A platform team hears these claims from founders, LPs, conference panels and vendor decks. None of them is a reason to introduce a company, and a few of them, repeated to a founder, set expectations nobody can meet.
Why AI headlines make poor introduction triggers
A headline describes a category. An introduction concerns one company, one set of records and one CEO's decision. The gap between the two is where referral diligence happens, and skipping it produces two kinds of waste: introductions of companies whose records were never theirs, and founders who expect a windfall because a panel said data was scarce.
Regulators have shown what testing an AI claim looks like. On March 18, 2024 the SEC announced settled charges against two investment advisers, Delphia and Global Predictions, for false and misleading statements about their use of AI; the firms agreed to pay $400,000 in total civil penalties, and Global Predictions had described itself as the first regulated AI financial advisor. In September 2024 the FTC announced Operation AI Comply, five law enforcement actions against companies that used AI hype or sold AI technology that could be used to deceive, among them business-opportunity schemes promising passive income from AI-powered online storefronts.
Those cases are not rules for referral partners. They show the three questions an outside reviewer asks of any AI claim: did it happen, what record backs it, and who said it. A platform team that forwards founders to a licensing conversation should ask the same three of every claim it is tempted to repeat. Deal teams testing a target's own AI story need a different tool; AI washing due diligence covers capability and data claims inside a sale process.
The claim-to-question translation worksheet
Five claims, each checked against the source most often behind it. The last two columns are written before anyone talks to the CEO.
| Claim as heard | What the source actually says | The company question | Answer that supports an introduction | Answer that stops it |
|---|---|---|---|---|
| "AI has run out of training data." | Epoch AI estimates the effective stock of quality-adjusted public human text at around 300 trillion tokens, with 80% confidence that it is fully used at some point between 2026 and 2032 if current trends continue | Which of this company's records were never on the public web? | Years of internal work records in a system the company controls | Most of the material is already public: published articles, marketing pages, open forums |
| "Synthetic data has made real records obsolete." | Microsoft's Phi-4 report says synthetic data makes up most of that model's training data, generated from seeds drawn from human-written web pages, books and code | Do the records pair a decision with what happened next? | A claim with its settlement, a ticket with its resolution, a bid with its result | Templates and boilerplate a model can already produce |
| "Every company's data is a goldmine, a passive income stream." | The FTC's Operation AI Comply targeted schemes promising passive income from AI; SourceX's terms count only compensation actually paid for selected data that reaches Closing, and call estimates projections, not promises | What does the CEO expect to do, and when does the CEO expect to be paid? | The CEO understands payment follows review, a signed agreement, delivery and Closing | The CEO wants a fixed figure before any review, or expects money for a meeting |
| "This company is AI-first." | The SEC's 2024 settled charges concerned advisers whose AI claims did not match their practice | Setting the AI story aside, what records did the work itself produce, and since when? | Records of real operations, made by the company's own people, regardless of its AI use | The AI claim is the only evidence offered, or most records were generated by a model rather than by staff |
| "Proprietary data is the moat." | SourceX's screening page treats material that belongs to a company's clients as usually not a fit | Whose contracts govern these records? | Records the company created itself; customer contracts silent or permissive, for SourceX's rights review to confirm | Records held under client contracts or inside clients' own accounts |
Method note. The five claims were chosen because they recur in AI data conversations; how often each is heard was not measured. Each "what the source says" cell was checked on the publisher's own site on 2026-10-26. The worksheet tests whether a claim reduces to a question about one company. It does not measure demand for that company's records, their quality or any payout, which only SourceX's review and a signed agreement settle.
Two habits keep the worksheet honest. Fill in the "stops it" column first, so a vivid supporting answer cannot rewrite your standard after the fact. And when a claim has no traceable source at all, the row ends there: nothing unsourced reaches a founder.
One founder's pitch through the worksheet (Illustrative)
Illustrative. A generic company and invented answers.
A portfolio founder forwards a slide: "Our 12 years of service data is worth eight figures to AI labs." The platform lead runs it through the rows.
- Source. None beyond the founder's estimate. The figure is not repeated to anyone.
