Consulting and recruiting

Survey platform terms: can the platform train AI on your studies?

By SourceX Editorial · Reviewed by Noah Loul · October 9, 2026

Short answer

Survey platform terms decide whether the vendor can use your questionnaires, responses and reports to train its own AI. Read five clauses together: the customer data definition, ownership, the license you grant the vendor, rights to aggregated or de-identified data, and any explicit AI-training statement. If the terms say nothing about AI training, ask in writing.

Key takeaways

The binding answer sits across the terms of service, the data processing agreement and any AI feature terms, not in a help-center FAQ.
A broad license to use customer data to improve the service can be read to include model training unless the terms say otherwise.
Rights to aggregated or de-identified data often survive termination, so check what the vendor keeps after you leave.
Published vendor commitments range from a flat no-training statement to authorization first to an admin toggle.
Your client contracts may forbid vendor AI training even when the platform terms allow it.

Where survey platform terms address AI training#

Survey platform terms address AI training across several documents at once: the terms of service, the data processing agreement, the privacy policy, product-specific or AI feature terms, and sometimes a trust center FAQ. The contract documents bind; the FAQ explains.

Which version governs depends on how you bought. A self-serve plan usually runs on online terms the vendor can update, while an enterprise order form may incorporate a negotiated agreement that overrides them. Save a dated copy of whatever applied when each major study was fielded, because the answer can differ from year to year.

The clause checklist#

The clause checklist below covers the terms that decide whether a platform can learn from your studies. Read the clauses together, because a narrow AI clause can be undone by a broad license elsewhere in the same agreement.

The clause checklist
Clause	What to look for	Reassuring wording	Wording to question
Customer data definition	Whether questionnaires, responses, uploads and reports are all included	Broad definition covering everything you and respondents submit	Narrow definition that leaves survey designs or metadata outside
Ownership	Who owns customer data	Customer retains all rights and the vendor acquires none	Silence, or shared ownership of outputs
License to the vendor	What the vendor may do with customer data	Use only to provide and support the service for you	Use to develop, improve or create new products and services
Aggregated or de-identified data	Rights to pooled or anonymized data	Limited to usage statistics under a defined de-identification standard	Undefined anonymization covering response content
AI and machine learning	An explicit training statement	No training on customer content without opt-in	Training by default, or an opt-out buried in settings
AI subprocessors	Third-party model providers	Providers barred from training on your data	No list, or providers acting on their own terms
Deletion and export	What happens after termination	Full export, then deletion within a stated period, with certification	Retention of de-identified copies without limit
Changes to terms	How updates take effect	Notice and a right to object to material changes	Changes effective on posting

How software vendors word AI training commitments#

Software vendors word AI training commitments in three broad ways, and published examples help you recognize each pattern in your survey platform's terms. None of the examples below is a survey platform; they show the range of language you are likely to meet.

Note where each commitment lives. A statement on an FAQ or principles page is useful evidence of intent, but if the contract grants a broader license, the contract usually controls. Ask the vendor to confirm its position in the order form or DPA.

Flat exclusion: Zoom's terms state that it does not use communications-like customer content, including poll results, to train its own or third-party AI models.
Authorization first: Box states that it does not train AI models on customer data and that any future custom-trained model would need the customer's explicit approval.
Admin control: HubSpot's knowledge base describes an AI model training switch customers use to control whether their account data trains HubSpot's models.

What aggregated and anonymized data rights really allow#

Aggregated and anonymized data rights let a vendor use pooled data beyond your account, and they are where AI training most often hides. Many SaaS agreements reserve the right to use de-identified data for benchmarking, analytics and product improvement, and that right often survives termination.

For a survey platform, the question is what gets pooled. Usage statistics, such as how many surveys you launched, are low risk. Response-level content, open-ended text and questionnaire designs are different: once de-identified to the vendor's own standard, they could in principle feed a model trained across all customers.

Ask three questions in writing: does the aggregated data include response content or questionnaire text, what de-identification standard applies, and does product improvement include training machine learning models.

Can the platform change its terms after you sign?#

A platform can usually change online terms by posting a new version, which is why the change clause matters as much as the AI clause. Negotiated agreements more often require notice, or a signed amendment, before changes apply.

