MorseSign inStart freeOpen Morse

Private AI for marketers: can you use ChatGPT with customer data?

The Morse teamPublished 11 min read

Marketers can use ChatGPT with customer data only where their company has a contract making the AI company its processor or service provider, and as of November 2026 the personal ChatGPT Free, Plus, and Pro plans don't come with one. OpenAI offers a data processing addendum (DPA) only for ChatGPT Business, Enterprise, and its API, and it says it may train on chats from personal plans unless you opt out. A customer list, a CRM export, or a webinar sign-up sheet pasted into a personal account is a disclosure your privacy notice probably never mentioned. Under California's rules, it may count as a sale or a "sharing" of that data.

On November 17, 2026, we read Articles 5, 6, and 28 of the GDPR on EUR-Lex, section 1798.140 of the California Civil Code (in an Internet Archive copy from October 2, 2026, because the legislature's site blocked us), the California Privacy Protection Agency's CCPA regulations, three posts from the FTC's technology blog, and the data pages of OpenAI, Anthropic, Google, HubSpot, and Salesforce. HubSpot deserves credit for publishing exactly which AI settings start on and which start off, and Salesforce for stating that prompts sent through its Trust Layer aren't kept by the outside model companies. This isn't legal advice.

Can you paste a customer list into ChatGPT?

Not into a personal account, if the list holds real people. Your company is the controller of its customer data, and GDPR Article 28 says a controller "shall use only processors providing sufficient guarantees," with the processing "governed by a contract" that sets out the data, the purpose, and the controller's instructions (EUR-Lex). OpenAI's enterprise privacy page, updated January 8, 2026, says it can sign a DPA "for their use of ChatGPT Business, ChatGPT Enterprise, and the API." Nothing on that page covers a Plus subscription paid for on a marketing manager's card. We compare the AI companies' DPAs plan by plan in what is a data processing agreement.

The personal plans also learn from what you paste. OpenAI says that on Free, Plus, and Pro "we may use your content to train our models," until you turn off Improve the model for everyone under Settings, then Data controls. Even after you opt out, a thumbs up or down means "the entire conversation associated with that feedback may be used to train our models" (OpenAI). Its consumer data page puts it plainly: "Please do not enter sensitive information that you would not want reviewed or used" (OpenAI).

Does GDPR allow customer data in an AI tool?

Only with a lawful basis for that use. Article 6 says processing is lawful "only if and to the extent that" one of six grounds applies, such as consent "for one or more specific purposes" or the controller's legitimate interests, where those aren't "overridden by the interests or fundamental rights and freedoms of the data subject" (EUR-Lex). Article 5 adds purpose limitation: data "collected for specified, explicit and legitimate purposes" may not be "further processed in a manner that is incompatible with those purposes."

A newsletter list was collected to send a newsletter. Asking a chatbot to profile those subscribers is a different purpose, and Article 6(4) lists what decides whether it's compatible: the link between the two purposes, "the context in which the personal data have been collected," the nature of the data, "the possible consequences" for the people on the list, and safeguards, "which may include encryption or pseudonymisation." Your privacy notice also has to name "the recipients or categories of recipients" of the data, under Article 13. A chatbot you signed up for yourself is unlikely to be one of them.

Is pasting a CRM export into ChatGPT a sale under the CCPA?

It can be treated as one. California's rules turn on whether the company receiving the data has a written contract with yours. Civil Code section 1798.140 defines a service provider as a company processing personal information "pursuant to a written contract" that bars it from "selling or sharing" the data or using it "for any purpose other than for the business purposes specified in the contract." Anyone who isn't the business, a service provider, or a contractor is a "third party" (California Civil Code, archive copy).

The CPPA's regulations, effective January 1, 2026, spell out what follows. Section 7050(e) says a company without a contract meeting section 7051 "is not a service provider or a contractor," and that disclosing personal information to it "may be considered a sale or sharing of personal information for which the business must provide the consumer with the right to opt-out of sale/sharing." Section 7051 also requires the contract to name the purpose specifically, never "in generic terms" (CPPA). OpenAI calls Free, Plus, and Pro its "services for individuals" (OpenAI), and the terms you accepted when you signed up for one aren't a contract between OpenAI and your employer.

Marketing tools get a worked example in the same rules. Section 7050 says "an email marketing service provider can send emails on a business's behalf using the business's customer email list," but "cannot use the original email list to send marketing emails on behalf of another business." An email platform signs that kind of contract with your company, and a personal chatbot account comes with none.

Does hashing emails or swapping names for IDs make customer data anonymous?

