Can freelancers use ChatGPT for client work? Who could see what you paste
The Morse teamUpdated 10 min read
Yes, freelancers can use ChatGPT for client work when the contract and the platform allow it, but a client's confidential material doesn't belong in a personal account that may train on it or keep it. As of November 2026, Upwork and Fiverr both allow AI tools. Fiverr forbids passing a client's data to third parties unless the client agrees, and Upwork requires reasonable steps to keep it confidential. The ChatGPT and Claude plans that don't train by default are sold only to two or more seats, so most solo freelancers work from one personal account with consumer terms.
On November 16, 2026, we read Upwork's User Agreement, Terms of Use, Optional Service Contract Terms, and help pages; Fiverr's AI guidelines and Community Standards; OpenAI's, Anthropic's, and Google's own plan, training, and retention pages; the American Bar Association's Formal Opinion 512 on generative AI; and Rule 34 of the Federal Rules of Civil Procedure. Fiverr deserves credit for spelling out what a client can expect when AI is used, and Upwork for not training its own AI on a job's messages unless both you and the client have opted in. This isn't legal advice.
Do Upwork and Fiverr let freelancers use AI?
Yes, with conditions set by the client. Fiverr's guidelines say it "permits the use of AI across all service categories," and that freelancers "must also honor a client's explicit request for non-AI work" (Fiverr). The same page tells clients to speak up early, because "freelancers do not necessarily disclose all the tools they use in their Gigs." When the client does ask, freelancers are expected to "disclose their intended workflow and align accordingly." If a dispute shows deception about how the work was made, "the order may be cancelled, the client may receive a full refund, and the freelancer's account may be permanently suspended."
Upwork's Terms of Use (version 4.12, effective September 10, 2025) list as misleading conduct "using generative AI or other tools to substantially bolster your job proposals or work product if such use is restricted by your client or violates any third-party's rights" (Upwork). On both platforms, then, AI is allowed until a client says otherwise, so read the job post and any message where the client sets terms.
What do Upwork's and Fiverr's confidentiality rules say about pasting client files into AI?
Fiverr's say no third party gets a client's data unless the client agrees, and Upwork's require reasonable steps to keep it confidential. An AI company is a third party. The AI rules on both platforms are about the work you deliver. The confidentiality rules sit elsewhere, and they decide what you may paste.
On Fiverr, the Community Standards say "freelancers are strictly prohibited from disclosing clients' data, including any personal data, to third parties (unless agreed otherwise with the client)," and that a breach "may lead to the permanent suspension of accounts involved." The AI section adds that freelancers must comply with each tool's terms of service and with "data protection, privacy, and consumer rights laws" (Fiverr). Fiverr's AI guidelines page doesn't mention what goes into the tool at all.
On Upwork, you and the client can agree your own confidentiality terms, and Upwork's help center says an outside NDA is "strictly between you and your clients" (Upwork). Without one, section 3.3 of the User Agreement (version 8.2, effective July 20, 2026) applies. It asks you to "take reasonable steps to keep that information confidential," and on the client's written request to "promptly destroy or return" it "and any copies of it contained in or on its premises, systems, or any other equipment otherwise under its control" (Upwork). The Optional Service Contract Terms (version 1.7, effective November 21, 2025) go further: no disclosure "to anyone without first obtaining the express written consent of the owner," and written certification that you've destroyed or returned it "within ten days" of a request.
A chat history that holds a client's brief is a copy on a system you control. Before you paste, ask whether you could find and delete it if the client asked tomorrow.
Does Upwork train AI on your client work?
Only if both you and the client are opted in. Section 2.4 of the User Agreement says that, unless you opt out, you grant Upwork a license to use "all User Content that you send or receive through the Upwork platform, including Work Product," to improve AI models that personalize your Upwork experience. Upwork's help page adds the condition that protects clients: "We require both you and the freelancer or client you are working with to be opted in. If either of you has opted out, none of the communication data or work product exchanged between you will be included" (Upwork).
