The tool that fits an insurance agency is the one that matches the systems the agency already uses, the data it is allowed to put there, and the workflows it wants to run. ChatGPT and Claude are general assistants: a person provides the document or the email, and a person at the agency reviews the draft before it goes to a client or a carrier. Microsoft 365 Copilot works inside apps such as Word, Excel, PowerPoint, Outlook, and Teams, on Microsoft 365 content the signed-in person can already open, so it fits an agency that already lives in Microsoft 365, and files outside Microsoft 365 are in reach only when a copy or an approved connector makes them available.
This page compares the three for renewal prep, carrier submissions and quote comparison, client email drafting, policy and document review, and benefits census cleanup. It then covers how each vendor describes business data, and what an agency should check before it rolls one out. Product statements below are limited to what OpenAI, Anthropic, and Microsoft publish on their own pages. Prices and plan details that those pages do not need for this comparison are left out.
ChatGPT vs Claude for an insurance agency
ChatGPT and Claude both draft from material a person puts in the conversation, and a person at the agency still reviews what goes to a client or a carrier. For an agency, the difference that matters is the account the work runs in and the data rules on that account. An agency that already keeps its email and documents in Microsoft 365 should read the Copilot section before it treats either assistant as the default.
On the work itself, the two assistants start the same way. Someone gathers the expiring policy, the quote pages, the submission, or the email thread, puts that material into the conversation, and asks for a draft in a format the agency already uses. A renewal summary, a side-by-side of limits and deductibles, or a client email can all be that draft. The producer or the account manager decides what is accurate and what gets sent. Neither product sits inside the agency management system on its own. Someone still brings the file to the conversation, unless the agency has separately connected another system and approved that connection.
The login is where an agency should separate them, and where it should separate both from a personal account. OpenAI says that on personal ChatGPT plans, including Free, Plus, and Pro, conversations may be used to train models unless the person turns that setting off. Anthropic says that on consumer Claude plans, including Free, Pro, and Max, the person chooses whether chats are used to improve Claude. An organization workspace is a different arrangement. On ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, and the OpenAI API, OpenAI says it does not use inputs and outputs to train its models by default. On commercial Claude products, including Claude for Work and the Anthropic API, Anthropic says it does not use inputs and outputs to train its models by default unless the customer opts in. Those are published defaults. They are not a substitute for the agreement, which is covered in client data, HIPAA, and PII.
OpenAI also says that, on its business products, workspace administrators control which connected apps are enabled, ChatGPT respects the organization's existing permissions, and each person signs in to a connected app before it is used. That is a setup choice, not something a personal login provides. Claude's commercial products are the account an agency would standardize on if Claude is the platform. When that is the choice, the rollout this site already describes is on Claude for insurance agencies. The same discovery can recommend Copilot or Hatz AI. Integrated AI is a Founding Claude Partner, and that partnership is not a claim that Claude wins this comparison for every agency.
Is Copilot good for insurance agencies?
Microsoft 365 Copilot fits an insurance agency that already does the work in Microsoft 365. It is a weak fit when the policies, quotes, and census files live somewhere else, unless those files are copied into Microsoft 365 or an approved connector makes them available. Copilot works inside Microsoft 365 apps and uses content the signed-in person can already open. It does not become a fit merely because the agency has heard the name.
On this page, Copilot means Microsoft 365 Copilot, the organization product Microsoft describes as working alongside Microsoft 365 apps such as Word, Excel, PowerPoint, Outlook, and Teams. Microsoft's service description also uses the name Microsoft Copilot for that product. This is not a personal chatbot login, and it is not a coding assistant sold under the Copilot name.
Microsoft's privacy documentation says Copilot provides answers anchored in organizational data the person can access through Microsoft Graph, including documents, emails, calendar, chats, meetings, and contacts. It also says Copilot does not access data that person is not allowed to open, and that prompts, responses, and data accessed through Microsoft Graph are not used to train foundation models. Microsoft's architecture documentation says the prompt can include text from input files or other content Copilot discovers, and that customer data stays within the Microsoft 365 service boundary, under the security, compliance, and privacy policies the organization already has. For a client email, that means Outlook can draw on mail and files that person can already open. For a renewal note or a quote grid, it means Word or Excel can work from a file that is already in that person's Microsoft 365 content.
