What is the Invalidity Search Agent?
The Invalidity Search Agent carries out a full patent invalidity search workflow end-to-end, using AI in every stage.
You give the Agent the patent you want to invalidate. The Agent reads the patent’s claims, extracts their key features, searches for prior art published before the patent’s effective date, reviews each discovered document for the presence of those features, and ranks the documents by relevance. The results are presented as an interactive Feature Chart, and the Agent can generate a full Written Report with novelty and inventive step analysis on demand.
The Agent automatically creates and populates a Project that you can continue to work on manually: analyze more documents, bring in documents you have found yourself, adjust relevance rankings, add comments, and export the results.
How to access the Invalidity Search Agent
To use the Invalidity Search Agent, your organization admin must
ensure that your organization has the Generative AI features (see Enabling AI Assistant Features) enabled, and that you are on Unlimited plan, and
ensure that the Invalidity Agent permission has been granted to your user account.
Your organization must also have an active Agent subscription or trial, or your user account must have free Agent runs left. The free quota is replenished on a monthly basis.
Once enabled, you can launch the Agent from the IPRally home page by selecting Agent mode (vs Standard mode) on the Home screen, and then selecting the Invalidity search type from the sidebar.
Launching an Agent search
To start a new Invalidity Search Agent:
Select the Invalidity search type from the home page sidebar. You can also start from anywhere in the app by opening the New menu and selecting Invalidity Search under the Agent section.
Enter the publication number of the patent you want to invalidate. The Agent fetches the patent and prepares its claims for review.
Choose which claims to target:
All claims — uses every claim of the publication for the invalidity search. Ideal when the whole subject matter of the patent is under attack.
Specific claims — pick the independent claims you want to invalidate. Expand Including Claims… under each independent claim to see its dependent claims, and deselect individual dependent claims to leave them out of the results and the report.
If the AI has classified a claim incorrectly, you can correct the claim structure with Mark as dependent / Mark as independent before launching.
Open Options to set the prior-art cut-off date and the level of automation (see below).
Click Launch Agent to start the automated workflow.
The Agent will begin processing and you can follow its progress in real time.
Choosing the prior-art cut-off date
For an invalidity search, only documents published before the target patent’s effective date count as prior art. The Options popover contains a Prior art published before date field that controls this cut-off: the Agent only searches for prior art published before this date. You can fill it from one of the target’s dates, or type a custom date:
Priority date (pre-selected) — fills the cut-off with the publication’s earliest priority date. This is the standard cut-off for an invalidity search and the default. If the publication has no priority claim, the application date is used instead.
Application date — fills the cut-off with the publication’s application (filing) date. Use this broader cut-off when the publication’s priority claim is expected to be contested.
A custom date — type any date directly into the field, for example when the effective date is disputed or you want to reproduce an earlier search situation.
The chosen cut-off is applied to every search the Agent makes, and the Written Report records the date and what it was based on — the priority date, the application date, or a user-provided date. Note that the Agent takes the priority date as given — it does not assess whether the priority claim is actually valid.
Reviewing features before the Agent runs
By default, the Agent runs fully automatically. If you want more control over the analysis, turn on Review features before agent runs in the Options popover. The Agent will then pause after extracting the key features from the claims, and let you review, edit, add, or delete features before the search and review begin. This is the most effective way to steer what exactly the Agent looks for in the prior art.
The Agent workflow
Once launched, the Agent executes the following steps automatically. In the user interface they are shown as expandable accordions with detailed information on each step, and you can monitor the progress as each step completes.
Step 1: Target Patent Submitted
Step 2: Search Target
Step 3: Features
Step 4: Search
Step 5: Review
Step 6: Relevance Evaluation
Step 7: Written Report
Step 1: Target Patent Submitted
Shows the target patent you submitted and the prior-art cut-off date the Agent resolved for it.
Step 2: Search Target
The search target is the target patent’s own claim set, exactly as you prepared it in the launch form — including any claim structure corrections you made there. The claims are referred to as C1, C2, … and the independent claims you selected are the invalidation targets.
Unlike a novelty search, the Agent does not need to invent a search target: the granted claims define it.
Step 3: Features
The Agent breaks each targeted claim into its key distinguishing features — the specific technical elements that will be checked against every prior art document. Each claim yields at least one feature. The Features are numbered per claim — F1.1, F1.2, … for the features of claim 1, F2.1 for claim 2, and so on — in the Feature Chart and the Written Report.
If you enabled Review features before agent runs, the workflow pauses here: review the generated features, edit them if needed, and click Approve to continue.
Step 4: Search
The Agent builds a broad pool of prior art candidates from several complementary sources, all limited to documents published before the cut-off date:
AI claim searches — IPRally’s AI search run on the claims: one search over the full claim set, plus one per claim chain, so that every claim combination is covered.
Number searches with the target patent’s family members — each family member of the target is used as a search seed of its own.
Examiner citations of the target patent — the documents cited against the target during prosecution.
The pool is de-duplicated by patent family, and the step shows how many documents each source contributed.
Step 5: Review
The Agent analyzes the pooled documents in batches of 100, checking each document for the presence of every key feature. The first batch — the most promising documents — is analyzed automatically as part of the run.
Documents that turn out highly relevant expand the pool further: their own citations are fetched and added as new candidates for later batches.
After the first batch you decide how far to go: analyze the next batch from the Feature Chart (see below), or stop when you have enough evidence. The review finishes automatically once every document in the pool has been analyzed.
Step 6: Relevance Evaluation
Based on the feature review, each analyzed document is assigned a relevance tier:
High — the document discloses all key features of at least one of the targeted independent claims. A potential anticipation of that claim.