No. The GDPR defines personal data as "any information relating to an identified or identifiable natural person," including one identified through "an identification number" or "an online identifier" (Article 4). Recital 26 says personal data "which have undergone pseudonymisation, which could be attributed to a natural person by the use of additional information should be considered to be information on an identifiable natural person" (EUR-Lex). A CRM export with customer IDs in place of names fits that sentence, because your CRM holds the key. California's definition of a "unique identifier" names a "customer number, unique pseudonym, or user alias" outright (California Civil Code, archive copy).

Hashed email lists, the format ad platforms ask for, are no exception. The FTC said on July 24, 2024 that hashes "can still be used to identify users" (FTC), and our guide to whether AI chats are anonymized covers the FTC's wider test. Most marketing questions can be answered from aggregated data instead: counts, rates, and averages by segment, with no row per person.

Is HubSpot's or Salesforce's built-in AI safer than ChatGPT?

On the contract question, usually yes, because the vendor already has a processor contract with your company. HubSpot's DPA is "incorporated into and forms part of" its Customer Terms of Service and covers HubSpot acting "as a Processor on your behalf" (HubSpot). HubSpot says it contractually requires its outside AI providers not to train their models on HubSpot customer data (HubSpot), and that it "enforces zero data retention with third party AI service providers wherever possible" (HubSpot).

HubSpot's own training is a different switch, and it starts on. HubSpot's settings guide says of AI Model Training: "By default, this setting will be toggled on." It covers "all eligible customer data" in the account, and the CRM data and Customer conversation data access switches for HubSpot's generative AI features also start on (HubSpot). Some of those models are trained "on data across many customer accounts," for email segmentation, spam detection, and deal-close predictions. Opting out "will only apply moving forward," because "it is not possible to delete previously used data from trained models" (HubSpot). A Super Admin turns it off from the settings icon, then AI under Account management, then the Access tab. Accounts with HubSpot's Sensitive Data setting turned on are opted out by default and can't opt in. Data enrichment is a separate setting, and turning off training doesn't turn it off.

Salesforce says its Trust Layer has "zero data retention" with outside model providers, so "prompts and generated responses are never stored or used to train the underlying third-party large language models," and that masking swaps personal details for tokens before a prompt leaves (Salesforce). Its Trailhead course adds two limits. Masking is on by default only "for certain sensitive data types," and "data masking for LLMs is disabled for agents" (Salesforce Trailhead).

Where does a customer list go in each AI tool?

We read each company's own pages on November 17, 2026. "Contract" means a processor or service provider agreement between the vendor and your company.

Tool and planTrains on what you paste by default?Contract with your company?How long a copy can stayCheck this
ChatGPT Free, Plus, ProYes, until you opt out (OpenAI)No; OpenAI's DPA covers Business, Enterprise, and the API (OpenAI)Cleared chats deleted within 30 days, unless de-identified or kept "for security or legal reasons" (OpenAI)A rated reply can train on the whole chat even after you opt out
ChatGPT Business, EnterpriseNo (OpenAI)Yes, a DPA on requestEnterprise admins set it; deleted chats gone within 30 days unless the law requires otherwiseBusiness admins can "view, access, export, and delete" your chats
Claude Free, Pro, MaxYes, if model improvement is on (Anthropic)No; Anthropic's DPA comes with its Commercial Terms (Anthropic)Up to 5 years with training on; flagged chats up to 2 yearsA thumbs up or down keeps the chat for 5 years
Claude Team, EnterpriseNo (Anthropic)Yes, accepted with the Commercial TermsFeedback kept up to 5 yearsOwners can turn off Rate chats for everyone
Gemini, personal accountDepending on your settings, while History is on (Google)None we found; Google says work accounts "may be subject to different data handling terms"18 months by default; around two months with History off; reviewed chats up to 3 yearsHuman reviewers read some chats
Gemini in Google WorkspaceNo, without permission (Google)Yes; prompts are customer data under Google's Cloud Data Processing AddendumUnder your Workspace agreementNot reviewed by humans without permission
HubSpot BreezeHubSpot's own models: yes, the switch starts on (HubSpot)Yes, HubSpot's DPA (HubSpot)Training can't be undone for data already usedAI Model Training in AI settings
Salesforce Einstein and AgentforceOutside model providers: no (Salesforce)Your Salesforce agreement; we didn't read its DPA todayZero retention at the outside model providers; masking logged in Data 360Masking is off for agents

Google's Gemini Apps Privacy Hub, last updated November 16, 2026, says temporary chats and chats with History off "are retained with your account for around two months."

Can an agency paste a client's customer data into AI?

Not without the client's written say-so. When an agency runs a client's campaigns, the client is usually the controller and the agency its processor, and Article 28(2) says a processor "shall not engage another processor without prior specific or general written authorisation of the controller" (EUR-Lex). An AI company is another processor. California says the same about service providers: one that engages someone else to help process the data "shall notify the business of that engagement," and the engagement must be "pursuant to a written contract" (California Civil Code, archive copy).

