How to fill an ACORD 125 with ChatGPT, step by step
- Get the free ACORD 125 skill. Download the free ACORD 125 recipe from this page and add it to your skills in the ChatGPT desktop app. The blank ACORD 125 travels inside the skill, so there is no fillable PDF to track down first.
- Open a new chat in work mode. Start a new chat in the ChatGPT desktop app and click the work mode icon at the top. This walkthrough uses the desktop app, not the browser version.
- Pull up your skills with a slash. Type a slash to open the skills menu, then start typing the name. Typing fill surfaces the fill ACORD 125 recipe. Pick it.
- Tell ChatGPT where the client information is. In the demo the Gmail connector is turned on, and the request is simply to find the email thread with the client, a coffee company owner named Marcus, and fill out the application from it. Any source works the same way, as long as ChatGPT can read it.
- Let it read the thread and map the fields. ChatGPT confirms it is running the ACORD 125 workflow, finds the thread, pulls the legal name, address, FEIN, square footage, revenue, and certificate holder details, and maps them onto the form. In the demo this ran from 7:48 PM to 7:50 PM.
- Read the missing-data list, then finish the form. It reports what the thread never established, in this case the carrier and NAIC code, the producer mailing address and phone, premiums, and the SIC code. Fill those from your own records.
- Make corrections by asking. Ask for changes in plain language, such as setting the carrier to AmTrust, adding a policy number, or changing annual revenue from 1.1 million to 2.2 million. The demo edit took about one minute, then the refresh button at the top shows the updated PDF.
- Save the PDF and review it. Use open in and download a copy, or open it in your own PDF app to finish the remaining fields. Check every figure against the client file before it goes to a carrier.
What a skill is, and why ChatGPT and Claude both have them
A skill is a recipe. It is a set of written instructions for one repeatable job, saved in a single file, so the same task runs the same way every time instead of depending on how well you phrase the request that day. Ben calls them recipes because that is what they behave like: the steps are already written down, you just say what you want cooked.
The ACORD 125 recipe is a good example of why the format matters. It carries the blank ACORD 125 PDF, a script that writes values into the form fields, and an instruction file that tells the assistant which piece of client information belongs in which box. None of that has to live in your head or in a long prompt, because it lives in the file.
Both ChatGPT and Claude now support this. The mechanics differ slightly, but the idea is identical, and one recipe file is a portable way to hand a coworker a process. That is the real reason to learn skills rather than collect prompts: a prompt is advice, a skill is a procedure.
| ChatGPT | Claude | |
|---|---|---|
| What it is called | Skill, sometimes called a recipe | Skill, sometimes called a recipe |
| Where you use it | The ChatGPT desktop app, in work mode | The Claude desktop app |
| How you start it | Type a slash, then the skill name, as shown in this video | Describe the task in plain English and Claude picks the skill from context |
| What is inside the file | Written instructions, plus any scripts and blank forms the job needs | Written instructions, plus any scripts and blank forms the job needs |
| Where the client data comes from | Connectors such as Gmail, or files you upload | Connectors, files, or a browser session |
| Cost of this ACORD 125 recipe | Free from this page | Free, covered in the Claude version of this tutorial |
How ChatGPT knows what to put on the ACORD 125
The recipe knows the form. It does not know your client. That gap is filled by pointing ChatGPT at wherever the real information already sits, and in this walkthrough that is Gmail, connected so ChatGPT can read the thread.
The sample thread is an ordinary agency email. The client needs a certificate before the first, and the reply asks for the legal name, the FEIN, and whatever else belongs on the certificate. What comes back is a normal email with the building details, the square footage, how much of it is open to the public, the annual revenue, and the additional certificate holder.
So the request is short: find the thread with this client and fill out this application. ChatGPT restates the job, searches Gmail, pulls the details out of the thread, and starts mapping them onto the form. The email was never written to be a data source, which is the point. The information agents need is usually already sitting in a thread nobody wants to retype.
What it filled, and what it deliberately left blank
The finished PDF comes back with the agency block filled in, the contact name and email, the quote box marked, the date, and the lines of business, in this case general liability and property. The next page carries the client details in the right fields, including annual revenue of 1.1 million, the description of primary operations, and the date the business started.
The more important part is what is missing. ChatGPT reports that it could not establish the insurance carrier and NAIC code, the producer mailing address and phone number, the premiums, or the SIC code, and it leaves those boxes empty. It states plainly that it is preserving the fields the thread does not establish, which is a polite way of saying it will not guess.
That is the behavior you want on an insurance form. A blank field is a task. A confidently wrong FEIN or limit is an errors and omissions problem that nobody catches until it matters. You get most of the document done and a short list of what is left, which is a much better trade than a form that looks complete and is not.
Editing the form by talking to it
Applications change after the first draft. The carrier gets decided, a policy number arrives, the revenue figure was a placeholder. Normally that means opening the PDF in another app and typing.
Here the edit is just a sentence: set the carrier to AmTrust, add the policy number, change the agency name, change annual revenue to 2.2 million. In the video that request is dictated rather than typed, using a dictation app that turns speech into text, which is worth knowing because talking through a change is faster than clicking into form fields.
The revision ran from 7:52 PM to 7:53 PM. A refresh button at the top of the document reloads it, and the updated carrier, policy number, and revenue are all in place. At that point you download a copy, or open it in your own PDF app and finish the fields ChatGPT flagged.
Where this fits in an agency day
One ACORD 125 is not the win. The win is that the shape of the work changes. Reading a client email and retyping it into a form is transcription, and transcription is exactly the kind of job to hand to software. Reviewing a mostly finished application against the client file is judgment, and judgment is what an agent is actually for.
It also runs on information you already have. There is no new system to populate and no data entry project first. The email thread, the declarations page, and the spreadsheet you already keep are the inputs.
You are still the one submitting the application. Check the carrier names, the FEIN, the limits, the revenue, and the dates against the source before anything leaves your office.