Should Freelancers Tell Clients They Use AI?

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Should Freelancers Tell Clients They Use AI?
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Should Freelancers Tell Clients They Use AI?

Yes, freelancers should tell clients when AI meaningfully shapes a deliverable, uses client information or changes an agreed way of working. Explain the intended use before starting, and ask for approval where permission is needed. For minor assistance, agree a simple standing policy instead of reporting every small edit.

This is a practical trust rule, not a claim that every task has the same legal disclosure requirement. Your contract, the client's instructions, the platform and the type of work may change the answer. If they conflict or leave a sensitive question unresolved, pause that use and get qualified advice.

Follow me on Instagram@sagnikteaches

Separate telling the client from asking for permission

Disclosure describes what you do. Permission authorises something the client needs to approve. They are related, but sending “I use AI” after uploading a confidential document has not given the client a useful choice about that upload.

Connect on LinkedInSagnik Bhattacharya

Put three questions beside the brief. What will AI do? What material will leave the client's existing systems? What will the client receive? Your answer may be different for brainstorming generic questions, rewriting an unpublished business plan and generating final illustrations.

Subscribe on YouTube@codingliquids

A statement such as “AI helps with my workflow” leaves all three questions open. A better explanation is: “I would like to use an approved writing assistant to suggest headings from your public brochure. I will write the final text and check all claims against your approved product sheet.”

For an existing client, raise the change before the next affected job. Do not assume that a friendly relationship answers a new data-sharing question. They may have promised their own customer that information will stay within a particular set of services.

The wider customer AI disclosure guide covers public-facing communication. A freelancer's first concern is more specific: whether the commissioned work matches what the buyer agreed to purchase.

Use the brief to choose the level of disclosure

Start with the signed agreement and the latest written instructions. Look for wording about AI, confidentiality, subcontractors, approved tools, originality, ownership and third-party processing. If “no AI” is written anywhere, clarify its scope before using an assistant. Do not quietly decide that your use is too small to count.

Proposed useSuggested conversationRecord to keep
Generic brainstorming without client materialAgree whether this fits the client's general AI policyStanding policy or project instructions
Editing client text in an external assistantExplain the tool and exactly what text would be sharedApproved tool, permitted material and purpose
AI-generated passages in the delivered copyExplain the contribution and your checking process before draftingAgreed production method and acceptance criteria
Generated images presented as illustrationsAgree their role and how the client may use themApproved approach and any presentation requirements
Client explicitly requests work without AIOffer a compliant workflow or decline that scopeClear written agreement on what is excluded

Do not use a percentage as your only description. “Ten per cent AI” could mean ten headings, an initial draft later rewritten or one critical calculation. Describe the actual step. Clients can make a better decision about “suggests alternative headings” than about an unexplained number.

For the low-risk end of that table, a standing policy saves you emailing about every small step. It goes in the proposal or onboarding pack once and covers every job unless a brief says otherwise. A filled-in version for an illustrative freelance translator:

On all projects I may use a spelling checker and a business-plan AI assistant for terminology lookups and alternative phrasings of individual sentences, without pasting in your documents. I translate and edit every sentence myself. I will ask before uploading any of your files to an AI service, before using machine translation for a first draft, and before producing any AI-generated images or audio. If you would prefer no AI assistance at all, tell me and I will note it on your account.

The policy works because it names three things a client can check: what happens routinely, what never happens without asking, and how to opt out. When a later brief arrives with a confidential contract, the translator doesn't need a new conversation about lookups, only about the upload.

Platform guidance needs checking too. Fiverr's current guidance permits responsible AI use, holds freelancers accountable for delivery and says they must respect a client's explicit request for work without AI. It encourages early discussion of preferences. That is not a universal instruction to label every tool action, nor does platform permission override your project agreement. See Fiverr's AI guidance for freelancers and clients.

A $900 writing commission, agreed before drafting

Consider an illustrative freelance writer commissioned by a packaging supplier to produce three product pages for a fixed $900 fee. The buyer provides public brochures and an unpublished spreadsheet containing supplier margins. The writer expects eight hours of work without AI and wants to try AI-assisted outlining.

The writer's first decision is about the input, not the wording of a disclaimer. The margin spreadsheet is unnecessary for outlining product pages. It stays out of the assistant. The writer asks the client to approve use of the public brochure and a short list of confirmed product facts.

A useful proposal message could read:

For these three pages, I propose using Claude to suggest outlines and alternative headings from your public brochure and approved product facts. I will choose the structure, write and edit the final copy, and verify each claim. I will keep your margin spreadsheet and customer records out of the tool. Please confirm whether this approach fits your policy, or whether you want the work completed without generative AI.

