What happens to your confidential deal information when investors analyze your deck with AI—and what you can do about it.
What happens to your confidential deal information when investors analyze your deck with AI—and what you can do about it.
You spent six months building it. The underwriting is flawless. The market analysis is thorough. You have letters of intent from your contractors. Your team's track record speaks for itself. The deal is real, the returns are solid, and the risk is mitigated. You've put $500K in soft costs and endless hours into due diligence.
Now you're sending your confidential offering memorandum to 300 potential investors.
Here's what you don't know: Of those 300 investors, probably 40-60% will paste your deck into ChatGPT to analyze it. They'll ask it questions about your assumptions, your fee structure, your team's experience, your acquisition strategy. They'll compare your deal to others they're evaluating. They might even ask ChatGPT to find holes in your underwriting.
And here's the uncomfortable question: What happens to that data after they upload it?
The honest answer? It's more complicated than you think—and the risks are real in ways you've probably never considered.
Let's start with the facts, because there's a lot of misinformation on this topic.
OpenAI's data policies have evolved significantly over the past two years. The headline that many people saw in late 2023 was: "OpenAI won't train on ChatGPT Plus data." This was a relief to many enterprise customers. But the full picture is more nuanced—and more concerning for GPs distributing confidential deals.
| ChatGPT Tier | Training Data Usage | Data Retention | Enterprise Security |
|---|---|---|---|
| Free Tier | Can be used for model training unless you opt-out in settings | Retained indefinitely (with opt-out available) | No |
| Plus/Team | Not used for training by default | Stored on OpenAI servers (~3 months standard) | Basic |
| Enterprise | Explicitly NOT used for training | Not retained beyond 30 days | Yes (SOC 2, encryption, VPC options) |
| API (IRDESK tier) | NOT used for training | Not retained beyond 30 days | Yes (same as Enterprise) |
Here's what matters: Most individual investors using ChatGPT are on the free tier. When they upload your deal to free-tier ChatGPT, OpenAI's terms of service explicitly state that conversations and uploaded files may be used to improve their models—unless the user has specifically toggled the data opt-out setting.
How many free-tier users know that setting exists? Not many. How many have actually enabled it? Even fewer.
If an investor is on ChatGPT Plus ($20/month), your data isn't used for training by default. But it's still sitting on OpenAI's servers. It's still accessible to OpenAI staff and potentially their security vendors. It's still subject to OpenAI's privacy policy, which has been updated multiple times.
The core issue isn't primarily about "spying." OpenAI isn't deliberately stealing your deal secrets. The problem is: once your confidential deck is uploaded to ChatGPT, you have zero control over that data. You don't know it happened. You can't delete it. You can't track it. You can't see what analysis was performed or what answers the AI gave about your deal based on (potentially misread) information.
Let's address the "spying" scenario directly, because it's worth understanding even if it's not the primary concern.
Could a competitor upload your deal to ChatGPT and extract information from models that were trained partially on investor-uploaded decks? Theoretically, yes. But here's why it's actually unlikely to be a major risk:
LLMs don't memorize and regurgitate documents like databases. They learn patterns, language, structures, and relationships from vast amounts of training data. When a model processes your 50-page investment memorandum, it doesn't create a searchable index of "this fund charges 2% management fee" that a competitor can query. Instead, it learns probabilistic relationships between concepts: how deal documents are typically structured, what language is common in real estate syndication, what assumptions underwriters typically make.
If a competitor asks ChatGPT "What fee structures do multifamily syndicators typically charge?" it might generate an answer that's informed by training data that included investor-uploaded decks. But will it mention your specific deal? Will it reveal your exact fee split? Almost certainly not. The model would have to somehow retrieve your specific document from its weights—which isn't how transformers work.
The honest assessment: Direct memorization and retrieval of your specific deal data is highly unlikely. But that's not the only risk.
The real dangers are more subtle but more significant:
When investors upload your deck to ChatGPT and ask it to analyze your market assumptions, assumptions about vacancy rates, rent growth, or cap rates—that conversation happens on OpenAI's servers. If that data is used for model training, it contributes to OpenAI's understanding of current market conditions in your specific property type and geography.
Over time, if dozens of deals in your market are uploaded to ChatGPT, the model learns your market's typical assumptions. A sophisticated competitor could use this to reverse-engineer what the "baseline" expectations are for deals like yours. This is indirect but meaningful competitive intelligence.
Your fee structure is proprietary strategy. It reflects your capital raising environment, your competitive positioning, and your underwriting rigor. If your typical investor—who's seeing multiple deals—uploads several offers to ChatGPT, they're creating a comparative data set that could inform models about typical fee structures in your market.
Your offering memorandum includes your team's background, prior exits, track record numbers. If that information is used in model training, it becomes part of the public probability distribution that ChatGPT understands about your team. An investor or analyst curious about your firm could extract insights about your historical performance from model outputs.
The specifics of how you model cash flows, how you stress-test assumptions, what CAP rate assumptions you use in what markets—these are methodological IP. If an investor uploads your deck and ChatGPT helps them understand your underwriting logic, those methodologies become part of the training landscape for future models.