Two limits apply. Training for fraud detection and trust and safety "is not subject to opt-out," and if you opt out later, content shared while you were opted in "may still be used." The same page says Upwork shares customer data with partners "including AWS and OpenAI" to run its AI features, under contracts that say the data "is never used to train their models." To change it, find the AI preferences section in your Upwork settings.
Can a solo freelancer buy ChatGPT Business, Claude Team, or Gemini for Workspace?
Only Gemini for Workspace sells to a single user. ChatGPT Business makes you pay for two seats, and Claude Team needs a second person. These business tiers are where the no-training defaults live, and two of the three are built for teams.
| Business plan | Can one freelancer buy it? | What the company's own page says | Source |
|---|---|---|---|
| ChatGPT Business | Only by paying for 2 seats, at $20 each a month billed annually or $25 monthly | A workspace "requires at least 2 Standard or Premium seats in total," and a seat is "paid capacity in a workspace, separate from the member who uses it" | OpenAI |
| Claude Team | No, not alone | "Team plans require a minimum of two members," and "each seat on a Team plan is assigned to one named individual, and only that person may use it" | Anthropic |
| Google Workspace Business Starter, with Gemini | Yes | Priced per user with a one-user option, at $7 a user a month on an annual plan |
So a freelancer working alone pays $40 a month, billed annually, for a ChatGPT Business workspace with one empty seat, and can't buy Claude Team at all without a second named person (Anthropic). Google's Workspace privacy hub, last updated October 6, 2026, says Gemini chats there "won't be reviewed by human reviewers or otherwise used to train generative AI models" without permission (Google). On the personal plans, ChatGPT and Claude train on your chats unless you opt out (Anthropic lists Claude's as "Opt-out"). As of November 2026, Google's help page says the settings that let it train generative AI on your Gemini chats (Chat content including text & uploads, and the audio and Live one) "are off by default" and need History on (Google). Google's page doesn't say whether accounts that had Keep Activity on before were switched over, so check Settings, then History. Our guide to private AI for startups sets out each plan's training and retention terms side by side, and is ChatGPT Business private covers what the business tier changes.
Who could see client material you paste into AI?
Four parties, and a different fix works for each. The facts in each row come from the companies' own pages and the rule's text, read November 16, 2026, with Google's Gemini pages re-read November 17, 2026.
| Who could see it | How, on a freelancer's usual setup | What helps |
|---|---|---|
| The AI company, through training | ChatGPT Free, Plus, and Pro "may use your content to train our models"; Claude trains unless you opt out; Gemini's training settings are "off by default," per Google | Turn training off in each app. Don't rate replies in client chats: OpenAI says a rated conversation "may be used to train our models" even after you opt out on a personal plan, though not on Business, and Anthropic keeps rated chats for up to 5 years, Team plans included, though a Team owner can switch off Rate chats for everyone. Morse doesn't train AI on your chats |
| The AI company's reviewers | Google says human reviewers, "including trained reviewers from our service providers," read some Gemini chats, and keeps reviewed ones for up to three years even after you delete your activity; Anthropic keeps flagged Claude chats for up to 2 years | In a Morse E2EE chat, only the verified enclave can read the message. In any other app, keep the client's names and figures out |
| A court, through a dispute | Rule 34 lets the other side request electronically stored information in your "possession, custody, or control"; OpenAI deletes a chat "within 30 days" unless it must keep it for "security or legal obligations" | Delete client chats when the job ends. Keep the client's identifying details out of any chat you wouldn't want produced |
| The platform | Upwork may train its own AI on job messages and work product while you and the client are both opted in; trust and safety training has no opt-out | Opt out in Upwork's AI preferences. Either party's opt-out keeps the job's exchange out |
Sources for the table: OpenAI, OpenAI, Anthropic, Anthropic, Google, Google, Cornell LII, and Upwork.
A freelancer with no IT department makes every one of these choices alone, and usually from one account that also holds personal chats and every other client's work. Google's Gemini privacy hub puts it in one line: "Please don't enter confidential information that you wouldn't want a reviewer to see."
Can a client dispute reach your AI chats?
