When we search for something, we expect accurate results quickly. Usually we look at the top ten or so, and move on.
Technical questions are never that easy. Problems exist and solutions to those problems exist, but the solutions are all slightly different from one another. So when you search for a problem, or for a solution to it, you do not get one answer. You get many, and they are all near neighbours. That leaves the researcher combing through hundreds of documents before reaching the few of interest. If you have experienced this, you are not alone. If your job is searching for technical information, you already know how much time goes into that exercise.
How do the current tools compare?
An independent benchmark published in BASES in January 2026 suggests this is not a skill problem, it is a tooling problem. François Libmann tested fourteen AI-assisted patent search tools on a single query, the use of ultrasound to accelerate whisky ageing, and manually verified every result against Orbit Intelligence.
The best tool returned a 46.9% relevance rate. The worst returned 6.7%. Nine of the fourteen fell below 40%.
More telling is what happened when he compared the two approaches. Pooling every relevant result from all fourteen AI tools gave 68 patent families. A carefully built Boolean strategy, using terms drawn from those same results, returned 86 hits of which 41 were relevant. Only 33 families appeared in both. Thirty-five were found by the AI tools alone, and fifteen only by the Boolean search.
His conclusion is that the two are complementary rather than competing. Ours is narrower and harder to argue with: on one simple query, in 2026, no single tool of either kind reliably gets you to the prior art that matters.
François Libmann, "IA générative et recherche brevets : ce que révèlent 14 outils analysés", BASES no. 443, January 2026, pp. 1–6.
Which leaves limited choices. Either learn to think in patent jargon, translating your idea into classification codes, Boolean strings and synonym chains, then processing the results by hand. Or hire a skilled professional to do it for you. Both cost time and money.
What we built instead
Imagine pointing a query at a Strata technology vault and getting answers back. Answers that help you learn the technology faster. Answers that show you how to work around the prior art. Answers that tell you whether something is an opportunity worth researching.
Traditional tools hand you a list of documents and expect you to resolve it. Strata gives you the answer, in a form you can act on.
An example
Say you want to know Airbus's filing strategy on fuel cells. You get something like this.
The core problem: fuel cells are too heavy for aircraft. Every Airbus patent circles back to the same constraint, gravimetric power density.
Three distinct approaches they are pursuing:
- Tubular and ring-shaped architectures (EP4366002B1, US20250273721A1)
- Eliminating interconnectors entirely (EP4340081A1, EP4641713A1)
- Manufacturing for thin, uniform layers (EP4525101A1, EP4531144A1)
That answer triggers the next questions. If they are not using interconnectors, what are they using instead? Who else is taking the same approach? Now you know this is worth pursuing. Or perhaps your own company is working on an interconnector-free design, and this is something to build on.
Notice what did not happen there. Nobody read a hundred documents, tallied assignees in a spreadsheet, or guessed at classification codes. The question went in and a position came out, with the records behind it, in a form that immediately suggests the next question. That is the whole difference between a list and an answer.
If you are tired of overpaying for patent search and want to see a different approach, we would be glad to work with you. We built Strata IP for people who work with technical information, and we are here to help you get to answers quickly.