A patent search is the work you do before filing an application to figure out whether your invention is actually new. The US Patent and Trademark Office (USPTO) calls the underlying concept a prior art search, and applicants are legally required to disclose any relevant prior art they know about on an Information Disclosure Statement (IDS). Skip the search and you risk a quick rejection, a wasted filing fee, or, worse, an unenforceable patent if material prior art surfaces later.
The mechanics of how to run a patent search changed substantially after the original version of this guide was written. The USPTO retired its old PatFT and AppFT databases on September 30, 2022, and replaced them with the Patent Public Search tool. The US Patent Classification (USPC) was retired for utility patents in 2013 and replaced by the Cooperative Patent Classification (CPC), jointly maintained by the USPTO and the European Patent Office. AI-assisted search platforms have moved from a curiosity to a standard option in the toolkit. The steps below are the same ones a professional searcher would walk through, updated for how the modern tools actually work.
Why a Patent Search Matters
Different goals call for different searches. Most do-it-yourself searches focus on the first one, but it helps to know what the others look like:
- Novelty / patentability search. Is this invention new and non-obvious enough to be worth filing? This is the search you run before drafting an application.
- Freedom to operate (FTO). Can I sell this product without infringing someone else’s existing patent? Different question, different scope — focuses on unexpired claims, not just any prior disclosure.
- State of the art. What does the field look like? Often used by R&D teams to scope a project or by investors during diligence.
- Validity / invalidity. Is a specific issued patent actually valid? Run by litigators looking to invalidate a patent being asserted against them, or by licensees deciding whether to take a license.
- Infringement. Does product X infringe patent Y? A claim-by-claim comparison, usually done by a patent attorney.
The steps below describe a novelty/patentability search. The same techniques apply to the other types — what changes is the database scope and how strictly you read the claims.
Step 1: Define the Invention Precisely
Before opening any database, write a one-page description of the invention that covers:
- What it does (the function)
- How it does it (the mechanism)
- What it is made of or composed of (materials, components, structure)
- What problem it solves and what it improves over the existing solution
- The patent type you would file: utility (most inventions — articles, machines, processes, compositions of matter), design (ornamental appearance), or plant (asexually reproduced new plant varieties)
This document does double duty: it disciplines your thinking, and it later becomes the seed for your application. If you cannot write the page, you cannot search well — go back and refine the invention until you can.
Step 2: Brainstorm Keywords and Synonyms
Patent attorneys and the engineers who wrote prior patents on the same idea may have used very different words for it. The earmuff that you call an “earmuff” might be filed as a “thermal ear covering” or “auditory protection device.” For each major aspect of the invention, list:
- The plain-language term
- Two or three technical synonyms
- Two or three layperson synonyms
- The verb form (“insulating” as well as “insulation”)
- Functional descriptors (“thermally protective”)
Patent search is fundamentally an exercise in vocabulary. A keyword list of 15-30 terms is normal for a small invention, and 50+ for anything more complex.
Step 3: Find Relevant Classifications
Classifications are how patent offices group inventions by subject. They are far more reliable than keyword search because they survive translation, terminology drift, and patentee creativity. Two systems matter in 2026:
- Cooperative Patent Classification (CPC). The current standard, used by the USPTO, EPO, and most major patent offices. CPC codes look like
F25D 23/02(a refrigerator section) — Section / Class / Subclass / Group / Subgroup. - US Patent Classification (USPC). Retired for utility patents in 2013 but still used for US design patents and plant patents. USPC codes look like
2/209.
To find your CPC codes:
- Go to the USPTO classification portal or Espacenet’s CPC browser.
- Search the scheme with one of your strong keywords.
- Read the class definition to confirm the scope matches your invention.
- Note the symbol and any closely related neighboring classes — your invention often straddles two or three.
Plan to spend an hour on classification alone. A clean set of three or four CPC codes saves you from running 20 sloppy keyword searches that miss the relevant art.
