-editing, collaborative workspace that allows you and ChatGPT to revise the same material at the same time—working line‑by‑line, highlighting text, requesting rewrites, and accepting/rejecting changes inline. Instead of copying and pasting individual outputs into traditional drafting software, Canvas allows you to refine a complete draft and then export the results. |
|
How |
In any chat, select the pencil icon with the small plus sign located below each output to open the Canvas workspace. Alternatively, drag and drop a document into ChatGPT and type “edit in Canvas.” |
|
|
Example |
Once in Canvas, highlight an overly wordy paragraph, click the Ask GPT button that appears, and type: “Maintain the ideas of this paragraph but condense it to around 150 words.” |
|
|
Search Chat History |
What |
You can search through past stored ChatGPT conversations, including inputs and outputs. Search covers chat titles and message text but not files; files are searchable only inside a Project. |
|
How |
Click on the magnifying glass icon (circle with small line at 5 o’clock) in the top left corner of the screen. |
|
|
Example |
Type “Bulk” to find every chat thread that included discussions related to the DOJ rule on transactions involving bulk sensitive personal data or government-related data, and read or resume the conversation. |
|
|
Persistent Memory |
What |
Persistent Memory lets ChatGPT remember details based on past usage—such as preferred formatting, communication style, analytical priorities, and role-based nuances—so it can reuse them in future conversations. These memories arise automatically but can also be triggered if you explicitly ask ChatGPT to “remember” something. While they can’t be edited directly, you should review them periodically to see if you agree with them, and, if not, delete them. |
|
How |
Click on the icon in the top right corner of the screen to pull down the account menu. Click on Settings → Personalization; toggle Reference Saved Memories to “on” and use Manage Memories to review or delete persistent memories. To add a new persistent memory, simply type “Please add this to your persistent memories,” in the chat window or keep giving the same instruction over time until ChatGPT infers the preference and saves it automatically. |
|
|
Example |
“Please add the following instructions to your persistent memories: always use Calibri 12‑pt bold font for headers in Word documents and use an Oxford comma in all your writing.” Every subsequent draft should arrive formatted to this specification. |
|
|
Custom Instructions |
What |
Similar to persistent memories, this allows you to create express instructions that become part of every prompt you submit to ChatGPT, but these are fully user generated and editable. |
|
How |
Click on the icon in the top right corner of the screen to pull down the account menu. Click on Settings → Personalization → Custom Instructions (or Customize ChatGPT for mobile). |
|
|
Example |
Enter: “I’m a senior attorney using ChatGPT for time-saving analysis. If you don’t know something, don’t guess. Just say you don’t know. Return answers in concise bullets and remain direct and to the point in all communications. When I share something with you that I drafted, I want you to tell me what’s wrong with it. Not just typos, but readability and persuasiveness too.” All subsequent chats will include these instructions automatically. |
|
|
Custom GPTs |
What |
Create specialized versions of ChatGPT by incorporating specific knowledge bases, tailored instructions, and customized response styles. These bespoke AI tools are optimized for targeted workflows and can be used by an individual user, shared with a select group of users, or made available Enterprise-wide. There is currently a limit of 20 files per Custom GPT with an Enterprise account. |
|
How |
Left sidebar → GPTs → +Create (upper right corner); manage in My GPTs |
|
|
Example |
Build an “Earnings Call Analysis GPT” that reviews earnings call transcripts and prepares a tabular summary of key topics discussed, analyst questions (with sentiment ranking), and potential deviations from prior public statements by the same people, especially on previous earnings calls. |
|
|
Projects |
What |
Organize and manage related chats, files, custom instructions, and memory—good for long‑running efforts under a unified umbrella. Chats created within the same Project become part of the reference data for new prompts, allowing information from one conversation to inform responses in another. There is a limit of 40 files per Project in Enterprise, and Projects are not currently shareable with other end-users. |
|
How |
Left sidebar → New Project (or the + icon on mobile), or move an existing chat via ••• (upper right corner of the screen) More → Add to Project. |
|
|
Example |
Create a “Cross‑Border M&A Transaction” project containing draft share‑purchase agreements, diligence memos, and a standing instruction to “summarize revisions in buyer‑ and seller‑focused bullet points.” |
|
|
Share Chats |
What |
Allows you to share an entire chat string with someone within your Enterprise account so they can continue the chat in their instance of ChatGPT, rather than just copying and pasting the last output. |
|
How |
At the end of a chat, click Share in the top right corner of the screen. |
|
|
Example |
After a long back and forth on an interesting research issue, share the whole string with a colleague so they can see what assumptions were made and what instructions were given (and how much back and forth it took) to reach the final output. |
To subscribe to the Data Blog, please click here.
The Debevoise STAAR (Suite of Tools for Assessing AI Risk) is a monthly subscription service that provides Debevoise clients with an online suite of tools to help them fast track their AI adoption. Please contact us at STAARinfo@debevoise.com for more information.
The cover art used in this blog post was generated by ChatGPT 4o, and content was partially generated by o3.
