What happens if an AI data buyer rejects part of your data?
· Company Data Advisors
Many AI data buyers pay only for records they accept. How rejection, de-identification losses and review windows work, buyer by buyer, and what to ask for.
At many buyers you are paid only for the part they accept. micro1's data license terms, updated on November 12, 2026, give it 30 business days after delivery to reject records in whole or in part, say it owes nothing for records it validly rejects, and let it reject everything if what remains is worth materially less. On November 17, a bankruptcy judge approved Spirit Airlines' sale of its business records to Google, after a revised agreement that, as WIRED reported, lets Google walk away if cleaning out personal information shrinks the data by more than 8 percent. Where payment depends on acceptance, the price is a price for accepted records, so the acceptance clause deserves as much attention as the number.
What does "acceptance" mean in a data license?
Acceptance is the point at which the buyer agrees that the records you delivered match what the contract described, and at buyers that pay on acceptance it is what starts the payment clock. Before acceptance the buyer can reject records it finds incomplete, unusable or different from what was promised. Records can drop out of a deal at three points:
- Scoping. You decide which systems, teams and date ranges are in, and which are out. Records you exclude here are your choice and should be written into the description of the data, not counted later as a shortfall.
- De-identification. Removing personal information can remove whole records, such as a thread that cannot be cleaned without losing its meaning. A de-identification step is the process of removing or replacing names and other details that could identify a person.
- The buyer's review. The buyer checks volume, quality and match to the agreed description, and can reject what falls short.
The first is in your control. The second and third are where an offer can shrink after you have done the work of exporting.
What do micro1's data license terms say about rejection?
micro1's Data License Master Terms, published on its website ("Last updated: November 12, 2026"), are, in our reading, the most detailed rejection terms published by any of the 17 programs in our index. Under section 1.3:
- A 30-business-day review. micro1 has "thirty (30) Business Days after completion of delivery" to reject data "in whole or in part," with written notice describing the deficiency. If it says nothing, the data is "deemed accepted at the end of the Review Period."
- Early acceptance only in writing, from the top. micro1 can accept early only by a written notice signed or sent by its General Counsel or Chief Financial Officer. A confirmation of receipt, or testing the data, does not count as acceptance.
- No payment for rejected records. micro1 has "no obligation to pay for any portion of the Data validly rejected."
- Partial rejection can become total. If rejecting part leaves the rest "no longer reasonably usable" or materially less valuable ("for example, due to reduced volume, coverage, completeness or diversity"), micro1 may reject all of it, end the agreement, and then "no amounts are payable to Licensor."
- A chance to fix. If micro1 rejects records but doesn't end the agreement, the owner is to correct the problem and resubmit.
- Claims after acceptance. micro1 has 60 days after acceptance to raise a latent defect in technical quality or a material mismatch with the agreed description; the owner then corrects or replaces the data, or refunds "the proportionate amount paid." That 60-day limit does not apply to claims about intellectual property, privacy, consents, data provenance or fraud, among others.
- Misstatements survive acceptance. Under section 4.3, if micro1 decides in good faith that the description of the data was materially "inaccurate or misleading," it may end the agreement with no cure period, owe nothing, and the owner "will refund any amounts already paid." The terms say this right "survives Licensee's acceptance or deemed acceptance of the Data."
- Who pays to export. The terms put "the costs of export and delivery" on the owner, and the owner may invoice only after acceptance.
The terms also say the license "attaches to a portion of the Data only once Licensee has accepted it" and paid for it, so rejected records are not licensed. On its program page, micro1 says it reviews delivered data "for completeness" and pays "in full within 30 days of acceptance"; the terms say the owner invoices after acceptance and is paid within the period set in the order form.
Our view: none of this is unusual for a buyer paying a fixed price for something it hasn't fully seen, and it is better to have it in writing than not. What an owner should notice is the order of events: you bear the export cost first, the review comes after delivery, and one rejection can, in some cases, unwind the whole deal. A misdescribed dataset can unwind it even after you have been paid.
What did the Spirit Airlines sale add?
The Spirit Airlines sale put a number on how much de-identification may shrink a dataset before the buyer can leave. WIRED reported on November 12, 2026 that the revised deal requires Spirit to hire a third-party de-identification company, Tonic AI, excludes employee medical records and documents about employee investigations and labor relations, and lets Google "renege on the sale if the data-cleaning process reduces the assets by more than 8 percent." On November 17, Judge Sean Lane of the U.S. Bankruptcy Court for the Southern District of New York approved the sale, according to Bloomberg Law, which described the data as nearly 2 billion emails from 90,000 accounts and 660 million Microsoft Teams items. We have not yet read the signed sale order, so check the docket (Case No. 25-11897) for the final wording.
Spirit is a closed airline selling through a court, not an owner signing a private license. But the clause answers a question every owner faces: if removing personal information takes out more than expected, who carries the loss? In Spirit's case the answer is a fixed threshold, agreed in advance, above which the buyer can leave. Our guide to licensing records that include employee and customer data covers the rest of the Spirit protections.
When does each buyer pay, and what does it review first?
