A bad patent search is worse than no patent search. No search leaves you appropriately nervous. A bad one leaves you confident, and confidence is what makes people spend twenty thousand dollars on a filing, order tooling, and sign a distribution deal for a product that a competitor's issued patent already covers. The failure almost never looks like failure at the time — you typed some words into a database, got a manageable number of results, read a few, and concluded that nothing matched.
Here are the specific ways that happens, in rough order of how much damage each one does.
1. Searching only in your own vocabulary
You call it a collapsible dog bowl. The patent that blocks you calls it a "portable animal feeding receptacle having a flexible sidewall movable between an extended and a collapsed configuration." Patent attorneys deliberately avoid marketing language, because a claim written in commercial vocabulary is easy to design around. The result is a systematic mismatch between how inventors search and how patents are written.
The fix is to build a synonym set before you search: the engineering term, the academic term, the term a chemist or mechanical engineer would use, the older term from decades ago, and the deliberately generic term an attorney would choose. Then search each one. Our guide to choosing keywords for a patent search covers how to generate that set methodically.
2. Never touching classification
This is the single biggest difference between an amateur search and a professional one. Every patent is assigned Cooperative Patent Classification codes describing its technical subject matter, assigned by people, independent of the words the drafter chose. Classification is how you find the patent that describes your invention perfectly using none of your terms.
The technique is straightforward: find one or two patents that are close to your invention by keyword, look at their CPC codes, then browse everything in those subclasses. It is slower than keyword searching and it is where the results that actually matter tend to surface. See searching patents by CPC and IPC classification.
3. Searching granted US patents only
Three separate omissions hide inside this one habit.
- Published applications. Applications publish at 18 months whether or not they are ever granted, and a published application is prior art against you. Many free tools default to granted patents, so make sure yours is not.
- Foreign patent documents. A Japanese utility model or a German patent published decades ago is prior art in the US even though it was never filed here and never translated. Machine translation is good enough to triage, and databases from other offices are free.
- Non-patent literature. Journal papers, conference proceedings, product manuals, catalogs, crowdfunding pages, forum posts, and videos all count. An examiner may not find them; a competitor's attorney trying to invalidate your patent absolutely will.
Starting with the USPTO's own full-text database is right, but treat it as one source rather than the whole picture.
4. Reading the abstract and stopping
Abstracts are marketing for the invention as a whole. The claims are the legal boundary, and they are frequently much narrower — or much broader — than the abstract suggests. A patent whose abstract sounds identical to your invention may have claims restricted to one specific configuration that you do not use. A patent whose abstract sounds unrelated may have an independent claim broad enough to cover you.
Read claim 1, in full, of anything remotely close. Then read the file history if the stakes are high. Our step-by-step guide to reading a patent covers the order to read the sections in and what each one is actually for.
5. Confusing two completely different questions
"Can I patent this?" and "Can I sell this?" are different searches with different scopes.
| Patentability search | Freedom to operate | |
|---|---|---|
| Question | Is my invention new and non-obvious? | Does my product infringe a live patent? |
| Scope | Everything published anywhere, ever | In-force claims in the countries where you sell |
| What matters | Disclosure of any kind | Claim language and legal status only |
| Expired patents | Still count against you | Irrelevant — free to use |
A clean patentability search says nothing about whether you can launch. Plenty of patentable improvements infringe a broader patent on the underlying device. The distinction is laid out in patentability search vs freedom to operate.
6. Ignoring legal status
A granted patent means nothing if it lapsed for unpaid maintenance fees, expired at the end of its term, or had its claims cancelled in a post-grant proceeding. Conversely, a patent that looks safely old may have a live continuation in the same family with claims still being written. Before you either panic about a patent or dismiss one, check whether it is actually in force in the country that matters — see how to check if a patent is still in force.
7. Getting the date logic wrong
Two date errors recur. The first is comparing against publication dates instead of priority dates: a reference published after you filed can still be prior art if its own priority date is earlier. The second is forgetting the blind window — applications filed in the last 18 months are not published, so no search of any database can see them. Your search is always a year and a half out of date — a limitation to plan around, and one more reason to file rather than wait.
8. Searching once and never again
A search is a photograph, not a subscription. New applications publish every week, and a competitor may have filed on your exact concept the month before you looked. For anything you are investing in seriously, set standing alerts on the relevant classification codes and on your main competitors' names, as described in setting up patent monitoring alerts. Ten minutes a month beats a surprise cease and desist letter.
9. Searching in order to be reassured
This one is psychological and it is the most common of all. You have been building this for a year. You want the search to come back empty, so you search narrowly, you stop early when nothing appears, and when something close does appear you find reasons why it is different. Every inventor does some version of this.
The countermeasure is to invert the exercise. Set out deliberately to destroy your own invention: your job for the next two hours is to find the reference that kills it. Note every hit that is arguably close, including the uncomfortable ones, and write down for each exactly which element of your invention it is missing. If you cannot articulate the missing element in one sentence, you have found a problem, not a false alarm. And when you do find something close, that is a starting point rather than an ending — a similar patent has several possible answers, most of them survivable.
When to stop doing it yourself
A careful self-search is genuinely worth doing: it costs time rather than money and teaches you the landscape you compete in. What it cannot do is give you a defensible opinion. Before tooling money, investor money, or a launch commitment, a professional searcher with paid databases and a classification specialist's instincts is cheap insurance — the tradeoff is quantified in DIY patent search vs hiring a professional.
Projects House is a product development engineering firm, not a law firm. Where we add value in this process is translating between engineering and patent vocabulary — working out what your invention's real technical distinguishing features are so the search targets them, and assessing whether a blocking patent can be designed around. Legal opinions on patentability and infringement belong with a registered patent attorney. To get an engineering read before you commit to a filing, use the contact form.