Permissions are the practical limit. If a SharePoint library is open to the whole staff, Copilot can use that library for anyone who can already open it. An agency should confirm sharing before it turns Copilot on, because the tool follows the access people already have. A policy PDF that exists only in the agency management system, or a quote that exists only on a carrier portal, stays outside that set until someone places a copy where that person can open it in Microsoft 365, or until an approved connector brings that source in. Microsoft's service description lists Microsoft 365 Copilot connectors as part of the product. Which systems a connector reaches is something the agency confirms in its own tenant.
An agency that does not run on Microsoft 365 will not get that benefit, and ChatGPT or Claude can still draft from material a person provides. An agency that does run on Microsoft 365 may still use one of those assistants for files that never land in Microsoft 365. The choice is which system already holds the work, plus the data rules in the next section. How enablement differs from buying a tool and stopping there is in AI enablement or an AI tool.
How ChatGPT, Claude, and Copilot compare on agency work
On each workflow below, all three can help a person draft or summarize. They do not decide coverage, they do not bind, and they do not replace the person who sends the work. The practical difference is where the source file already lives, and which account is approved to see it.
Renewal prep
A renewal starts from the expiring policy, the loss runs, and the notes already on the account. In ChatGPT or Claude, a person puts that material into the conversation and asks for a draft in the shape the producer already reviews. In Microsoft 365 Copilot, the same kind of draft can start from a Word file, an email, or another Microsoft 365 file that person can already open. The producer still decides what to remarket and what to tell the insured. We do not publish a property and casualty hours figure for this work.
Carrier submissions and quote comparison
A submission summary or a quote comparison lines up what the pages already say, including premiums, limits, deductibles, and the differences a person would otherwise re-key. ChatGPT and Claude do that from the quote pages a person provides in the conversation. Copilot can draft from those pages in Word, or work from figures already in Excel, when the file is Microsoft 365 content the person can open or an input file included in the prompt. Recommending a coverage stays with a licensed person. Published times for benefits workflows, including carrier quote comparison, are in the five use cases every agency builds in week one. Those figures describe the workflow. They are not a score for ChatGPT, Claude, or Copilot, and they are not a property and casualty hours number.
Client email drafting
All three can draft a client email from material the tool can see. ChatGPT and Claude draft from the thread or the notes a person puts in the conversation. Copilot can draft in Outlook from email and other Microsoft 365 content that person is already allowed to open, which follows from Microsoft's description of Graph grounding. Nothing goes out until a person at the agency reviews it. That review is the same control described in the client-data guide.
Policy and document review
Policy checking, as this site uses the phrase, means comparing the policy with the binder or the quote. Any of the three can help a person read that difference when the documents are available to the tool. ChatGPT and Claude have the documents because a person provided them. Copilot has them when they are Microsoft 365 content that person can open, or input files in the prompt. The person still owns the exception. A wrong limit in a draft is still the agency's error if it is sent.
Benefits census cleanup
A census is a spreadsheet, and it often contains Social Security numbers, dates of birth, and other fields that do not belong in an unapproved account. Copilot in Excel can work on a workbook the person can open in Microsoft 365. ChatGPT or Claude can work from material a person provides. Neither path is appropriate until the agency has decided which fields may enter the tool. Census work already described on this site refuses Social Security numbers and dates of birth on input. A benefits brokerage that handles protected health information should read the client-data guide before it picks a login, and it should start from a written policy rather than from a tool trial.
How ChatGPT, Claude, and Copilot handle business data
OpenAI, Anthropic, and Microsoft each publish a default for whether business data is used to train models, and that default is not the same on a personal login as it is on an organization account. Those pages are a starting point. A statement on a webpage is not a contractual term and can change, so the agency confirms training, retention, access, and the agreements it needs in writing before client information goes in.
ChatGPT
On personal ChatGPT plans, including Free, Plus, and Pro, OpenAI says conversations may be used to train models unless the person turns the setting off. On ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, and the OpenAI API, OpenAI says it does not use inputs and outputs to train its models by default, unless the customer opts in. On those business products, OpenAI says administrators control which connected apps are enabled, the product respects existing permissions, and each person authenticates to a connected app. OpenAI also says it does not, by default, train on data accessed from those apps. A personal login is the pattern the client-data guide calls unsanctioned use: staff pasting client information into a consumer tool the agency has not approved.
