The options, and when each fits
1FDA’s 510(k) and Product Classification databases
Fits when: You know your product code, or can find it, and have time to read the summaries yourself.
Free and authoritative: the primary record. The 510(k) database searches record fields, such as product code, applicant, device name, decision date, review panel and whether a predetermined change control plan was authorized, not the text of the summaries. The Product Classification database lists every device type with its product code, class and regulation. FDA also publishes downloadable files of 510(k) records.
2Bearnier AI
Fits when: You want the candidates under your product code compared on intended use and technology, with the page behind every fact.
Searches FDA’s 510(k), De Novo, classification and guidance records, accepts plain descriptions as well as product codes, and links every precise fact to the FDA page it came from. US FDA only today. We make it, so weigh this entry accordingly and test it on a device you already know.
3General AI assistants
Fits when: You want to understand concepts, such as what substantial equivalence means, or to draft text.
They answer from training data and the open web, not from an index of FDA’s record, and can state K-numbers that do not exist or belong to a different device. Treat any K-number, date or clearance they give you as unverified until you have opened it in FDA’s 510(k) database.
4Regulatory consultants
Fits when: The predicate choice is contested, the device is novel, or you need someone accountable for the strategy.
Experienced consultants bring judgment the databases cannot: which predicate a review team is likely to accept, and when a De Novo or a Pre-Submission is the better path. They cost more and take longer, and they search the same FDA databases.
Side by side
| Option | Cost | Sources searched | Reads 510(k) summaries | Cites the page | Exports |
|---|
| FDA databases | Free | FDA 510(k) records and product classification | You open each summary | It is the primary source | Downloadable 510(k) files |
| Bearnier AI | Free for 3 answers a month; Pro $250/month | FDA 510(k), De Novo, classification, guidance, 21 CFR | Yes, including scanned ones | Yes, every precise fact | Predicate Finder CSV; Word export on Pro (coming soon) |
| General AI assistants | Free and paid plans | Training data and the open web | Only documents you upload | No; K-numbers need checking | Copy and paste |
| Regulatory consultants | Hourly or project fees | FDA databases, plus experience | Yes, by hand | In their report | Their deliverable |
Bearnier AI’s entries describe our own product; the FDA entries come from FDA’s database pages, checked Oct 16, 2026. Prices for other tools vary, so none are quoted.
How to check a predicate
Whichever tool found it, check each candidate against FDA’s criteria before you build on it. Items 4 to 7 come from FDA’s 2023 draft guidance, which is not final.
1. Same intended use
A device is substantially equivalent when it has the same intended use as the predicate. FDA establishes this first.
2. Same technology, or differences that raise no new questions
The technological characteristics must be the same, or different in ways that do not raise different questions of safety and effectiveness, with data showing the device is as safe and effective.
3. Legally marketed
The predicate must be legally marketed: on the market before May 28, 1976, reclassified from Class III, found substantially equivalent through a 510(k), or granted through De Novo. Any legally marketed device may be a predicate, not only a recent one.
4. Cleared using well-established methods
FDA’s draft guidance recommends a predicate whose testing used well-established methods, such as FDA-recognized consensus standards or FDA guidance.
5. Meets or exceeds expected safety and performance
Also from the draft guidance: prefer a predicate that meets or exceeds expected safety and performance for its device type.
6. No unmitigated use-related or design-related safety issues
Check FDA safety communications and adverse event reports in MAUDE for the candidate.
7. No design-related recall
Search FDA’s recall database for the candidate and its product code.
8. Verify the K-number and read the summary
Open the record in FDA’s 510(k) database and read its 510(k) summary, which states the intended use, the technological comparison and the testing behind the decision.
9. Record the evidence
Keep the K-number, page and passage behind each comparison for your substantial equivalence table.
Try it: the Predicate Finder
Free, no account. Recent clearances like yours, live from FDA’s 510(k) database, each linked to its summary.
Ask which candidate is strongest
Describe your device; Bearnier AI compares candidates on intended use, technology and testing, with the page behind each point.
Questions about predicate search
What is the best tool to find a 510(k) predicate?
It depends on the job. FDA’s 510(k) and Product Classification databases are free and authoritative when you can search by product code and read each summary. Bearnier AI searches the same FDA records for you and cites every fact to its page. General AI assistants help with concepts but their K-numbers must be verified, and consultants add judgment for contested or novel devices.
Is FDA’s 510(k) database free?
Yes. FDA’s 510(k) database on accessdata.fda.gov is public, and FDA also offers downloadable 510(k) files. It searches record fields such as product code, applicant, device name and decision date.
Can ChatGPT find my predicate device?
It can explain what a predicate is and help you describe your device, but it does not search FDA’s record and can produce K-numbers that do not exist or belong to other devices. Check every K-number it gives you in FDA’s 510(k) database.
Can any cleared device be a predicate?
Any legally marketed device can be a predicate, FDA says, although recently cleared devices are often chosen. That includes devices found substantially equivalent through a 510(k), devices granted through De Novo, and devices on the market before May 28, 1976.
Is FDA’s predicate selection guidance final?
No. Best Practices for Selecting a Predicate Device was issued as a draft on September 7, 2023 and is not for implementation. Its four recommendations are still a useful checklist.
Sources
- Premarket Notification 510(k) · FDA, content current as of Aug 22, 2024
- Best Practices for Selecting a Predicate Device to Support a Premarket Notification [510(k)] Submission · FDA draft guidance, Sep 7, 2023, not for implementation · docket FDA-2023-D-3134
- The 510(k) Program: Evaluating Substantial Equivalence in Premarket Notifications [510(k)] · FDA final guidance, July 2014 · docket FDA-2011-D-0652
- 510(k) Premarket Notification database · FDA, accessdata.fda.gov
- Downloadable 510(k) files · FDA
- Product Classification database · FDA, accessdata.fda.gov
- Medical Device Recalls database · FDA, accessdata.fda.gov
- MAUDE: Manufacturer and User Facility Device Experience · FDA, accessdata.fda.gov
- De Novo Classification Request · FDA
- Requests for Feedback and Meetings for Medical Device Submissions: The Q-Submission Program · FDA final guidance, May 2025 · docket FDA-2018-D-1774