The client's illustrative reply is: “Outlines and headings are fine. Please do not use AI-generated product descriptions.” That is a limited approval. The writer records it against the job and keeps the assistant out of paragraph drafting. A general “AI allowed” tick would lose the detail that matters.

The completed job takes six hours, including a 20-minute AI preparation step and 40 minutes checking headings and source facts. The agreed fixed fee remains the starting point for the invoice; the parties have not agreed to change it. For future quotes, the writer can reassess the scope and price using actual time records.

If this were instead an hourly job at an illustrative $60 an hour, six actual billable hours would total $360. The eight-hour estimate would have been $480. Record actual eligible work under the agreed billing rules, including genuine preparation and checking, rather than billing the manual estimate as though it happened.

The handover note is short: “AI was used for the outline and heading options as agreed. I wrote the page copy and checked the specifications against the approved facts.” The writer keeps the client's approval, source sheet and drafts. There is no need to send a pile of prompts unless the agreement calls for them.

This example does not establish a universal pricing rule. It shows why the method, deliverable and charging basis should be clear before work begins. For the broader commercial question, see whether to charge less when AI speeds up work.

Three smaller jobs that need different conversations

A proofread that becomes a rewrite

In an illustrative assignment, a freelancer checks a laboratory's public service brochure. The agreed AI use is spelling and punctuation suggestions. The assistant proposes replacing “results are reviewed before release” with “every result is guaranteed accurate”. That is a new claim, not a correction.

The freelancer rejects it and records the error. If they want help rewriting the brochure, they ask the client to approve that wider role first. A proofreading agreement should not expand just because the software offers a more persuasive sentence. At delivery, a simple note that suggestions were reviewed against the original may be enough under the standing agreement.

A concept image that could look like evidence

An illustrative freelance designer creates a mood board for a home-care provider. The proposed image shows a fictional care worker in a fictional home. Before producing it, the designer explains that it is a concept illustration and asks whether generated imagery is acceptable for the intended use.

The client approves it for an internal discussion, then later asks to put it beside a testimonial. That is a new context. The designer should flag that the picture does not show the person or visit described and agree appropriate presentation or choose an authorised real image. The original approval for a mood board does not settle every later use.

A useful analysis that would expose unnecessary data

An illustrative freelance analyst helps an import-export business group 120 enquiry reasons. The source file includes names, contact details, supplier terms and free-text requests. Saying “I use AI for analysis” does not explain that an external service would receive those details.

The analyst proposes a smaller input: enquiry IDs and manually reviewed reason descriptions, with identifying and commercially sensitive details removed. The client approves that specific process. The analyst checks a sample against the originals locally and keeps the mapping in the client's approved system.

Removing names alone is not a reason to call a file anonymous. A rare order, reference number or quoted message can still identify a person or business. Where personal data or confidentiality obligations are involved, agree the process with the client's responsible adviser before using the service.

Write a useful disclosure in five lines

Allow about 15 to 30 minutes to write your first project-specific note, then ask someone to read it as a buyer. Can they tell what they are approving? Can they tell what still needs checking? You can do this in an ordinary document; no disclosure software is necessary.

Project: three product pages for a spare-parts manufacturer.
Proposed AI role: outline options and checks for missing source facts.
Permitted input: public catalogue and the approved specification sheet.
Human responsibility: final wording, source checks and delivery approval.
Boundary: no customer records, no invented compatibility claims, no final AI-written paragraphs.
Client decision: approved / approved with changes / work without AI.
Approval record: date, authorised approver and agreed changes.

This is an illustrative working record, not a ready-made legal clause. Adapt it to the job. If you need binding contract wording about ownership, confidentiality or liability, have a solicitor review it. The guide to AI clauses and approvals helps identify the issues to take to that discussion.

Avoid promises such as “Your data is completely private”, “All output is guaranteed original” or “Nothing is ever stored”. Those statements need evidence about the actual service, plan, settings and contract. A precise statement about the permitted input is more useful than a sweeping reassurance.

AI assistants tend to write exactly those reassurances when asked to draft the note for you. An illustrative request and reply:

Prompt: Write two sentences for a proposal telling a client
how I use AI on their website copy.

Illustrative output: "Your data is completely safe with me:
I use secure, enterprise-grade AI tools that never store your
information. All content is 100% original and human-quality."