Most agency contracts and NDAs add their own limits on who may see a client's material. Our guide for freelancers doing client work covers what an AI clause in a client contract should say.

Is it safe to put an unreleased campaign into ChatGPT?

On a personal plan, a launch date or a rebrand goes to services that ask you not to send them anything confidential. Google's Gemini hub says "Please don't enter confidential information that you wouldn't want a reviewer to see," and keeps reviewed chats for up to three years even after you delete your history (Google). Our guide to private AI for startups compares consumer and business terms for confidential company documents, and on ChatGPT Business a workspace admin can still read every chat.

Does your privacy policy cover AI tools?

Read it before you paste, because the FTC holds companies to those promises. In a February 13, 2024 post, FTC staff wrote that "a business that collects user data based on one set of privacy commitments cannot then unilaterally renege on those commitments after collecting users' data," and that sharing data with third parties through "a surreptitious, retroactive amendment" may be unfair or deceptive (FTC). The post was aimed at AI companies, but its sentence about "a business that collects user data" describes yours too. A January 8, 2024 post noted that customers of AI companies "may reveal sensitive or confidential information," including "even their own users' data" (FTC). If your notice lists the kinds of vendors you share data with, a chatbot nobody contracted with isn't on that list.

How Morse keeps a marketer's drafts private

The personal plans above keep a readable copy of your work on an AI company's servers, under terms your company either never signed or signed for a different tool. On Morse, your chat history is stored in your browser, encrypted, not on Morse's servers. Which customer records may go into any AI tool, Morse included, is still your company's call, so customer names, emails, and IDs stay in the CRM. Take them out first, aggregate what's left, and pick the mode by what's still in the chat.

  • E2EE, for an unreleased campaign or client brief your contract allows in an AI tool. Your browser checks the secure enclave's hardware attestation, then encrypts the message so only that enclave can read it. Morse's servers relay text they can't read, and the reply is decrypted in your browser. The enclaves are run by third-party providers, not by Morse, and the protection rests on the enclave hardware and attestation working as designed. E2EE runs open-weight models like GLM, Kimi, and DeepSeek, with no web search or image generation, and it comes with Pro at $19 a month. Here's how end-to-end encrypted AI chat works.
  • Private, for copy, segment strategy, and numbers already aggregated. New chats start here. Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't (zero data retention, explained).
  • Anonymous, for asking GPT, Claude, or Gemini how legitimate interests or a CCPA opt-out works. Our proxy removes your account, email, and IP address and sends the question under Morse's own account, so the provider sees it without knowing it's you. The provider may keep the message for a limited time to check for abuse, under its own policy, so keep client names out of it.

Morse doesn't train AI on your chats, sell your data, or show ads. Because we don't store chats, we usually have very little to give authorities (privacy). Clearing your browser or losing the device deletes your chats, and Morse can't recover them, so keep finished copy and briefs in your team's own files.

What can go where?

  • Customer names, emails, phone numbers, and IDs: only into tools your company has a processor or service provider contract with, such as your CRM's built-in AI with its training switch checked.
  • Hashed or pseudonymized lists: treat them like the originals.
  • Aggregated results with no row per person: a Morse chat in Private.
  • Unreleased campaigns and client material your contract allows in an AI tool: a Morse chat in E2EE.

Try a chat from the Morse home page without an account, or start free and add E2EE with Pro when the launch plan needs it.

Common questions

Is a customer's work email address personal data?

Yes, under the GDPR, when it can identify one person, since the regulation's definition covers anyone identifiable through a name or an online identifier. The CCPA's definition of a unique identifier also covers a customer number, a unique pseudonym, and a user alias, so a B2B contact list is still a list of people.

Can I paste aggregated campaign results into ChatGPT?

Counts and rates with no row per person, such as open rates by segment or revenue by region, aren't records about anyone. Check that no segment is small enough to point to one customer, and that the numbers aren't confidential to a client.

Does turning off HubSpot's AI model training delete what it already learned?

No. HubSpot says it is 'not possible to delete previously used data from trained models,' and that an opt-out applies only from then on. Turning it off also doesn't turn off data enrichment, which HubSpot manages as a separate setting.

Is ChatGPT Business enough for a marketing team handling EU customer data?

It's the plan where OpenAI offers a DPA and doesn't train on your data by default, which covers the processor contract. Your company still needs a lawful basis for the use and has to name the AI tool, or its category, among its recipients, which a plan can't do for you.

Ask what you actually want to ask.

80+ models, characters, and images, with every chat encrypted and kept on your device. Free to start.

Start a private chat

Read next