Chat with Set
ο»ΏChat with Set lets you analyze an entire patent dataset using natural language. Instead of reading patents one by one, you ask a question and the AI reads your dataset and answers for you.
In this section, we explain to you how this feature works and give you some best practices when using "Chat with Set".
Chat with Set allows you to ask questions about a set of patents and get insights across the entire dataset. It helps you quickly understand trends, identify relevant inventions, and explore your results without reviewing each patent individually.
Chat with Set includes two modes:
- Chat with Set β Explore
- Chat with Set β Deep Dive
These modes are designed for different levels of analysis, from quick exploration to structured, in-depth workflows.
ο»Ώο»ΏHow it worksο»Ώο»Ώ
Chat with Set works on top of your current result set. It does not change the dataset itself, but helps you analyze it through natural language queries.
Depending on the mode you choose, the system either:
- Generates a single aggregated answer (Explore), or
- Builds and maintains a structured analysis layer across all patents (Deep Dive)
The workflow is always the same:
- You ask a question
- The AI determines whether any filters need to be applied
- It retrieves the relevant patent data
- It answers your question using the patent data
ο»ΏChat with Set β Exploreο»Ώ
ο»ΏChat with Set β Explore is designed for fast, high-level understanding of your result set.
ο»ΏWhat you can doο»Ώ
- Ask broad questions about your patent set
- Get summarized insights across multiple patents
- Quickly explore themes, trends, and relationships
ο»ΏHow Explore worksο»Ώ
When you ask a question, the system analyzes the patents in your current set and generates a single, aggregated answer. This mode is optimized for speed, simplicity, and high-level exploration.
ο»ΏWhen to use Exploreο»Ώ
Use Explore when you:
- Want a quick overview of a dataset
- Are exploring a new topic
- Need directional insights before deeper analysis
ο»ΏLimitations of Exploreο»Ώ
- Answers are aggregated and not structured per patent
- Limited visibility into which patents support a conclusion
- May miss details that exist only in patent descriptions
For each response, the AI draws on the Title, Abstract, Claims, Organization, and Patent Office fields of the patents in your dataset.
At the moment, LLMs have a limited context window (think of this as its memory). This means that the LLM cannot query your entire dataset at once if it is too large. As a rule of thumb, the LLM will do the following:
- Fewer than 1,000 patent families: The AI will analyze all families in the dataset.
- More than 1,000 patent families: The AI will sample and analyze up to 1,000 families.
The above-mentioned number of families is a rule of thumb. The exact number of families analyzed depends on text length. Domains with especially long abstracts or claims will result in fewer families being processed, and vice versa.
ο»ΏChat with Set β Deep Diveο»Ώ
ο»ΏChat with Set β Deep Dive is designed for detailed, structured analysis across a set of patents. Your question is applied to each patent in your dataset (up to the first 200 for the Research subscription and 1000 for the Strategic subscription). Instead of one combined answer, you get an individual answer for every patent.
ο»ΏWhat Deep Dive doesο»Ώ
Deep Dive analyzes multiple patents at the same time and builds a Resource Table in the background β think of it as a spreadsheet that grows as you ask questions.
π‘ As you work, Deep Dive builds a table in the background β Resource Table β each row is a patent, each column is a question you've asked. This table stays available throughout your chat session and can be exported (e.g. to XLS).
You can keep refining your analysis as you go:
- Add new questions (up to 50 columns)
- Update or rephrase existing ones
- Remove columns you no longer need
All changes are applied consistently across the dataset. This makes it easy to compare patents side by side and build up a structured view of your dataset.
π‘ The Resource Table stays available throughout your conversation and evolves as you keep working.
ο»ΏHow Deep Dive worksο»Ώ
- Select Deep Dive mode
- Ask a question
- Deep Dive checks whether it can answer right away or needs to add structure to the Resource Table
- If a table update is needed, it proposes a plan β you can approve or adjust it before anything runs
- The analysis is applied across all patents
- You get both a summary answer and the detailed per-patent results in the Resource Table
π Before Deep Dive makes any changes to the Resource Table (e.g. adding or updating a column), it first shows you a plan β what it will do, what prompt it will apply, and what to expect. You can approve, adjust, or revise the plan before it runs.
