Reinventing Patent Search with AI
October 2025 - Traditional patent searches typically take 4 to 8 hours. There are two main reasons for this:
Keyword uncertainty. Patent attorneys often describe inventions using highly specific — and sometimes intentionally obscure — terminology to protect novelty. As a result, the first few hours are spent identifying a handful of relevant patents just to understand the right wording and classifications.
Comprehensive recall. Once the right terminology is found, the next phase involves ensuring that no relevant patents are missed which can be equally time-consuming.
As the number of patent filings continues to grow, speed and accuracy have become critical.
With the help of AI, this process can now be reduced from hours to minutes — without compromising quality.
Phase 1: Precision. AI models like ChatGPT can instantly identify one or two highly relevant “seed” patents that capture the core invention. This reduces the keyword discovery phase from hours to seconds.
Phase 2: Recall. Tools such as Ambercite AI can then expand from these seeds to map the entire landscape of related patents with exceptional precision — often in a single iteration.
The result: a professional-grade patent search in minutes.
Finally, the dataset exported from Ambercite can be fed back into ChatGPT or similar tools for advanced analytics — such as clustering, inventor insights, or trend statistics — transforming how IP professionals work.
What sets Ambercite apart from other search engines? Unlike traditional tools that rely heavily on keyword matching, Ambercite uses a sophisticated computational AI to analyze patent citation networks. It evaluates two types of citations: known citations (directly linked to the seed patent) and unknown citations (indirectly connected patents). Ambercite’s AI calculates the probability that two patents are similar by assessing factors such as citation direction, centrality, structure, and more. This approach helps uncover relevant patents even when they use different terminology.
Ambercite Use Case: Patentability search for the umbrella folding mechanism
We previously described a search strategy in our article - Patent Olympiad 2024 Search Strategy Case. Let’s now see if the same case can be addressed just as effectively by starting with ChatGPT.
Description of the Innovation for conducting a Novelty Search :
This innovation presents a lightweight, foldable umbrella made from durable stainless steel. Featuring a telescopic design, the umbrella is optimized for portability through miniaturization. Its collapsible structure allows for easy opening and closing, offering both convenience and space-saving benefits. Suitable for both rain protection and as a sunshade, this umbrella combines functionality with a sleek design, making it ideal for on-the-go use.
Part 1: Use ChatGPT or similar AI tools to identify one or two highly relevant seed patents as a starting point.
Part 2: Transition to Ambercite for Conducting a Novelty Search
Step 1: Use the Seed Patents Identified in Part 1 to Initiate Your Search in Ambercite
Input Key Patents: Take the most relevant patent identified and input them into Ambercite. Show 100 results and click on “Find Similar Patents”
Step 2: Analyze the Results and Identify Additional Relevant Seed Patents
Understanding Ranking and Scores in Ambercite
In Ambercite, two key scores help evaluate patents: the Similarity Score and the AmberScore.
The Similarity Score reflects how closely a patent relates to your seed patents. Ambercite’s AI calculates this score by running a probability analysis based on various criteria such as citation networks, citation direction, centrality, and the structure of citation relationships between patents.
The AmberScore, on the other hand, indicates the patent's personal value within the network, reflecting its popularity and significance. This score shows how influential or valuable the patent is in the broader context of the citation network.
By analyzing both scores, you can select the most relevant patents to reinforce your seed patents. Up to 200 additional patents can be included based on their similarity and AmberScores, helping to create a well-rounded and effective search strategy.
Explore Similar Patents: Ambercite will suggest similar patents within the network that may be relevant.
Identify Influential Patents: Focus on patents that are highly connected within the network and closer to your invention. Click on like to organize them into a folder.
To facilitate analysis, consider highlighting key terms : fold* umbrella open* telescopic
The patent results will then appear in a list.
You can click on the description to open the summary panel and quickly review each one to identify the most relevant hits you'd like to keep. Once open, you can navigate from one invention to another and mark the patents that are relevant to your search by clicking the "like" button.
All the patents you've liked will be saved in the "Liked" folder.
For further analysis, you can open the full text in Google Patents or Espacenet.
Step 3: Promote patents that are highly relevant and specific to the search criteria.
By clicking on the title or description of a patent, the summary panel will open. From there, you can review each result individually and click the "Promote" button to include the patent as a seed patent in your next search. For more detailed information, such as the full text of the patent, click on the "Google Patent" or "Espacenet" button to access the complete document and make more informed decisions.
Once promoted, these patents will be highlighted in the input field. Click on "Find Similar Patents" to refine the search with additional inputs, enhancing both recall and precision. By exploring the citation network of these patents, you'll gain a more comprehensive view compared to relying on a single patent. The AI will then be able to run more accurate probability analyses on the similarity.
You can then review the patents in the "New Patents" category, which includes only the newly discovered patents since the previous search. This allows you to focus on new findings without reanalyzing patents you’ve already reviewed. In our case, new patents have been discovered.
Benefits of Iterative Searching with Ambercite
Ambercite excels at finding relevant patents where keywords and semantic searching fails because of imprecise language, and class codes are too inconsistent. Providing you have a reasonable starting patent, you can iterate your way towards the best patents. The “New Patents” button makes this process easier than ever before.
How can I start using Ambercite?
You can try Ambercite yourself in our no-cost trial version, found here:
However, to run these sorts of analyses discussed in our blog, you will need to contact us about a corporate subscription. If you do contact us, we would be very happy to provide an online demonstration of the above approach on one of your subject.