Verifying a Company’s Claim of 98 Artificial Intelligence Patents
A clear company statement that it owns 98 patents in artificial intelligence refers to a count of distinct intellectual property records linked to AI techniques, systems, or applications. That number can mean different things depending on how patents are counted, where they were filed, and whether they are granted or still pending. Below are the main points to check: what patent counts include, which databases to consult, how ownership and assignments affect the total, the gap between applications and granted patents, the role of filing dates and jurisdictions, and how to judge relevance to AI work.
What a “98 patents” count often represents
A patent count can come from raw search hits, family-group totals, or corporate reporting. One company might count every published application and granted patent linked to AI terms. Another might count only granted patents in a single country. Patent family counting groups documents that share a common priority date. A granted patent is a right issued by an office after examination; a published application is an earlier public filing that may never mature into a grant. The simple number alone does not say how broad, enforceable, or commercially useful the patents are.
Where to check: common patent databases and search methods
Use multiple public sources to cross-check a 98-patent claim. Different services index different fields and update on different schedules. Searches usually start with company names, inventor names, patent classifications related to AI, and focused keyword phrases. Combining database searches with company assignment records gives a clearer picture than a single query.
| Database | Coverage | Useful for |
|---|---|---|
| USPTO | U.S. patents and published applications | Official ownership records and grant status |
| European Patent Office (Espacenet) | European and global patent families | Cross-jurisdiction family views and classifications |
| WIPO PATENTSCOPE | International filings under the international system | Published PCT applications and global reach |
| Google Patents | Broad, easy search across jurisdictions | Quick keyword searches and linkages to family members |
| Lens.org | Integrated scholarly and patent data | Citation analysis and exportable data for analytics |
How ownership and assignments change a count
Patent records list the original filer and may later show recorded transfers. Corporations frequently move patents between subsidiaries or to holding companies. Public filings and assignment registries indicate current legal ownership only when the transfer has been recorded with the relevant patent office. Counting patents under a company name requires checking aliases, parent and subsidiary names, and recorded assignments to avoid double counting or missing transfers.
Granted patents versus published applications
A granted patent is the result of examination and confers enforceable rights in the jurisdiction that issued it. A published application is publicly visible but not yet enforceable; it may be refused, withdrawn, or later granted. Some companies report the total of both for headline numbers. For many assessments, distinguishing grants from applications is essential because only grants represent confirmed patent rights.
Timing and jurisdiction: filing dates and territory coverage
Patents are territorial. A U.S. patent applies in the United States; a European grant applies in nominated member states. The filing date and the earlier priority filing influence what is covered and when protection begins. Patent families can include many national filings from the same root filing. A single invention patented in ten countries can appear as ten patents in a count. That multiplies the apparent size of a portfolio without increasing the number of distinct inventions.
Interpreting relevance to artificial intelligence technology
Relevance is judged by claim scope, technology class, and citations. Look for independent claim language that describes core AI operations, such as model training methods or data processing pipelines. Classification codes tied to machine learning or neural networks can help filter results. Citation patterns and family breadth hint at technical influence. A portfolio of narrow, incremental patents tells a different story than a few broad patents covering foundational techniques.
Practical constraints and trade-offs
Data sources differ in update frequency and search logic. One database may show 98 records; another may show a different number because of how it treats continuations, family members, or expired documents. Not all transfers are promptly recorded. Jurisdictions have different publication rules and claim formats, which can change how results map across systems. Access to raw file histories and translations may be limited. Deep relevance assessment requires reading claims and prosecution histories, which takes time and subject matter knowledge. Finally, quantity is not the same as quality. A count is a starting point, not a valuation.
How to run a patent search for AI
When to use patent analytics services
How patent ownership affects valuation
Next steps for verification and further research
Start by compiling search queries in at least two major databases and by checking assignment records where available. Map patent families to see whether 98 items represent distinct inventions or many national filings of fewer inventions. Flag granted patents versus published applications, and note jurisdictions and filing dates. For relevance, read representative claims and examine classifications and citations. If deeper analysis is needed, export searchable datasets for clustering and citation mapping or consider a professional patent analytics report that includes prosecution history. Each of these steps sharpens confidence in what the 98 figure actually reflects.
Legal Disclaimer: This article provides general information only and is not legal advice. Legal matters should be discussed with a licensed attorney who can consider specific facts and local laws.
