DataPacts

Seller track · Lesson 9 of 9 · 9:03

When to say no, and spotting scams

What you'll learn

  • Six ways to say no
  • What each one costs you
  • Five scam patterns
  • Seven checks before you reply

Free lesson

Enter your email to watch this lesson and the rest of the course.

Transcript

Here's the answer. Saying no is a perfectly good outcome. A company can keep its data, use it in its own models, release it openly, share it with researchers, license a narrower slice, or delete it. Not licensing at all is valid.

First, the six ways to say no, and what each one leaves you.

Each of the six paths keeps or gives up different rights. Only some can be undone later. So as you look at them, keep asking how reversible each choice is.

Path one: keep it. You do nothing for AI buyers, so every right, every later option and every confidentiality protection stays with you. No grant exists to recall. The cost is simple: no license income.

Income is uneven, so what you'd forgo is uncertain too. One stock photo company reported its data licensing revenue down seventy-six percent in a year, from sixty-five point seven million dollars to fifteen point eight million.

Path two: use it in your own models. The value stays inside your company, and nobody else's model learns from it. But the privacy exposure stays too. A European data protection board opinion finds that models trained on personal data can't always be considered anonymous. And you carry the cost and the legal duties yourself.

Path three: release it openly. Anyone can use it, including every AI developer, without paying. You can't recall it. A public web archive says its files are immutable, with removals only from later crawls. And buyers stop paying for what's already public.

Path four: share it with researchers. You keep commercial rights while the data gets studied. Income is little or none. And once the data is with a research partner, statutory exceptions apply whatever the agreement says.

Path five: license less. Go through a collective, set terms for crawlers, or keep the grant narrow. You get some income and more control. But collective and crawler payouts aren't published, and it's still a license.

Path six: delete it. That ends the value and the risk. But it's permanent, with no later license, research use or own use. You can't delete records the law requires you to keep. And deleting after a license doesn't reach models already trained on the data. Talk to a lawyer about what you must keep.

So which can you undo? Keeping can be undone, because a license stays possible later. Using your own models, partly. Open release and deletion can't be undone. Neither can an irrevocable grant, once signed.

The source page puts it in one line: a no today can become a license later, but an irrevocable grant, once signed, cannot be reversed. A yes is the step you can't walk back, so give it all the time it needs, and never let a stranger's deadline set the pace.

Now the other side of saying no. Spotting the emails you should never answer.

Data licensing emails come from four kinds of sender. Buyer staff, writing from the program's own domain. Paid referrers, who earn when a company signs. Intermediaries, who sell data for companies. And impersonators. Only the last kind is a scam. But the first three have their own incentives, so check every sender.

Five patterns show up again and again. Impersonation, an upfront fee, access before a contract, changed payment details, and pressure. Each has a check that costs nothing.

A fair warning on evidence, because we only state what the sources show. The site knows of no court or regulator record of a scam impersonating a named data buyer. So these patterns come from general fraud records and from the buyers' own published rules.

Pattern one: impersonation. Someone uses a known buyer's name on a domain or profile the buyer doesn't control. In the United States, posing as a business violates the FTC Impersonation Rule, sixteen CFR Part 461. One program even forbids its own referrers from posing as its staff.

Pattern two: an upfront fee. Someone asks you to pay for a valuation, a listing or a compliance check. That's the clearest red flag, because programs that publish their terms state the company pays nothing. Each of those is the program's own claim.

General fraud data shows the scale. The FBI's complaint center logged seven thousand ninety-seven advance-fee fraud complaints for 2024, with over a hundred million dollars in reported losses. That's general fraud, though, not data licensing.

Pattern three: access before a contract. Programs differ here, so ask what comes first. One states its steps as scope, connect, process and payout. Another offers a written estimate, then a read-only connection. An email demanding credentials or full access, with no program step named, has no legitimate counterpart.

Pattern four: changed payment details. A message that swaps the bank account for a payout should be checked by phone, on a number you already hold. Never by replying. The legitimate order is a wire after delivery or acceptance, so a new account number outside that sequence is a reason to call.

Again the scale is large. The FBI's complaint center reports twenty-one thousand four hundred forty-two business email compromise complaints for 2024, and over two point seven billion dollars in reported losses. Again, that's general fraud.

Pattern five: pressure. A one-day deadline or a seven-figure headline doesn't prove a scam. But it's a reason to slow down, because some legitimate incentives reward speed. One program states a ten thousand dollar fast signing booster, so a referrer can have a reason to rush you.

Headline figures can clash too. One program's page says earn a hundred thousand to four million plus. On that same page, its estimator says up to a hundred thousand. Treat them as separate claims.

Now the checks to run before you reply to anyone.

Seven checks, and none needs a reply first. One: find the program page yourself, not through the email's link. Two: compare the sender's domain with the program's. Three: ask which legal entity signs and pays. Four: ask whether the sender is paid for the introduction, by whom and how much.

Five: compare the steps they ask for with the program's published process. Six: send no records, credentials or account access before signed terms. Seven: confirm any bank details by phone. And if a program isn't on a list you find, that alone isn't a scam. Reach it from its own site.

What if the sender is a paid referrer? That's not a scammer. But the referrer is paid by the buyer when you sign or deliver, so the advice isn't independent. In the United States, a payment connection to a seller must be disclosed clearly and conspicuously. So ask.

One more point. A real email still isn't a good offer.

Once the sender checks out, the question changes to whether the terms are worth taking. Declining is a valid answer. A reply can get you terms to compare. It carries risks too: one program's master terms give thirty business days to reject a delivery, after which it's deemed accepted.

You have options besides replying. You can leave the email unanswered. You can decide not to license the data at all. Or you can ask several programs for their terms and compare them. And remember, a reply confirms your address is active.

If you do spot a scam, save the email with its full headers, and leave any attachments unopened. Report it to the FBI's complaint center, and tell the impersonated buyer through its own published contact.

Here's the whole track in one line. Decide whether to sell, check your rights, strip out personal data, value it, read the terms that matter, get more than one offer, get paid properly, and know that every step can still end in a no.

That's the end of the seller track. Everything in it, with sources, is at datapacts dot com.

Read and watch the full lesson free. Enter your email to unlock every lesson of the academy.

Sources on DataPacts