- Never public? Yes: work orders and technician notes since 2013 in one field-service system, never published.
- Decision paired with outcome? Mostly: each work order records the fault, the fix and whether the customer called back within 30 days.
- Whose contracts? Work orders are the company's own; customer site notes include residents' names and phone numbers, which would have to be removed if they were licensed at all.
- Expectation? The CEO, who runs the 260-person company, wants a conversation but has a price in mind.
The worksheet does not kill this row; it changes the introduction. The platform lead tells the CEO that SourceX publishes no prices, that payment depends on review, scope and Closing, and that the referral reward, if any, is SourceX's and never comes out of the company's payout. The introduction then carries metadata only: system, record families, years, ownership as the CEO understands it. The customer questions guide shows how these signals surface in ordinary client conversations.
What the worksheet needs from the relationship and the records
A translated claim still needs three facts before anyone writes an introduction: a named person who owns the CEO relationship and will ask, a CEO who decides whether the company discusses licensing and signs for it, and records the company made itself. An introduction the CEO never heard about is, on SourceX's own screening page, usually not a fit.
The published floor applies to every row: 50+ full-time employees at peak, contractors excluded, documented operating history and rights to license. Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English. Funding status, AI branding and press coverage appear nowhere in that list.
Before any introduction, ask whether the company has already corresponded with SourceX. Under the terms, an earlier lead, account or email exchange makes it an existing relationship, reviewed as a duplicate rather than credited to a new partner. A partner submits only after signing the SourceX Channel Partner Agreement; agreements signed under its former name remain in force. What no worksheet can answer is left to SourceX's review: whether the records match a current program, whether customer contracts restrict reuse and what a license would cover. The one-system records question and the buyer diligence questions show what that review tends to probe.
A ten-company test of the worksheet
The worksheet is itself a thesis: that AI headlines are a worse source of introductions than operational facts. Test it before making it policy.
- List the last ten companies someone on the team suggested because of an AI headline, a founder's AI claim or a conference conversation.
- List ten suggested for operational reasons: a system migration, a long-running records business, a retiring owner.
- Run all twenty through the worksheet and the fit facts at metadata level, through each relationship owner.
- Count three numbers per group: rows where the claim reduced to a company question, companies passing the fit facts and CEOs who chose an introduction.
What would change the thesis. If the headline group produces as many consented introductions as the operational group, headlines are not misleading your team and the worksheet can shrink to the "stops it" column. If it produces fewer, keep the worksheet as a gate before any founder hears the word licensing. SourceX has not published conversion data by referral source, and no results from this test exist yet, so the comparison can only be made within your own network.
Where this worksheet goes wrong
- Over-skepticism. A founder who talks in headlines may still run a 300-person company with fifteen years of clean records. Judge the answers, not the vocabulary.
- Sources move. Epoch AI's 2022 analysis put full use of high-quality text at 2024; its later estimate moved the date back. A cell checked today needs a date and a recheck.
- Enforcement is not a standard. The SEC and FTC cases illustrate scrutiny of AI claims; they do not define what a referral partner must verify.
- Synthetic substitution. If generated data keeps replacing human records in more domains, the second row weakens and demand for some record types could narrow. SourceX checks relevance to each current program in review, and criteria differ by program.
- Endorsement drift. Diligence on a claim is not diligence on a company. An introduction says the CEO wants a conversation, nothing more; introducing without endorsing draws the line.
Writing to founders about AI data has its own pitfalls, handled in writing about AI data without a gold-rush pitch. Program basics for funds are on the venture platform teams page.
Next step
Refer a company. No cap on rewards. Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.
When a claim survives the worksheet as a company question with a supporting answer and a CEO who wants the conversation, register for a partner link, confirm the company against who qualifies and read the program terms before the introduction goes out.
Readers who are weighing SourceX itself as an investment face a different diligence task from this worksheet; investor materials are available on request through the SourceX investor overview.
This article is for information only and is not an offer to sell or a solicitation of an offer to buy securities. Any offering by SourceX is made only to verified accredited investors through its offering documents.