25 Questions to Ask a Data Buyer Before You Sign

How to use the list

Ask early, ask in writing, ask everyone

The best time for these questions is after an NDA and before you send anything beyond a manifest and a few samples. Three habits make the answers useful.

Ask in writing

A conversation gives you impressions. An email gives you a record you can hold up against the draft agreement. If an answer matters to you, it belongs in the contract text.

Ask every buyer the same

Identical questions turn different sales pitches into comparable answers. Practitioners advise getting more than one offer; see getting more than one offer.

Bring a lawyer in early

The questions help you spot gaps. Your own lawyer reads the agreement against your client contracts and staff notices. General information, not legal advice.

Published payout figures from buyer programs are tiers and ranges. micro1, for example, publishes "$100k+" for qualified partnerships, and Mode publishes "$100K-$5M" (both as published, checked 7 October 2026). Those numbers tell you a market exists. They do not tell you what your agreement will say about exclusivity, acceptance or liability, which is where two offers with the same price can be worth very different amounts. The eligibility checker shows which programs you may fit; this list is for the conversation after that.

25 questions to ask a data buyer before you sign

Sell Data to AI, selldatatoai.com/questions-to-ask-a-data-buyer/. General information, not legal advice. Buyer: ____________________ Date: __________

The checklist

The 25 questions

Each question has a short note on why it matters, plus what a clear and a weak answer tend to sound like. When printed, each question gets a line for the buyer's answer.

Money Scope Privacy Liability Exit

Money Questions 1 to 5: what you are paid, when and on what condition

What is the price, and is it fixed, per unit, or a range that depends on your review?

Published figures are tiers, not offers. Ask what would move the number up or down, and get the final figure stated in the agreement.

Clear answer: A stated figure or formula, the factors that could still change it, and a date by which the number becomes final. Weak answer: “We will know once we have seen everything.”

Is payment one-off or recurring, and is it split into milestones?

Some arrangements pay once for a delivered dataset; others involve ongoing participation. Know which one you are agreeing to.

Clear answer: A payment schedule with amounts tied to named events such as signing, delivery and acceptance. Weak answer: “Payment follows our usual process.”

What are the acceptance criteria that trigger payment, and who decides?

The most overlooked question. If payment follows acceptance, vague criteria can delay or reduce what you receive.

Clear answer: Written checks, a review period counted in days, and a fix-and-resubmit route if a batch fails. Weak answer: “We accept data that meets our quality bar.”

How long after acceptance is each payment due, and how is it paid?

Practitioners cite 60 to 90 days just to close. Ask for payment dates in the contract, not in an email.

Clear answer: Payment terms in days from acceptance, plus the currency and the payment method. Weak answer: “Finance pays on its normal cycle.”

Can any amount be reduced, withheld or reclaimed after payment, and in which cases?

Look for set-off, refund and holdback language. You want the cases listed, not left to discretion.

Clear answer: A short list of specific cases, such as a proven breach, with nothing reclaimable purely at discretion. Weak answer: “We reserve the right to adjust amounts.”

Scope Questions 6 to 10: what exactly is licensed, and for what

Exactly which systems, date ranges and record types are in scope?

Ask for the scope to be attached as a manifest. "Your Slack" is not a scope; "these channels, these years, excluding DMs" is.

Clear answer: A manifest naming systems, channels, date ranges and exclusions, attached to the agreement as a schedule. Weak answer: “Send what you have and we will pick.”

Is the license for training only, evaluation only, or both?

Evaluations built on data are valued far above raw data, practitioners say. Know which use you are pricing.

Clear answer: Named uses, such as training and evaluation, with anything beyond them needing your written consent. Weak answer: “Any AI-related purpose.”

Who downstream can receive or use the data, and can they pass it on?

Many data companies serve several AI labs. Ask whether your data goes to one named customer, several, or anyone.

Clear answer: Categories of permitted recipients, no onward resale unless agreed, and the same restrictions passed down to them. Weak answer: “Our customers are confidential, so we cannot say.”

Is the license exclusive, time-limited or non-exclusive?

Exclusivity decides whether you can ever sell the same records again. See exclusivity and resale rights.

Clear answer: Non-exclusive, or exclusive only for a stated period and field, with the price reflecting that. Weak answer: “Standard exclusivity applies.”

Is this a one-time delivery, or do you expect ongoing work from our staff?

Feedback on AI outputs or recurring exports takes employee time. Ask how much, from whom, and whether it is paid separately.

Clear answer: An estimate of hours and roles needed, for how long, and how that time is paid. Weak answer: “Just a little help from your team now and then.”

Privacy Questions 11 to 15: people in the data, and how they are protected

Who de-identifies the data, by what method, and can we review samples first?