Step 4: Run Searches in the Right Databases
No single database covers everything. A reasonable novelty search hits at least three of these:
USPTO Patent Public Search (PPS)
The current free USPTO search at ppubs.uspto.gov replaced the old PatFT/AppFT system in 2022. It supports both a Basic mode (Google-style keyword search) and an Advanced mode (Boolean queries with field codes). PPS uses different field tags than the legacy system — .spec. for specification, .ccls. for current classification, .aclm. for claims, and so on. The Help screens inside PPS list every field code; print the cheat sheet once and keep it next to your screen.
Espacenet (European Patent Office)
Espacenet is the EPO’s free global database, covering more than 150 million patent documents from 100+ countries. It includes machine-translated abstracts and full text where available. Espacenet’s smart search and classification browser are excellent for finding the CPC codes that PPS will then let you exhaustively query.
PATENTSCOPE (WIPO)
PATENTSCOPE is the World Intellectual Property Organization’s database. It is the canonical source for PCT (Patent Cooperation Treaty) international applications, and it covers national filings from many smaller jurisdictions that do not appear in Espacenet’s coverage.
Google Patents
Google Patents indexes more than 120 million patent documents and links each one to cited non-patent literature, machine translations, and a Prior Art Finder feature that suggests potentially relevant references. It is the easiest tool to start with and the worst tool to finish with: do not rely on it as your only search.
Lens.org
Lens.org is a free, well-funded global patent + scholarly literature platform from Cambia. It is particularly strong for combined patent + non-patent literature searches, citation graphs, and assignee analysis. Many universities and small firms use Lens.org as their primary search platform.
FreePatentsOnline
FreePatentsOnline still exists and offers a usable search interface plus PDF downloads, but its data is largely a subset of what is in PPS, Espacenet, or Google Patents. Useful as a backup, not as a primary tool.
Step 5: Read the Claims, Not Just the Abstract
The single most common amateur mistake is reading the title and abstract of a hit and moving on. The legal scope of a patent is defined by its claims, not the abstract or the figures. A patent might be titled “thermal beverage container” and look superficially close to your invention, but its claims may be narrowly limited to a specific double-walled stainless construction that your design avoids.
For each potentially relevant hit:
- Read the independent claims (claim 1, plus any other independent claim — usually claims that do not refer back to a previous claim).
- Compare element by element against your invention. The all-elements rule means a single claim infringes only if every element is present.
- Check the patent’s status (active, expired, abandoned). Expired patents do not block new patents but can still count as prior art.
- Note the assignee, inventor, and CPC codes for further searches.
Step 6: Citation Analysis
Once you have a single solid hit, citations are the fastest way to expand the search:
- Backward citations (“references cited”): the prior art the examiner used to evaluate the patent. Reading these tells you the prior-art landscape the patent was carved out of.
- Forward citations (“cited by”): later patents that referenced this one. These often describe variants, improvements, and competing approaches, and they are usually impossible to find with keywords alone.
Espacenet, Google Patents, and Lens.org all surface forward and backward citations cleanly. Plan to follow each major hit one or two citation hops outward.
Step 7: Document Everything
Keep a running search log with: every database used, every query string run, the date, the number of results, and a one-line note on the most relevant hits. This serves three purposes:
- It prevents you from running the same search twice.
- It feeds directly into the Information Disclosure Statement (IDS), which the USPTO requires within three months of filing or patent issuance for any prior art the applicant or attorney is aware of.
- If you ever need to defend the patent, the search log is part of the prosecution history.