ο»ΏWhen to use Deep Diveο»Ώ
Use Deep Dive when you need to go beyond a high-level overview:
- Detailed, patent-level insights
- Side-by-side comparisons across patents
- Structured outputs like matrices or feature tables
- Higher accuracy grounded in patent descriptions
π‘ Typical use cases include competitive benchmarking, freedom-to-operate (FTO) analysis, feature extraction (materials, processes, etc.), and art matrix creation.
ο»ΏUnderstanding which patents were consideredο»Ώ
After every response, you can see exactly which patents the AI used to answer your question.
π Click "Click here to see the patents that were considered" to expand the list.
This helps you verify the answer is grounded in the right data - especially useful when working with large or filtered datasets.
ο»ΏHyperlinks in answersο»Ώ
Whenever the AI mentions a specific patent in its response, it will hyperlink it directly. Click the link to open and explore that patent further.
ο»ΏUsing Filters with Chat with Setο»Ώ
- Filtering manually before you start
You can apply any filters from the right-hand panel before using Chat with Set. The AI will only analyze the patents currently visible in your filtered dataset.
Example: Filter by US patents + a specific organization first β Chat with Set will only answer based on those patents.
- Letting Chat with Set apply filters automatically
Chat with Set can also apply certain filters on its own, based on how you phrase your question. Here's what it can control:
Filter | Example Prompt |
|---|---|
Publication year | "What are the main innovation trends in the past 3 years?" |
Patent office | "Identify the main topics of invention in Chinese patents in this dataset." |
Organization | "Analyze all inventions by Samsung and explain their innovation efforts." |
Publication number | "Compare patents US-10462849-B1 and US-10383371-B2 and explain the main differences." |
Status | "Analyze all pending patents and tell me about their main topics." |
For other filters such as similarity or keyword filters - you still need to apply them manually before using Chat with Set.
ο»Ώ
Each question you ask is a fresh interaction. The AI re-queries the dataset for every question, including follow-ups. This means:
- The same question asked twice may return slightly different answers if it can be interpreted in more than one way
- Follow-up questions do not carry context forward automatically - phrase each question with full context if needed
ο»Ώο»ΏSaving Chat with Setο»Ώο»Ώ
You can save any Chat with Set conversation for future reference.
To save a chat:
- Click the Save button in the right upper corner of the chat panel
- Click the pen icon to rename the chat before saving
- A success message will confirm it's been saved - click "Open in folder" to jump straight to it
To find saved chats:
- Click "Show Insights" in your workspace
- All saved chats, comparisons, and insights are stored here linked to the datasets
- Each item is labeled by type so you can tell them apart at a glance
You can view your saved chats by clicking on "Show Insights." Here, you'll find all your saved chats, comparisons, and insights, conveniently stored in one place. Saved insights are easily distinguishable by the labels displayed beneath each item.
π‘ You can save multiple chats for the same dataset. They'll all be accessible from that dataset's Insights panel.
ο»ΏSuggested use casesο»Ώ
Not sure where to start? Here are some prompts to inspire you:
ο»ΏTrend Analysisο»Ώ
You could ask:
- What are the latest trends in this domain?
- What are the main trends in this domain over the past 5 years?
- What are the topics being discussed in this dataset and how many inventions belong to each topic?
- What are the patent numbers that relate to [your topic]? List the publication numbers in a comma-separated list.
ο»Ώο»ΏOrganization and Portfolio Analysisο»Ώο»Ώ
- What are the top 3 startups in this domain? Provide a summary of their most important inventions.
- Compare the portfolio's of Organization X and Organization Y and explain the main differences to me.
ο»ΏComparing Inventions and Tracking Developmentο»Ώ
- What are the differences between patent X and patent Y?
- How has this technological domain developed? Compare 5 years ago with today.
For prompt examples, and prompting tricks, read Prompting like a ProPrompting like a Pro.