{/* Schema recommendation: BlogPosting + FAQPage.
- BlogPosting: author Nalin Vahil, datePublished and dateModified 2026-08-28. Emitted by the post template from frontmatter.
- FAQPage: emitted from the frontmatter faq block. Do not duplicate the FAQ in the body; the template renders it below the article. Internal linking: first SBIR mention links to /insights/sbir-guide-for-startups; companion piece linked inline (/post/dod-sbir-2026-reopening-guide); CTA links to /roadmap-intake. */}
Can ChatGPT find SBIR grant opportunities? Not reliably. In a systematic comparison, web-search-based discovery missed every one of 115 specific open DoD SBIR topics that the Department of Defense's own SBIR/STTR Innovation Portal (DSIP) API correctly listed as open. Not most. Not some. All 115.
That gap isn't a fluke. It's structural, and it matters if you're a founder using ChatGPT or Google to find your first grant opportunity.
Can ChatGPT Find SBIR Grant Opportunities? We Ran the Test.
Here's the setup. In April 2026, when SBIR/STTR was reauthorized and DOD published 115 new topics on DSIP the same day, we ran a direct comparison on DoD SBIR topic discovery for the automated discovery pipeline behind Cada's own Grant Roadmap Builder. On one side: general web search, the same approach ChatGPT, Perplexity, and most AI grant tools use under the hood. On the other: the DSIP API, the Department of Defense's own structured database of every SBIR/STTR topic and its live status.
The comparison worked like this. Our discovery agents first searched the open web for currently-open DoD SBIR topics, the way a founder using ChatGPT or Google would search. We then pulled the same window of topics directly from the DSIP API, which returns verified open and close dates for every topic. We checked the two lists against each other, topic by topic.
An "open topic" has a specific meaning here. It's a solicitation with a defined scope, an open date, and a close date, currently accepting proposals. This is different from a general program area or a standing announcement. It's a narrow, time-boxed window.
Across that comparison, the DSIP API returned 115 currently open topics with verified open and close dates. Web-search-based discovery, run the way a founder would actually run it, surfaced zero of those 115 as specifically open and correctly dated. Zero.
This is the single most common failure mode we see in AI-assisted grant discovery: reporting closed programs as open, or missing the specific topics that are actually live.
Why AI Web Search Gets Grant Status Wrong
The mechanism is simple once you see it. Web search tools, including the ones powering most AI chat assistants, index pages that a crawler visited at some point in the past. That's fundamentally different from querying a live database.
A topic's status field in DSIP can flip from "pre-release" to "open" to "closed" within days. The cached web page describing that topic might not get re-crawled for weeks or months. When an AI tool answers your question, it's often pulling from that stale cache, not from the government's live record.
This isn't hypothetical. During that same April 2026 comparison, our discovery agents logged specific instances of it: cached search results showed DoD topics A234-027 and A244-021 as still open weeks after DSIP had marked them closed. A founder relying on that search result would spend hours drafting a proposal for a topic that had already stopped accepting submissions.
The fix sounds obvious in hindsight: query the authoritative source directly instead of the web's cached copy of it. In practice, almost no AI search tool does this, because it requires knowing which structured API is authoritative for which agency, and most tools are built for general web queries, not government data schemas.
The Always-Open BAA Trap That Wastes 40+ Hours
There's a second failure mode that's arguably worse, because it doesn't even require stale data. It requires misreading a solicitation that was never time-sensitive in the first place.
A perpetual BAA (broad agency announcement) is a solicitation that's effectively always open. Some accept submissions on a rolling basis. Others carry deadlines years out. Real examples: DARPA's office-wide BAAs, USSOCOM's multi-year BAA, and the DEVCOM Soldier Center BAA, which runs from 2025 through 2030.
These are legitimate ways to pitch the government. But they're not "new opportunities" in any meaningful weekly sense, and an AI search tool has no way to distinguish "I just found this BAA today" from "this BAA has been open for three years and will be open for three more."
Here's why that distinction matters to you. A founder who sees "DARPA BAA open now" and treats it like a normal SBIR deadline will often write to the BAA in general terms, without matching it to a specific active topic the program office is actually funding right now. That's 40 to 80 hours of writing time aimed at a moving target, with meaningfully lower odds of a response, let alone an award.