Claude
On consumer Claude plans, including Free, Pro, and Max, Anthropic says the person chooses whether chats are used to improve Claude. On commercial products, including Claude for Work and the Anthropic API, Anthropic says it does not use inputs and outputs to train its models by default, unless the customer opts in, for example by sending product feedback. For a Claude for Work account, Anthropic says the customer organization controls the data its users submit and Anthropic processes that data to provide the service. Integrated AI's own engagements add a separate term this site already states: nothing an agency provides is used to train external or shared models. That term belongs to the engagement. It does not automatically transfer to a Claude account an agency buys on its own. Confirm the equivalent in the vendor agreement. When the platform is Claude, the account setup this site describes is on Claude for insurance agencies.
Microsoft 365 Copilot
Microsoft says prompts, responses, and data accessed through Microsoft Graph are not used to train foundation models, including the models Microsoft Copilot uses. Copilot uses content the signed-in person can already access, such as documents, email, calendar, chats, meetings, and contacts, and it does not open content that person cannot open. Microsoft's architecture documentation says customer data stays within the Microsoft 365 service boundary. The control the agency actually holds is permission. An over-shared site is still over-shared after Copilot is turned on, because Copilot can use what that person can already open.
None of these defaults makes a tool compliant with HIPAA by itself. HIPAA compliance depends on the agency's controls and agreements, which is the point of the client-data guide. Before protected health information goes into any of the three, confirm the current vendor agreement, including whether a business associate agreement is available for that product, with counsel or the agency's privacy officer.
The vendor pages behind the statements in this section are OpenAI's enterprise privacy page and its help center article on model improvement, Anthropic's privacy center articles on commercial model training and consumer privacy settings, and Microsoft's pages on Copilot data, privacy, and security, Copilot architecture, and the Microsoft Copilot service description. Read the current version before you rely on it.
What to check before an agency rolls one out
Four checks come before the first client file. Write the governance policy, settle client data and protected health information, keep licensed judgment with a person, and match the tool to the systems and the workflows the agency actually has.
- Sign a governance policy before anyone builds a workflow. The policy names the approved tools and the data that may enter them, so a deadline does not become a reason to open a personal account. Start from the free AI governance policy template, edit it for the agency, and have counsel review it before adopting it. The sections a policy needs are in what belongs in an agency's AI governance policy.
- Decide where client data and protected health information are allowed. Classify the fields that may enter the tool, and build hard stops for the fields that may not, including Social Security numbers and dates of birth on census work. In the vendor agreement, confirm whether inputs train the vendor's models, how long data is kept and whether it can be deleted, who can access it, and whether the vendor will sign the agreements the agency's compliance position requires. A benefits book that handles protected health information needs the HIPAA-aware version of that policy. The working version of these controls is in client data, HIPAA, and PII.
- Keep licensed judgment with a person. Recommending coverage and binding a policy stay with a licensed person at the agency. The tool can line up what the documents already say. It does not hold an insurance license, and it does not send the message to the client or the carrier. Integrated AI does not issue a credential or a certification, and it does not provide insurance continuing-education credit. How that credit works is on where AI stands with insurance CE credit.
- Match the tool to where the files live, then pick 5 to 10 workflows. If email, documents, and spreadsheets already live in Microsoft 365, Copilot can use the content people are allowed to open there. If the book lives in an agency management system and on carrier portals, ChatGPT or Claude still need a person to bring the file, unless an approved business connection already does that. Choose workflows whose inputs already exist and whose finished output already has a format the agency uses. The method is in how to pick the first workflows.
How Integrated AI helps an agency choose and implement a tool
Integrated AI helps an agency choose one platform and implement it on 5 to 10 workflows in the agency's own accounts. The platform recommendation is made in discovery, and it can be Claude, Copilot, or Hatz AI. Integrated AI is an implementation partner, not a training provider, and it does not sell courses or issue credentials or certifications.
The engagement is the Foundation. Discovery includes a data and exposure audit of every tool and account already in use, and the recommendation of one platform to consolidate on comes out of that audit. Integrated AI is a Founding Claude Partner. The same engagement builds in the agency's own environment when the recommendation is Copilot or Hatz AI, so the workflows stay in accounts the agency controls. How that work is described when the platform is Claude, including what the partnership does and does not change, is on Claude for insurance agencies.
