Updated
For owners holding an offer
Should We Accept an AI Company's Offer for Our Data?
Short answer
Not right away. First find out who will use your records, what they can do with them, what stays out and when you get paid. A fair offer will still be there after a week. Ask a lawyer to check it before anyone signs.
Should we accept an AI company's offer for our data? Not the first one, and not before you check a few things. Find out who will use the records and what they may do with them. Find out what stays out and when the money arrives. Many of these offers are real, and some are fair. A good offer will still be there after a week of careful review.
This page gives you a checklist for the offer in front of you. It covers red flags, questions to send back and records to keep out at any price. It also shows which programs with public rules fit your company. That gives you something real to hold the offer up against.
- Red flags to spot first
- Questions to send back
- You decide, on your timeline
Why now
Why are AI companies calling working businesses?
AI makers want writing that shows how real companies work. Think of the email thread that closed a sale, or the chat that fixed a problem. That kind of writing is rare on the public web. Your files hold years of it.
For a while, most sellers were startups that had shut down. That has changed. Lawyers at Frankfurt Kurnit Klein & Selz wrote about this on October 1, 2026. They said operating companies are now getting the same calls. Their note also points out a second path that owners often miss.
Buyers also reach out to the software vendors that store a company's data. A vendor is a company you pay for a service, like email or file storage. The note warns that a vendor may sell that data when it is bought, shuts down or goes bankrupt.
So an offer that shows up out of the blue is not strange by itself. What matters is how the buyer acts once you ask questions. A serious buyer puts terms in writing and gives you time. For background on the wider market, read our guide on how businesses sell data to AI companies. It also covers which record types qualify.
Red flags
What are the warning signs in an offer for your company data?
One of these alone does not prove a bad deal. Two or more is a reason to slow down. Bring in a lawyer before you reply.
A very short deadline
Some offers come with a tight reply date. They push you to decide before you have read the terms. You may live with a license for years, so it deserves more than a weekend. A serious buyer will give you time to bring in a lawyer.
No answer on who uses the data
The buyer may not say what kind of company trains on your records. It may not say whether the records get resold. Then you cannot judge the risk. At the very least, ask what type of company the end user is.
Any fee you must pay first
A real data buyer pays you. Some ask you to pay first, such as an appraisal fee, a listing fee or a processing fee. That is the clearest sign to walk away.
Use rights that say almost nothing
Watch for phrases like "any purpose" or "all present and future uses" in the rights. Words like these let your records go far beyond model training. Use rights should be named and narrow.
No sample review
Before anything leaves your hands, you should see cleaned samples. Check that names and private details are gone. A buyer that skips this step is asking you to trust it blindly.
No plan for removing personal details
Ask who removes names, when it happens and how. To anonymize means to strip out anything that shows who a person is. Saying "we anonymize everything" with no written process is not a plan.
A request to plug a tool into your systems
An API is a tool that lets one app pull data out of another. Slack's API terms set rules for outside apps that connect this way. Under them, an outside app may not use what it gets to train a large language model. Those are the models behind AI chat apps. Other platforms set their own limits. An export your own admin runs keeps you in control of what leaves.
Pressure to send samples before any review
Raw samples are your data. If you share them before your lawyer reviews the deal, you give away the very thing being priced.
Your reply
Questions to ask the AI company before you say yes
Send these back in writing and keep the answers.
| Question | What a clear answer looks like | Why you ask |
|---|---|---|
| Who will end up using our records? | The type of company that trains on them, such as a frontier AI lab. Also whether the buyer resells them. | You are deciding who holds a copy of your company's history |
| What exactly may you do with them? | Training and testing AI models. Resale, publishing and sharing raw records are ruled out unless you agree. | Wide rights are worth more to the buyer and cost you more |
| How long do the rights last? | A set length of time, or a plain list of what lasts after it ends | Rights that never end need a much closer read and a better price |
| How do you remove personal details, and can we check samples? | A written method, who runs it, and cleaned samples you review before anything moves | Customers and staff show up all through work email and chat |
| What stays out? | Your list of what is left out, written into the contract by system, folder and record type | Spoken promises about what gets skipped are hard to enforce |
| When and how do we get paid? | A fixed amount and a due date tied to delivery or a set review period | Pay tied to an open-ended approval can drag on for months |
| What happens to raw copies afterward? | Raw files deleted on a set date, confirmed in writing | Deleting files cannot undo training, so at least you want the copies gone |
| Who pays for our legal review? | A clear yes or no, settled before your lawyer starts | Review costs money, so count it when you weigh the price |
Keep out
What should you leave out of any data deal, whatever the price?