Regulators have noticed the pattern. In a February 2024 post, FTC staff said that a business which expands how it uses consumer data, for example to train AI, and tells people only through a hidden, retroactive edit to its terms or privacy policy may be acting unfairly or deceptively. That guidance speaks to consumer data, but it is a useful reference point when you ask a B2B vendor how future AI changes will be notified.

Compare the terms at each renewal and keep a dated copy each time. A side-by-side comparison of two versions is often the fastest way to spot a new AI clause.

Steps to take before you sign or renew#

The steps below work best as an annual routine owned by one person, usually operations or the compliance lead, rather than a one-off scramble after a client asks.

Pull the current terms, DPA, privacy policy and any AI feature terms, and save dated copies.
List every survey and qualitative tool in use, including free tools staff signed up for on their own.
Check admin settings for AI features and training toggles, and record how each one is set.
Send the vendor written questions on training, aggregated data, subprocessors and post-termination retention.
Compare the answers with client contracts that restrict AI use and with your respondent privacy notice.
Negotiate an addendum where needed, or move sensitive studies to a tool whose terms fit.
File the decision with the study records so a later audit or licensing review can see what applied.

Illustrative: an agency audits its three survey tools#

Illustrative: a fictional research agency with about 90 employees runs client trackers on an enterprise survey platform, holds video interviews on a qualitative platform, and finds that project managers also use a free form builder for quick pulse surveys.

The enterprise platform's negotiated agreement bars training on customer content, so it stays. The qualitative platform's online terms grant a license to improve its services, and its AI summary feature is on by default; the agency switches the feature off for client studies and asks for written no-training confirmation. The free tool's terms reserve broad rights over de-identified responses, so the agency retires it for client work.

The audit also answers a client's supplier questionnaire, which asked exactly which tools had touched its respondents' data.

Why platform terms matter if you license your own data later#

Platform terms shape what you can license later, because a buyer will ask who else already holds rights to the same records. In the Rights step of the SourceX five-step transaction, SourceX reviews platform and vendor terms alongside client contracts and respondent consent, and the result is recorded in the SourceX Evidence Packet under provenance and licensing rights.

If a vendor already holds a broad license to your response data, an exclusive license to an AI developer may not be possible, and a non-exclusive one needs that history disclosed. Data is licensed, not sold outright, so the agency keeps ownership either way.

Frequently asked questions

Does paying for a higher plan change the AI terms?

Sometimes. Enterprise agreements can override online terms, and some vendors tie AI training commitments to certain plans or contract types. Check which document your order form incorporates and whether an AI addendum applies to your plan, rather than assuming the most protective version on the vendor's site covers you.

Is a platform's AI summary feature the same as training on our data?

Not necessarily. A feature can send your data to a model to produce an answer without that data being used to train the model. The terms should say both whether the feature processes your data and whether the vendor or its model provider may retain it or train on it. Ask about each separately.

Do respondents need to be told which survey platform we use?

Privacy notices commonly describe categories of service providers rather than naming each tool, though rules vary by jurisdiction and by panel terms. What matters most is that the platform's use of the data stays within what the notice describes. A platform that trains its own products on responses may fall outside that description.

Can we export our data and use it for our own AI work?

Usually you can export your own account data, subject to the export tools your plan offers. Exporting does not change the respondent consent or client contract limits on how you use the data, so the same review applies before any internal model training or outside license.

What should we do if we find a problematic clause mid-contract?

Raise it with the vendor in writing, turn off any AI settings you control, and move the most sensitive studies to a tool with better terms. Document what applied and when. If client studies were affected, check your client contracts for notice obligations and involve counsel.

Sources

Zoom's Terms of Service state that Zoom does not use audio, video, chat, screen sharing, attachments or other communications-like Customer Content, such as poll results, whiteboard and reactions, to train Zoom or third-party AI models. Source
Box's AI FAQ states that Box does not train AI models on customer data and that any future custom-trained model would require the customer's explicit approval. Source
HubSpot's knowledge base says customers can use an AI model training switch to control whether HubSpot uses their account's customer data to train its AI models. Source
FTC staff warned on February 13, 2024 that adopting more permissive data practices, such as using consumer data for AI training, through a surreptitious, retroactive change to terms or a privacy policy may be unfair or deceptive. Source

Related resources

See if your company qualifies

A short company assessment. No data uploads are needed.

List your survey tools and records Talk to SourceX