The Foundation is $12,000 for the first 10 seats, plus $500 per additional seat, and it includes the first 60 days of the Partner retainer. After that the retainer is $1,500 a month for up to 10 people, $2,000 a month for 11 to 25 people, and $2,500 a month for 26 people and up. It is month to month with 30 days' notice. Pricing is valid for engagements signed through December 31, 2026. The agency buys its AI subscriptions directly and owns them. The governance policy inside that work can start from the AI governance policy template. Where a workflow touches client data or protected health information, the rules to confirm first are in client data, HIPAA, and PII.
The two days on site are working sessions on the agency's own accounts and systems. The agency keeps the policy, the shared library, and the workflows. The first conversation is a free 30-minute discovery call. Bring the systems the team already uses and the workflows that repeat. We will say which platform question to settle first, or we will say when this is not a fit.
Questions agencies ask
ChatGPT vs Claude for an insurance agency: which one fits?
ChatGPT and Claude are both general assistants. An agency uses either one by putting the renewal, the quotes, or the email into the conversation and asking for a draft a person will review. Neither assistant is the default for every agency. The fit depends on the agency's data-handling rules, on whether the account is an organization workspace or a personal login, and on the workflows the agency wants to run. Microsoft 365 Copilot is a separate choice when the work already lives in Microsoft 365.
Is Copilot good for insurance agencies?
Microsoft 365 Copilot fits an insurance agency that already works in Microsoft 365, because it works inside apps such as Word, Excel, PowerPoint, Outlook, and Teams and uses content the signed-in person can already open. It is a weak fit when the policies, quotes, and census files live only in an agency management system or a carrier portal, unless a copy is in Microsoft 365 or an approved connector has made that system available to Copilot. The agency still needs a governance policy, a rule for client data, and a licensed person reviewing the output.
Can an insurance agency put client data into ChatGPT, Claude, or Copilot?
Not into a personal login. OpenAI says personal ChatGPT plans, including Free, Plus, and Pro, may use conversations to train models unless the person turns that setting off. Anthropic lets a person on a consumer Claude plan, including Free, Pro, and Max, choose whether chats are used to improve Claude. On ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, and the OpenAI API, OpenAI says it does not use inputs and outputs to train its models by default. On commercial Claude products, including Claude for Work and the Anthropic API, Anthropic says it does not use inputs and outputs to train models by default unless the customer opts in. Microsoft says Copilot prompts, responses, and data accessed through Microsoft Graph are not used to train foundation models. Protected health information still needs a current vendor agreement that covers it. A statement on a vendor webpage is not the contract.
What should an agency check before rolling out ChatGPT, Claude, or Copilot?
Write and sign an AI governance policy before the first client file goes in. Confirm whether the account is an organization workspace whose vendor terms leave model training off by default, and get those terms in the agreement. Confirm whether protected health information is covered by the current agreement. Keep coverage advice, binding, and anything sent to a client or a carrier with a licensed person. Then match the tool to where the files already live, and to the 5 to 10 workflows the agency actually wants to run.
Does Microsoft 365 decide between ChatGPT, Claude, and Copilot?
It decides whether Microsoft 365 Copilot is a candidate. If the agency's email, documents, and spreadsheets already live in Microsoft 365, Copilot can work from content the signed-in person is allowed to open. If they do not, ChatGPT or Claude can still draft from documents a person provides in the conversation. The data-handling rules and the workflows still decide which assistant the agency standardizes on.
How does Integrated AI help an agency choose among these tools?
Integrated AI is an implementation partner for independent property and casualty agencies and employee-benefits brokerages. The Foundation recommends one platform to consolidate on, which can be Claude, Copilot, or Hatz AI, and then implements the agency's first 5 to 10 workflows in the agency's own accounts. Integrated AI is a Founding Claude Partner. It is not a training provider, and it does not sell courses or issue credentials or certifications. The Foundation is $12,000 for the first 10 seats, plus $500 per additional seat, and it includes the first 60 days of the Partner retainer. After that the retainer is $1,500 a month for up to 10 people, $2,000 a month for 11 to 25 people, and $2,500 a month for 26 people and up. It is month to month with 30 days' notice. Pricing is valid for engagements signed through December 31, 2026.