These records come with duties that a license fee does not cancel. Name each one in the contract as left out. Our page on whether it is legal to sell company data explains the laws behind them in more detail.
HR, payroll and benefits
Staff files, reviews, pay and leave records hold the most private facts about your people. The CCPA is California's privacy law, and it covers staff information too. A paid transfer of it may count as a sale. The exception is when the details are properly de-identified first, so no one can be named from them.
Staff talk about pay and conditions
Federal labor law can protect these talks, whether or not your team has a union. The rule is Section 7 of the NLRA, the National Labor Relations Act. Leave out HR channels, private messages and any thread about schedules or wages.
Mail with your lawyers
Talks with your lawyer are usually covered by privilege, which keeps them private. Sharing that advice outside the company can weaken that protection. Keep your lawyer's mailboxes and legal folders out entirely.
Clients' and partners' private material
NDAs and client contracts often cover what others told you by email. An NDA is a promise to keep certain information secret. Their plans and prices are theirs to protect. Your agreement with them may forbid sharing.
Customer personal data
Contact lists, payment records and anything built to identify a person stay out. Your old privacy policy also limits what you can do with data you collected under it. The FTC, the federal agency that protects consumers, has warned against quietly changing those promises later.
Health and other regulated records
Patient records and similar files follow their own laws. Some programs do not take them by default. Others send them through a separate review.
Before you write back, see what fits your company
Tap through a short quiz on staff size, years open and the software you use. You will see a careful estimate. You will also see the programs whose public rules match your company, and why each one fits.
A reference point
Programs that publish their rules, and who each one fits
An offer is easier to judge when you know what programs with public terms ask for. Here are the kinds we can match you with. The facts below come from each program's published rules. If anything differs, the signed agreement always wins. We don't name them here. The estimate shows which of them suit your company.
Many-system programs
One program licenses a copy of a company's operating records, like email, chat, documents and CRM, for AI training and research. Through our referral link it takes firms with at least 50 office staff, or 15 at law, accounting, medical, banking and finance firms. Those firms must be based in America, Canada, Australia or an EU country. The program reviews each company and decides whether to make an offer.
Written know-how programs
One program values written know-how most. That means SOPs (step-by-step work guides), how-to guides, internal documents, CRM history and project records. CRM is the system sales teams use to track customers. It needs 30 employees or more, with records in English. US companies come first, then the UK, Canada and Europe. Sign-up starts with a short AI interview. It pays after it accepts the data, with a review period of up to 30 business days. Closed companies, and ones closing within six months, go through a separate shut-down program instead.
Code and chat programs
One program licenses team chat, email, files, tickets, project history and code to frontier AI and data labs. In its own words, data is never resold through brokers. A team of at least 20 full-time people qualifies, with work done mostly in English. US companies are preferred. It de-identifies everything before licensing. Owners get an upfront payment, then a share of every license after that.
Long-history programs
One program wants firms with a real operating history and licenses their full work record. That includes email, chat, files, CRM and accounting. It reads a sample before it names a price, and it anonymizes everything first.
What to do this week
How to handle the offer, one step at a time
We do not review, compare or negotiate offers. We show which programs fit your company and why. You and your lawyer make the call.
Reply that you are reviewing
A short note saying you need time for a legal review is enough. Send no samples. Sign no NDA that lets the buyer keep anything you share.
5 minutesList what you hold
Write down your systems, the years each one covers and about how much is in each. Mark each one in or out using the keep-out list above.