AI-Assisted Patent Search in 2026
The last three years have produced a mature commercial market for AI-assisted patent search. The pattern is the same across vendors: paste your invention disclosure (or upload an early draft application), the system uses semantic embeddings to retrieve patents and non-patent literature, and the output is a ranked list with relevance scoring. Notable platforms include:
- IPRally — graph-based semantic search aimed at search professionals
- Patsnap Eureka — large commercial platform with AI assistant features
- IPlytics (LexisNexis) — analytics-focused, especially strong on standard-essential patents
- Questel Orbit Intelligence — established commercial platform with AI-augmented modes
- The Lens AI features — built into the free Lens.org platform
- Google Patents Prior Art Finder — free, less polished, but a good first pass
None of these replaces a careful classification + claims-reading workflow. They are good at retrieval (finding things you would not have found with keywords) and bad at judgment (whether what they found actually anticipates your claims). Use them as additional eyes, not as a substitute for the human ones.
When to Hire a Professional
A do-it-yourself search is appropriate when you want a quick read on whether your idea is obviously already taken, or when you want to refine the invention before incurring attorney fees. Hire a registered patent attorney or agent (or a dedicated search firm) when:
- The novelty search is going to lead to an actual filing decision involving real money.
- You need a freedom-to-operate opinion for a product going to market.
- The technology field is dense, fast-moving, or international.
- Litigation is foreseeable.
Professional novelty searches typically run $1,000-$3,000 in 2026 in the US; FTO and invalidity searches run higher. The cost is small relative to a $10,000-$25,000 utility-patent prosecution and dwarfed by the cost of an unenforceable patent or an inadvertent infringement.
Decide Whether to File
After the search, you have three reasonable outcomes:
- Clean. No close prior art. Move on to drafting the application — most likely a provisional first to lock in a priority date, with a non-provisional within 12 months.
- Close prior art exists, but your invention is distinguishable. Refine the claims around what is actually new and non-obvious in light of the prior art, and file with full disclosure.
- Already done. Stop, save the filing fee, and either pivot the invention or move on.
Remember the AIA first-inventor-to-file rule (effective March 16, 2013): the priority date is what matters, not when you first conceived the idea. If you are confident about novelty, file the provisional sooner rather than later.
FAQ
Is a patent search legally required? The search itself is not, but the duty to disclose known prior art is. The USPTO requires applicants to file an Information Disclosure Statement listing material prior art they are aware of. Failure to disclose can render the patent unenforceable and expose the applicant to charges of inequitable conduct.
How long should a do-it-yourself patent search take? A reasonable amateur novelty search takes 8-15 hours: 1-2 hours on definition and keywords, 2-3 hours on classification, 4-8 hours on database searches and reading hits, 1-2 hours on documentation. Less than that and you have probably missed something.
Can I just rely on Google Patents? No. Google Patents is excellent as a first pass and a citation explorer, but its keyword interface and ranking can miss patents that are clearly relevant under classification search. Combine it with PPS, Espacenet, and (ideally) Lens.org.
Do AI patent search tools actually work? Yes for retrieval, no for judgment. The semantic embedding-based platforms are good at surfacing patents that share concepts but use different vocabulary. They still require a human to read the claims and decide whether something genuinely anticipates the invention.
What is the difference between a provisional and non-provisional patent application? A provisional application establishes a priority date, costs less to file, does not get examined, and expires after 12 months. A non-provisional is the actual application that is examined and that, if granted, becomes a patent. Most inventors file provisional first and convert within 12 months.
How much does it cost to obtain a US utility patent in 2026? Filing fees alone for small entities are roughly $700-$1,000. Total prosecution including attorney fees typically runs $10,000-$25,000 over 18-36 months. Maintenance fees over the patent’s 20-year term add another $13,000+.
Bottom Line
The mechanics of patent search have changed in the last decade — PatFT and USPC are gone, CPC and Patent Public Search are the current standards, AI-assisted search is real and useful. The fundamentals have not. Define the invention, build a real keyword list, find the right CPC codes, search several complementary databases, read the claims, follow citations, and document everything. Do that, and you will know whether to spend the next $10,000 on a filing or the next month on a different idea.
Patent search is one of those tasks where the cheapest hour you spend up front saves the most expensive month later.