Redaction and consistent pseudonyms leave different residual risk. See de-identification before selling data.

Clear answer: The method named, who runs it, and representative samples for you to review before any use. Weak answer: “Our pipeline handles privacy automatically.”

Can we set an exclusions list that is removed before anything leaves our systems?

Customer details, client-confidential matters, health and financial records, HR files and legal advice are the usual candidates.

Clear answer: Yes: your exclusions list becomes a schedule, applied before export, and the buyer confirms nothing excluded was received. Weak answer: “We will filter sensitive material later.”

Where is the data stored during processing, and who can access it?

Ask about the number of people with access, devices, separation from other clients, and retention periods.

Clear answer: Named storage locations, access limited to named roles, separation from other clients, and a retention period. Weak answer: “It is all secure.”

Will we get audit rights or written evidence over de-identification?

If you are asked to warrant that the data is safe to share, you need a way to check what was done to it.

Clear answer: Evidence on request, such as a de-identification report or a sample review, within a stated number of days. Weak answer: “You will have to trust our process.”

What must we tell employees, and what consent statements will you ask us to make?

Staff messages and email are in most workflow data. See employees and selling company data.

Clear answer: A short list of notices or consents you must confirm, worded so you can actually confirm them. Weak answer: “You warrant that every consent was obtained.”

Liability Questions 16 to 20: who pays if something goes wrong

What warranties will we give about ownership, consent and third-party rights?

Every promise you make is a possible claim later. Ask for the list before you agree a price.

Clear answer: Warranties limited to what you can reasonably know, qualified by your knowledge where that is fair. Weak answer: “Full warranties on everything in the dataset.”

If de-identification misses something, who pays, especially if the buyer did the work?

The party that controls the process is usually best placed to carry its risk. Ask how the draft allocates it.

Clear answer: The party that runs de-identification carries the risk of its own errors. Weak answer: “The seller is responsible for all content, whoever processed it.”

Is our total liability capped, and how does the cap compare with the price?

An uncapped indemnity can outweigh any payment. See indemnities and warranties in data deals.

Clear answer: A cap tied to the price paid, with any exceptions named and kept narrow. Weak answer: “Liability for data issues is unlimited.”

How long do warranties and indemnities survive after the agreement ends?

A survival clause can keep your exposure open for years after the money arrives.

Clear answer: A fixed survival period, in months or years, for each obligation that continues. Weak answer: “Obligations survive indefinitely.”

How does this fit our duties to our own clients, and will you accept carve-outs?

Law, accounting, M&A, agency and healthcare firms hold client secrets that may never be licensable.

Clear answer: Written carve-outs for client-confidential material, and no requirement to include it to earn the price. Weak answer: “Your client contracts are your problem.”

Exit Questions 21 to 25: how it ends, and what remains afterward

When are our originals and your copies deleted, and will you confirm it in writing?

Ask for a deletion schedule and a written certificate, covering both raw exports and processed copies.

Clear answer: A deletion date for originals and copies, with a written certificate when it is done. Weak answer: “We delete data when it is no longer needed.”

Does deletion cover derived datasets, and what happens to models already trained?

Deleting files is different from undoing training. Ask what is realistic, and get it described plainly.

Clear answer: A plain description of what deletion covers, including derived datasets, and what it cannot undo. Weak answer: “Deletion is handled according to policy.”

Can either side terminate, on what notice, and which clauses survive?

Termination rights matter if the buyer changes direction or a client objects after signing.

Clear answer: Termination rights for both sides on stated notice, plus a list of the clauses that survive. Weak answer: “The agreement runs until we end it.”

Can you assign the agreement or the data to another company, and will we be told?

If the buyer is acquired or sells a business line, your license may move with it.

Clear answer: No assignment without notice to you, and the same terms binding any successor. Weak answer: “We may assign the agreement freely.”

Which law governs the agreement, and where are disputes resolved?

A foreign forum can make any dispute too costly to pursue. Ask your lawyer what this choice means for you.

Clear answer: A governing law and a forum that you and your lawyer can realistically work with. Weak answer: “Whatever our standard paper says.”

General information, not legal advice. Talk to your own lawyer before you sign. For the clause-by-clause outline behind these questions, read the AI data licensing agreement guide.

Reading the answers

Vague answers and clear answers

These are general patterns, not descriptions of any named buyer. A clear answer is specific, written down and ends up in the agreement.