About 30 minutesGather the papers your lawyer needs
Pull every privacy policy you have posted and your staff handbook. Add client contracts or NDAs that mention data. These decide what can be licensed at all.
Under an hourTake the estimate
Answer the quiz on this page. You will see a careful number and your best program matches. Each match gives the reason it fits. It also flags terms worth a close read.
About 2 minutesHave your lawyer read everything
Your lawyer reviews the offer, any program agreement you like and your list of what stays out. Our data licensing agreement guide lists the clauses that need the closest read.
Depends on your lawyerDecide
Accept the offer, sign up with a program directly, or say no to both. Each is a sound choice when you make it with the facts in hand.
Your callThe market
Real deals that explain the calls owners are getting
These are public reports of AI companies paying for business records. None of them are DillisData deals.
Google's winning bid when Spirit Airlines' work records went to a bankruptcy auction in August 2026. The set includes millions of staff emails and Teams chats. Passenger and loyalty data were kept out. The sale still needs a judge's OK.
Chapter 11 Cases, Oct 2026The date a tech law firm warned of a shift. Buyers now call operating companies and their software vendors, not only closed startups.
Frankfurt Kurnit Klein & SelzWhat one AI data company offered for the same Spirit records once the auction had closed. That was 25% above Google's winning bid.
Fortune, Sep 2026One AI data company's own pay guide. It says a qualified deal can top $100,000 and big ones can top $500,000. Rare data can pass $1 million.
The Next Web, Oct 2026Timing
Don't rush your reply, but don't set it aside for good
A week of review costs little. Sitting on the question for a year may cost more.
The interest in your records is real today
Buyers are reaching out because labs want private business writing now. That demand is why you got the call.
Rare records get more money
Buyers pay most for kinds of records they do not have yet. As more companies in your field license, later offers there tend to get smaller.
Published terms can change
Programs may change their rules and prices as more companies sign up. What you can read and accept now may look different next year.
Nobody knows how long this lasts
Labs may collect enough of some record types. Or they may find other ways to train. No one can name the date, so a long delay is a gamble too.
Questions
Questions owners ask
Is a surprise offer for our company data a scam?
Not always. Many buyers are real, and they are contacting working companies directly. Stop and check if a buyer asks you for money. Do the same if it will not name the end user or gives you a 48-hour deadline.
How long can we take before we answer?
As long as a careful review needs. That is usually a week or two. Some buyers pull an offer because you asked for a legal review. That tells you something useful about how the deal would go.
Can DillisData review our offer or tell us if it is fair?
No. We do not review contracts, compare offers or negotiate. We show which programs with public rules fit your company and why. You and your lawyer decide what to do.
Should we send the buyer a sample so it can price our data?
Not before your lawyer has read the terms. If a sample is needed, remove names and private details first. Also get a written agreement on what the buyer may keep.
Our software vendor got an offer for data in our account. Can it sell it?
Check your contract with the vendor. Look for any right to sell, license or transfer your data. That includes de-identified or combined data. Ask your lawyer to tighten those terms before the vendor accepts anything.
Do we need to tell our employees?
That depends on where you operate, what your handbook promised and which records go in. If you leave out HR files and private chats, the touchiest material is gone. Ask your lawyer if the law requires you to tell staff or get their OK.
Will anyone find out we licensed our records?
Possibly. California's AB 2013 took effect in January 2026. It covers makers of generative AI, which is AI that writes text or makes images. It requires them to post general facts about their training data. Lawyers note that this includes where the data came from and whether it was bought or licensed. Ask the buyer how it handles that, and ask for a clause about naming you.
Who pays for the legal review?
Ask the buyer before your lawyer starts. Either way, count that cost when you weigh the price. Our matching is free for you; the program is the one that pays us.
Before you sign
Know which programs fit you before you answer the offer
Two minutes of questions gives you a careful estimate and your best program matches, with reasons. Then you, your lawyer and your team decide what comes next.
Free for your company. Your records never pass through us. Have your lawyer read any license.