AI data buyers differ on what is checked before you commit and what happens after delivery. At some programs you approve a sample or the buyer prices from one before you sign; at others the main review comes after delivery. Our view: a sample reviewed early helps, but it doesn't remove a later acceptance step where the buyer pays only on acceptance. The table sets out what ten of the 17 programs in our index say on their own pages about review and payment.
| Program | What is checked before or after delivery | When it pays |
|---|---|---|
| micro1 | The buyer reviews the full delivery for quality, accuracy and usability; 30 business days to reject | Program page: within 30 days of acceptance; terms: invoice after acceptance, paid within the order form's period |
| Appen | You approve a sample pack before anything is licensed; payment still waits for acceptance | "On acceptance rather than submission" |
| Centific | You review an anonymized sample before the data is used | On acceptance |
| Polyshares | The buyer reads a sample of the record to price it | Net 30 to 60 once the data is approved, shared and anonymized |
| SuperAnnotate | The buyer reviews a sample for volume, quality and uniqueness; you review a sample too | On delivery, and again on each relicense |
| Turing (Project Lazarus) | Quality check on samples within 5 business days, alongside the revenue share agreement | A share of each sale; nothing owed if nothing sells |
| Mercor | You approve before data leaves your systems; its team does not manually review your data | By wire on delivery of the anonymized dataset, typically 2 to 4 weeks after signing |
| Collinear | The buyer confirms delivery is complete | Per the contract, after delivery is confirmed |
| SourceX | A privacy review before delivery; the payout is fixed in a signed Dataset Deal Confirmation | Within 60 days of a valid invoice, sent after closing |
| Mode | You approve the scope before export | On the schedule in the written agreement, set before export |
Sources: each program's own page, read on November 18, 2026; micro1's master terms. Each buyer's page in our index links to its program.
Appen says "payment is on acceptance rather than submission," and that you review "a representative sample pack before anything is licensed." Centific says you review an anonymized sample and "Without your approval, it does not move forward." Polyshares says it reads a sample because "we do not pay for something we have not seen." SuperAnnotate reviews a sample "for volume, quality, and uniqueness" before it values the data. Turing lists a quality check of samples "within 5 business days" at the same step as its revenue share agreement. Mercor pays "upon delivery of the anonymized dataset to the buyer" and says its team "does not manually review your data." Collinear says "We confirm delivery is complete, then pay according to the contract." SourceX fixes the payout in the Dataset Deal Confirmation and is invoiced after closing. Mode says the written agreement "sets the amount and payment schedule before export."
Most of these pages describe the sequence, not the remedy. Of the programs we read, only micro1 publishes what happens when it rejects part of a delivery. For the others, the answer is in the contract you are sent, which is where to look before you export.
Who pays for the work if records are rejected?
The cost of exporting and cleaning records falls on different sides at different buyers, which matters most when part of a delivery is rejected. micro1's terms put export and delivery costs on the owner. Mercor says it "covers all costs associated with data extraction, anonymization, and transfer," and Appen says it covers "extraction, anonymisation and transfer, so there are no costs to you." Turing says de-identification, cleanup and packaging are "on us," recovered from sales.
Our view: when the owner pays to export, a rejection costs twice, in the payment you don't receive and the work you already did. Ask who pays before you agree to export anything.
What should you ask for before you export?
These are the questions we would put to any buyer before records leave your systems. None is legal advice; your counsel should read the final contract.
- What exactly is being bought? Get a written description of the data (micro1 calls it the Data Manifest) listing systems, date ranges, volumes and what you excluded. A buyer can reject against that description, so make it accurate, not optimistic.
- Is the price per accepted unit or for the whole? A price per accepted record, ticket or mailbox makes a partial rejection a smaller payment, not a renegotiation.
- How much can be rejected before the deal can end? Spirit's 8 percent figure is one public example of a fixed threshold. In our view, without one, "materially diminished" is decided after the fact.
- Does de-identification loss count as a defect? If the buyer's own cleaning removes records, ask that those losses not be treated as your shortfall.
- How long is the review, and what happens if the buyer is silent? In our view, micro1's rule that silence means acceptance protects the owner; check whether other buyers have one.
- Can you see a sample first? In our view, a buyer that has priced from a sample of your real records has fewer surprises to reject later.
- What survives acceptance? Ask which claims the buyer can still bring afterwards, for how long, and whether the remedy is a fix, a partial refund or a full one.
Before you start, read what you're agreeing to when you license company data for the other clauses (exclusivity, term, onward use), and one payment or a revenue share for how the payment structure moves risk between you and the buyer.
Where we come in
Owners comparing several programs rarely get to see each one's acceptance terms side by side before they choose, and those terms decide how much of an offer is actually paid. Company Data Advisors is an independent advisory firm. We track every major program's published terms, have direct contacts at the major data buyers and at AI labs, and read the contract with the owner, including what can be rejected, by when, and who pays for the export. Nothing about your company is shared until you approve a specific, named buyer. Participating buyers may pay us referral fees, and we tell you which before any introduction. Owners pay us nothing, and you can apply to any buyer directly.
Before you sign with anyone, write down what you will exclude, ask the seven questions above in writing, and keep the answers with the contract. If you want to see which programs fit your company first, start with the screener.
Programs in this post
More from the blog
- Is my company too small to license its data to AI buyers?
Most AI data buyers with a size floor set it at 20 to 50 employees, but law, accounting and medical firms qualify smaller, and nine programs state no minimum.
- What company data AI buyers want: November 2026
Across the 17 AI data-buyer programs we track, accounting, healthcare and software lead the industries, and CRM, ticket, email and finance records lead.
- How to check that an AI data buyer is legitimate
Six checks an owner can run in an afternoon before sharing records. Who the buyer is, what you pay, what it asks first, personal data, payment and license terms.
See which buyers fit your company
Eight quick questions against every program in our buyer index. Participating buyers may pay us referral fees.