It can, because a chat history is a record like any other. In a US civil case, Rule 34 reaches documents and "electronically stored information" in a party's "possession, custody, or control" (Cornell LII), and a chat in your own AI account fits that description. Anthropic says it may keep chats "as required by law, to resolve disputes" (Anthropic). A scope dispute over a missed deadline could put your prompts about that client in front of their lawyer. We cover the court orders and rulings so far in can AI chats be subpoenaed.
What should an AI clause in a freelance contract say?
It should name the tools and plans you'll use, and what client material may go into them. Lawyers have the most worked-out rule on this. The ABA's Formal Opinion 512, issued July 29, 2024, says a lawyer needs the client's informed consent before putting client information into a "self-learning" AI tool, and that "merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient" (ABA). The opinion doesn't bind freelancers, but its test makes a good checklist for any contract:
- Which tools, on which plan. "ChatGPT Plus with training off" tells a client more than "AI tools."
- What may go in. For example, published brand copy yes, unreleased financials no.
- No ratings in client chats, since a rated chat can be used for training on personal plans.
- Deletion when the job ends, matching the return-or-destroy clause your platform or NDA already has.
- Disclosure on request, the standard Fiverr already sets.
Translators and writers face versions of this with their own codes, covered in is it safe to use ChatGPT for translation and private AI for writers.
How Morse keeps a freelancer's client chats private
Every row in the table above starts with a readable copy of your work on a company's servers, kept under terms you didn't negotiate. Morse is built so your chat history is stored in your browser, encrypted, not on Morse's servers, and you pick who can read each chat. Your contract still decides what client material may go into any AI tool, Morse included.
- E2EE, for client material your contract or client has cleared for 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 ciphertext 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 guarantee 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 comes with Pro at $19 a month (how it works).
- Private, for drafts and questions with the client's details taken out. The default for a new chat. 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 (what that means).
- Anonymous, for general research with GPT, Claude, or Gemini. Our proxy removes your account, email, and IP address, so the provider sees the question without knowing it's you. It may keep it for a limited time to check for abuse, under its own policy, so leave client names and figures out.
For the training row, Morse doesn't train AI on your chats, sell your data, or show ads. For the reviewer row, Morse's servers don't keep messages or log message content. For the court row, because we don't store chats, we usually have very little to give authorities (privacy). A discovery request to you still reaches the chats in your own browser, so delete client chats when the job ends. When a client asks you to destroy their material, Morse's servers hold no copy of your chats, so delete them in your browser; in Anonymous mode, the provider may also keep the message for a limited time to check for abuse, under its own policy. Clearing the browser deletes the copies in it, and Morse can't recover them, so keep every deliverable in your own project files.
Which setup fits which job?
- Public material, such as a client's live website copy: any Morse mode.
- A client's confidential material: only where your contract or the client allows an AI tool, then a Morse chat in E2EE, pasting only the passages you need.
- Your own drafts and questions, with the client stripped out: a Morse chat in Private mode.
- Work where the client said no AI: no AI tool at all, as Upwork's and Fiverr's rules require.
Try a chat from the Morse home page without an account, or start free with no card needed.
Common questions
Do I have to tell a Fiverr client that I use AI?
Only if they ask, by Fiverr's own guidance. Fiverr says freelancers "do not necessarily disclose all the tools they use in their Gigs," but when a client asks for non-AI work before or at the start of an order, you're expected to disclose your workflow and honor the request.
Can I buy two Claude Team seats and use both myself?
No. Anthropic says each Team seat "is assigned to one named individual, and only that person may use it," and a Team plan needs at least two members.
Does turning off ChatGPT training cover the thumbs-up button?
No, on Free, Plus, and Pro. OpenAI says that if you rate a reply, the entire conversation may be used to train its models even after you've opted out. On ChatGPT Business, rating a reply doesn't permit training on the conversation.
What does Upwork mean by returning or destroying client information?
Under Upwork's default terms, on the client's written request you destroy or return their confidential information and any copies on systems under your control. If your contract uses Upwork's Optional Service Contract Terms, you also confirm it in writing within ten days of being asked.