The tell is almost always in the deadline. If a submission window is rolling, or more than 12 months out, or the title contains "Office-Wide," "Long Range," or "Enterprise," you're probably looking at a perpetual BAA, not a fresh opportunity.
How to Check If an SBIR Topic Is Actually Open
This is the exact verification sequence we run before recommending any government opportunity to a client. None of it is secret. Here it is:
- Identify the authoritative source for that agency. For DoD SBIR/STTR, that's the DSIP API (dodsbirsttr.mil). For civilian agency SBIR/STTR (NSF, DOE, DHS, NASA, USDA, EPA), that's the SBIR.gov API. For broader federal opportunities, that's Grants.gov. A vendor's search tool, a blog post, or a cached web page is never the authoritative source.
- Treat that source's live status field as ground truth, full stop. If a general web search or an AI tool tells you a topic is open but the authoritative API doesn't list it, the topic is closed. No exceptions.
- Check the deadline length before getting excited about a BAA. Rolling submissions or a deadline 12+ months out means you're looking at a standing announcement, not a time-sensitive window. Treat it as a long-term relationship-building opportunity, not an urgent deadline.
- Check for recent amendments. Agencies update BAAs and topic areas without always re-announcing loudly. A BAA you dismissed six months ago may have added a topic area that fits you now.
- When in doubt, go to the source document. Read the actual solicitation PDF or the API record, not a summary. Eligibility rules and scope details are frequently oversimplified in secondary sources, including AI-generated summaries.
This is tedious. It's also the entire difference between spending 40 hours on a genuinely open opportunity and spending 40 hours on one that closed a month ago.
Where ChatGPT Helps -- and Where It Doesn't
To be fair to the tools: AI search is genuinely useful for parts of grant discovery. It's good at explaining what an agency's review process looks like, summarizing a program's general focus area, and helping you draft language once you know what you're applying to.
It's unreliable for anything with a status that changes on a government timeline: is this specific topic open right now, what's the actual close date, has this BAA been amended. That category of question needs a live, structured, authoritative source, not a general-purpose search index.
Frankly, this is why Cada's own discovery process doesn't rely on AI web search for topic status at all. It queries the DSIP, SBIR.gov, and Grants.gov APIs directly for that layer, and reserves AI and human judgment for the parts that actually require judgment: is this program genuinely competitive for your technology, is it worth 40+ hours of writing time, and how does it fit into a broader portfolio across federal, foundation, and state programs. AI plus structured data plus expert judgment beats any one piece alone.
Get a Free Opportunity Verification
This piece opened with a question: can ChatGPT find SBIR grant opportunities on its own? Not reliably enough to skip verification. If you've already found a solicitation through ChatGPT, Google, or an AI grant tool and you're not sure whether it's actually open, send it our way. We'll check it against the agency's authoritative source and give you a straight answer: open, closed, or perpetual BAA that's worth a longer-term play. No pitch, no obligation.
If it turns out to be a real opportunity, that's usually the moment to talk about the bigger picture: which other federal, foundation, and state programs fit your technology, and how to sequence them so your first grant isn't your last.
Sources
- DSIP -- dodsbirsttr.mil -- the Department of Defense's SBIR/STTR Innovation Portal and API, the authoritative source for DoD topic status used in this comparison
- SBIR.gov and Grants.gov -- the authoritative sources for civilian-agency SBIR/STTR and broader federal opportunities, respectively
- Cada's DOD SBIR 2026 reopening guide -- the April 13, 2026 SBIR/STTR reauthorization (S. 3971) and the 115 DOD topics published on DSIP the same day, the window this comparison was run in
- SAM.gov -- DEVCOM Soldier Center BAA (solicitation W911QY-25-R-0023), active February 2025 through February 2030
- The 115-topic web-search-vs-DSIP comparison, the two topic-code examples (A234-027, A244-021), and the verification checklist are Cada's own internal methodology and testing, not third-party research
Guidance reflects DSIP, SBIR.gov, and Grants.gov practice as of August 2026. Solicitation and topic status change constantly; verify against the live authoritative source before writing.