Every clause needs fixing. “Completely safe” and “never store” are claims about a service you don't control; “enterprise-grade” is marketing; “100% original” is a promise nobody can prove. A version you can stand behind: “I use a paid AI writing assistant to suggest outlines from your public materials. I write and check the final copy myself, and I won't upload your private documents to it without asking first.” Shorter, and every word describes something you actually do.

For example, Claude Team does not use content for training by default. That fact alone does not answer how long information is retained, who can access it or whether your client permits the upload. Consumer Claude plans have a model-training opt-out; do not present that switch as equivalent to every business-plan control.

Keep the approved tool and account type in your private project record. If you switch to another service, add a connected tool or start using a new class of data, review the agreement again. A process that once used only public copy may become materially different when it gains access to the client's files.

Handle silence, objections and earlier undisclosed use

If permission is needed and the client does not reply, continue only the work that does not depend on that permission. Prepare the source list, conduct an agreed interview or draft manually. A missed reply is not a useful basis for deciding that a restricted upload is acceptable.

When the client objects, ask what outcome they are protecting. They may want confidential material kept out of external systems, wholly human-written final copy or a process their own customer has approved. Record the answer in task terms. “No uploaded interview notes” is clearer than “use AI carefully”.

Offer a revised method and explain any effect on scope, schedule or price before proceeding. Do not pressure the client to accept a tool because it suits your margins. Equally, you can decline work if the requested method is not one you are willing or able to deliver.

A quick sum makes the price effect easy to explain. Take an illustrative freelance video editor quoting $540 for a 12-minute interview film: nine hours at $60, with captions generated by an AI transcription tool and corrected by hand in about an hour. The client objects, because the interviews discuss an unannounced product and they don't want the audio uploaded anywhere. Captioning by hand adds roughly two and a half hours, so the revised quote becomes $690. The message says exactly that: “Without AI transcription, captions take about 2.5 extra hours, so the quote moves from $540 to $690. Everything else is unchanged.” The client can then choose on facts, and some will decide the upload is acceptable once they see the cost of avoiding it. Either answer goes in the approval record.

If you already used AI without a needed conversation, establish the facts first: what was submitted, to which service, under which account, and what reached the deliverable. Tell the client accurately and propose a correction. Do not delete records simply to make the issue harder to trace.

An illustrative correction is: “I used an assistant to reorganise the public brochure into an outline before we agreed the process. No private files were included. I can rebuild the outline manually and have paused further AI use while we agree your preference.” Use that wording only if every part is true.

If confidential or personal information was involved, involve the client's responsible contact promptly and get qualified advice about any further steps. A disclosure message alone does not resolve a data-handling problem. Keep your explanation factual and avoid unsupported assurances about what the service has deleted.

Check the final handover against what was agreed

Before delivery, compare the actual workflow with the approval record. Check the tool, inputs, AI contribution and human checks. If they differ, resolve that difference before presenting the work as compliant. Keep the record brief enough that you will use it on the next job.

  • The client had a clear choice before the affected work began.
  • The material shared matched the approved input.
  • The final output meets the agreed quality and evidence requirements.
  • Your invoice follows the agreed charging basis.
  • Your handover describes your contribution accurately.

A successful disclosure process produces fewer surprises. The client understands the method, you know the boundaries, and a later question can be answered from the record. That is a better test than whether a long disclaimer was included in every email.

Disclosure questions beyond the first agreement

What if a subcontractor uses AI?

Give the subcontractor the same agreed boundaries that apply to your work. Ask which tools they intend to use and what client material they would provide. Check whether your client agreement allows that arrangement before sharing files. Keep a record of their contribution so you can accurately explain the final process to the client.

What if the client sends me an AI-generated draft?

Ask what they expect you to do with it: edit wording, check facts or rebuild the work from sources. Their use of AI does not automatically approve your use of another service. Agree which source materials you need, what verification is included and whether your own AI assistance is permitted before quoting or starting.

Should my portfolio say that a piece involved AI?

Describe your contribution accurately and follow the client's publication and confidentiality agreement. If the example demonstrates your writing or design process, explain any AI contribution that would change a reader's understanding of your role. Ask permission before publishing client prompts, drafts or screenshots, because these may reveal material beyond the finished public work.

Further reads

Sources: Fiverr Help, Using AI on Fiverr: Guidelines for freelancers and clients; Upwork, How To Write a Freelance Contract and What To Include. Product privacy details from the vendors' help pages, checked 27 September 2026.

Need clear AI boundaries for your freelance work?

An AI implementation consultation can map where AI enters your process, what client material it needs and which approvals to record. On a 1:1 call, we can turn that into a practical working routine.

Book a 1:1 call with me