The headline that should concern you isn't "Are competitors stealing my data?" It's "I no longer control what happens to my deal information once it leaves my office."
Here's what happens when an investor uploads your offering memorandum to ChatGPT:
This is the "zero insight problem." Your deal information has become something you can't observe, can't control, and can't verify. In an industry built on information asymmetry and trust, that's a real problem.
The more fundamental issue: You're operating in an environment where the default behavior of capital allocators is to upload your confidential information to consumer AI tools. You can't stop them (and shouldn't try—it's how due diligence works now). But you can change the infrastructure they use to do that analysis.
The uncomfortable truth first: You cannot prevent investors from uploading your deck to ChatGPT. And frankly, you shouldn't want to. Investors need tools to evaluate deals. If you tell them "don't use AI to analyze this," you're just pushing them toward proprietary tools you have even less visibility into.
What you can do is control the infrastructure they use when they want to perform AI analysis on your deal.
Modern investors expect AI-assisted analysis. They want to upload a document and ask questions. They want analysis, summaries, comparisons. The question isn't whether they'll use AI—it's where they'll use it.
If you provide a deal room with integrated AI analysis (built on enterprise-grade APIs, not consumer ChatGPT), you accomplish several things:
Specifically, this means using API-tier access to major AI models (like OpenAI's API or similar), not consumer-tier ChatGPT. Here's why that matters:
API-Tier vs. Consumer ChatGPT:
Free/Plus ChatGPT: Your data may be used for training. Retention varies. No enterprise security. You have no control.
API (Enterprise-grade): Data explicitly NOT used for training. Data deleted after 30 days. Encryption in transit and at rest. Access logging. SOC 2 Type II compliance. This is the difference between hope and guarantees.
When you send your offering memorandum, you can include a note: "This investment opportunity has been uploaded to [Your Deal Platform], which includes AI-powered analysis tools. If you have questions about your investment, we encourage you to use our built-in AI analysis rather than uploading this document to external tools. Our platform uses enterprise-grade security and data privacy protections."
This accomplishes two things: It tells sophisticated investors that you're thinking about data security (which builds confidence), and it gives them an alternative to the default behavior of pasting your deck into free ChatGPT.
Beyond AI, deal rooms with comprehensive audit trails mean you know which investors are engaging with your information and how. You can see which sections they're focusing on, which analyses they're running, and when. This isn't surveillance—it's information parity. Right now, you have none.
The core shift happening in capital formation right now is this: Your deal documents are no longer contained. The moment you send your offering memorandum to an investor, you have to assume it will be processed by AI systems you don't control.
This isn't unique to ChatGPT. It's becoming true across the entire ecosystem. Every investor who sees your deal might run it through Claude, or Gemini, or Perplexity, or specialized real estate AI tools you've never heard of. Some of those are consumer-grade. Some are enterprise-grade. Some have data privacy protections. Others don't.
You don't get to choose. Your only leverage is in the infrastructure you provide alongside your offering.
GPs who recognize this are already adapting. They're moving toward deal rooms with:
The GPs who don't adapt? They're betting that the default behavior—investors uploading decks to free ChatGPT—is acceptable. Maybe it is. But they have zero visibility into it and zero control. They don't know if their market assumptions are being extracted and used against them. They don't know if their fee structures are being analyzed and compared. They don't know what misinformation about their deal is spreading through investor networks based on AI analysis of partially understood documents.
The uncomfortable truth: Your competitors are probably already losing control of their deal data to consumer AI. You have an opportunity to be the GP who didn't—who set up infrastructure that allows analysis while maintaining control and visibility.
If this resonates, here's what to consider immediately:
1. Audit your current data practices. Where are you distributing decks? What's your process for sharing confidential information with LPs? Are you warning investors about uploading to third-party AI tools?
2. Evaluate deal room platforms with enterprise AI. Look for platforms that offer AI-powered analysis on API-tier infrastructure (not consumer ChatGPT). Verify they have explicit data privacy guarantees. Check for SOC 2 compliance. Confirm that data is not used for training and has short retention periods.
3. Update your offering materials. Don't be alarmist, but be clear: "We've provided integrated AI analysis tools in our deal room to enable thorough due diligence while maintaining data security and confidentiality." Sophistication reads as strength.
4. Monitor and learn. As you implement deal room infrastructure, pay attention to which analyses investors run. What questions are they asking? What sections of the deal are they scrutinizing? This intelligence is valuable for future deal marketing.
5. Build institutional knowledge. Talk to your investor relationships about how they evaluate deals. Understand which of them are already using AI tools (spoiler: most are). Position your deal room as the solution that acknowledges this reality rather than pretending it isn't happening.
Are other real estate funds spying on you through ChatGPT? Probably not directly. But your confidential deal information is floating through AI systems you don't control, generating analysis you can't verify, accessible through data retention policies you didn't negotiate.
The question isn't whether to accept this reality—you have to. The question is whether you're going to shape the infrastructure around it or just hope it doesn't cause problems.
The GPs raising the largest capital checks in 2026 won't be the ones who ignore AI in diligence. They'll be the ones who integrated it thoughtfully, transparently, and securely—while maintaining visibility and control.
That's not paranoia. That's competitive advantage.