Topic	A vague answer	A clear answer
Scope	"We will work out what we need once we see it."	A manifest of systems, dates and exclusions, attached to the agreement.
Acceptance	"Payment follows once we are happy with quality."	Written criteria, a review period in days, and what happens if a batch fails.
Downstream use	"Our customers are confidential."	The permitted uses and categories of recipient, even if names stay under NDA.
De-identification	"We handle all of that."	The method, sample review before use, and evidence you can check.
Deletion	"We delete data when it is no longer needed."	A retention period, a deletion date and a written confirmation.

Before the first question

What to have ready on your side

The questions work best when you already know what you are offering and what you will never include.

Your documents

A manifest: systems, years, rough volumes and record types, with no content.
A few samples you are comfortable showing under NDA.
An exclusions list: what never leaves the building.
A list of client contracts with confidentiality or data clauses.

Your decisions

Who signs internally, and who must approve first.
Your walk-away terms: for example, no perpetual exclusivity or uncapped liability.
How staff will be told, and when.
Which programs you will approach, so you can compare answers. See prepare your data for sale.

Why it is worth the time

Slow questions now, fewer surprises later

A data deal moves through NDA, buyer review, agreement, export, de-identification and acceptance. Each step is a chance to fix terms that are hard to change afterward.

60 to 90 days

What practitioners cite as a typical time to close. There is room to ask questions without losing the deal.

More than one

Offer, that is. Practitioners advise sharing a manifest and samples with several buyers, never the full dataset before price.

One document

Only the signed agreement binds both sides. Program pages and emails describe intent; the contract is the deal.

Check eligibility

When to ask what

Not all 25 at once: a sequence that fits the deal

A deal moves through NDA, buyer review, agreement, export, de-identification and acceptance. Each group of questions has a natural moment, and asking too early or too late both cost you.

Before the NDA: is this worth pursuing? Ask questions 1, 2, 6 and 9 in general terms. You only need to know whether the buyer's model of price, payment, scope and exclusivity could ever suit you. If the answers are far from what you need, stop here and save everyone weeks.
After the NDA, with a manifest and samples: ask questions 7, 8 and 10 on scope, then 11 to 15 on privacy. The buyer can now see what you have, so it can answer concretely. This is also when you hand over your exclusions list and ask to see how it will be applied.
With the draft agreement: work through questions 16 to 25 on liability and exit with your lawyer, clause by clause. Compare each written answer you collected earlier with the contract text, and flag every place where they differ.
Before signing: come back to questions 3, 4 and 5. Acceptance criteria, payment dates and reclaim clauses are often the last terms to settle and the first to cause trouble. Check them against the final text, not the earlier draft.
After signing: keep the answers file with the signed agreement. If a dispute or a staff question comes up a year later, the record of what was asked and answered is often the fastest way to resolve it.

Common mistakes

Six ways sellers weaken their own position

None of these is about a particular buyer. They are patterns on the seller's side that the 25 questions are meant to prevent.

Asking only about price

The headline figure is the easiest number to compare and the least informative. Exclusivity, acceptance and liability decide what the price is actually worth to you.

Sending the full export for a quote

Once a buyer holds everything, your leverage on scope and price is gone. Practitioners advise a manifest and samples first, and never the full dataset before price.

Relying on spoken answers

Answers given in conversation are easy to remember differently later. Ask for the answer in writing, then check that it appears in the agreement.

Talking to one buyer only

With a single offer there is nothing to compare terms against. Asking several buyers the same questions shows quickly which terms are normal and which are not.

Bringing in the lawyer last

A lawyer who sees the deal only at the final draft can flag problems but has little room to fix them. Involve yours when the first draft arrives.

Forgetting people inside the company

Client partners, HR and IT each know about restrictions the owner may not. Ask them before you answer the buyer's questions about consent and exclusions.

FAQ

Questions about the questions

When should I ask a data buyer these questions?

After an NDA is in place and before you send anything beyond a manifest and a few samples. Ask in writing, and ask every buyer you are talking to the same questions so the answers can be compared.

Which questions matter most?

Exclusivity, scope of use, who carries the risk if de-identification misses something, and the acceptance criteria that trigger payment. Those four decide whether a high headline price is worth it.

Is a verbal answer from a buyer enough?

No. A program page or an email describes intent. The signed agreement is what binds both sides, so each answer that matters to you should appear in the contract text.

Do I still need a lawyer if I ask all 25 questions?

Yes. The questions help you understand the deal and spot gaps. A lawyer reads the agreement against your client contracts, employee notices and any sector rules. This page is general information, not legal advice.

Should I send a full dataset so the buyer can price it?

Practitioners advise against it. Share a manifest and samples, get more than one offer, and send the full export only after price and terms are agreed in writing.

Keep reading

Eligibility checker Buyer programs compared AI data licensing agreement Exclusivity and resale rights Indemnities and warranties Getting more than one offer