Medium — the document comes close on at least one targeted claim (features found or unclear).
Low — the document discloses some features, but the overall match is weak.
Documents below these bars are marked as having no relevance.
Note: Relevance is always evaluated per independent claim — a document that fully anticipates one targeted claim is ranked High even if it misses the others. Dependent claims are reviewed informationally within their independent claim’s results — they never change a document’s tier. The per-claim feature presence is shown in the Feature Chart.
Step 7: Written Report
The Written Report is generated on demand: click Generate on this step (or Generate Report on the Written Report tab) whenever you want a report of the findings so far. You do not have to analyze the whole pool first — an early report can be generated between review batches.
Viewing results
Once the first batch of documents has been analyzed, you can access the results in the tabs next to the Agent Workflow tab.
Feature Chart tab
The Feature Chart is a table view of which features were found in which documents, giving you a clear overview of feature coverage across the prior art.
The Feature Chart has
A Full view and a Compact view
Claims and Features on the horizontal axis
Full texts can be seen by hovering over them
The documents found, sorted by relevance, on the vertical axis
Clicking the document’s title opens it in a new browser tab
A Relevance column with the Agent’s tier (High / Medium / Low) — you can override it manually at any time
A Vs. Target Priority Date column, unique to invalidity searches, telling you how each document relates to the target’s priority date:
Before — published before the target’s priority date; fully usable as prior art for both novelty and inventive step arguments.
After — published after the target’s priority date. For US target patents such documents can still be used for both novelty and inventive step; for other jurisdictions they can support a novelty challenge but not inventive step arguments.
Unknown — the relationship could not be determined; verify manually before relying on the document in legal arguments.
A Source column showing how each document was found: AI search N, Number search N, Cited (target), Cited (relevant doc), or User for documents you added yourself
A Batch column showing in which review batch the document was analyzed
An interactive Present/Unclear/Absent matrix
Hover on a cell: see the AI’s reasoning for an individual Feature/Document pair
Click on a row: see the reasoning for all Features for that Document
Hover on a paragraph citation: see the original text of the citation
Click a paragraph citation: open the document in highlighted view
Ability to add Comments
Continuing the review from the Feature Chart
The Analyze controls in the Feature Chart header drive the batched review:
Analyze Next Batch (N) — analyzes the next batch of up to 100 unanalyzed documents from the candidate pool, most promising first. Documents you have added yourself are analyzed first.
Analyze Selected (N) — if you select unanalyzed documents in the chart, exactly those documents are analyzed in the next batch.
Already-analyzed documents are never re-analyzed, so continuing the review only consumes analysis effort on new documents.
Exporting
The Feature Chart, including full review contents and paragraph citations, is exportable as a whole or limited to selected documents, relevance levels, features and fields. Invalidity searches offer three export targets:
Feature chart — XLSX, PDF or DOCX
Prior art PDFs (ZIP) — the original documents of the selected prior art as a ZIP bundle (up to 50 documents)
Report + feature chart + prior art PDFs (ZIP) — a complete evidence package in one download
Written Report tab
The Written Report is a document summarizing both the Agent workflow and the findings, with an AI-generated analysis of novelty and inventive step for each targeted claim.
The report adapts to the target patent’s jurisdiction: US patents get an anticipation (35 U.S.C. § 102) and obviousness (35 U.S.C. § 103) analysis, European patents get a novelty (Art. 54 EPC) and inventive step (Art. 56 EPC, Problem–Solution Approach) analysis, and other jurisdictions get a jurisdiction-neutral analysis.
The structure of the Written Report is the following
Executive summary
1. The patent under review
1.1. Target patent
1.2. Key features per selected claim
2. Search results
3. Novelty analysis
3.1. Independent claims
3.2. Dependent claims (fallback positions)
4. Inventive step analysis
4.1. Independent claims
4.2. Dependent claims (fallback positions)
5. Other findings relevant for validity
6. Search methodology
7. Disclaimer
The most relevant documents are cited as D1, D2, … and the analysis references the key features (F-numbers) with paragraph citations to the cited documents. Section 5 additionally checks the claims against the target’s own specification for other validity findings (for US patents, under 35 U.S.C. § 112).
The Written Report is exportable in PDF and DOCX formats in its entirety. If more documents are analyzed after the report was generated, the tab shows a notice and you can regenerate the report to include them.
Note: The search report is automatically generated and relies heavily on AI, including large language models. The search coverage, the feature analysis, and the drafted assessments are subject to the limitations of those systems and of the underlying data. Paragraph references in the form [0000] are IPRally-assigned identifiers for text blocks in the ingested document text and may not match the original publication's own paragraph or page/line numbering. The report does not constitute legal advice, and no decision should be based on it without independent professional review.
Relevant Documents and Non-relevant Documents tabs
These tabs contain the documents evaluated as Relevant (High, Medium, Low) or Not relevant, respectively, in the normal “result list format” of IPRally: full bibliographic data, full document view, (Multi-patent) Ask AI, etc.
Document Sources tab
The Agent documents its work transparently as ordinary IPRally search cases and collections in the project:
one search case per AI claim search, so you can reopen any of them and see (and extend) exactly what the Agent searched,
one number search case per family member of the target, and
a collection named “Examiner Citations of target patent and relevant documents” holding the citation-based candidates.
Analyzing your own documents with the Agent
You can extend the analysis to documents you have found yourself:
Go to the Document Sources tab of the Agent’s project.
Add Collections or new Search Cases containing the documents you want to include.
The new documents appear in the Feature Chart with the Source label User.
Analyze them with the Analyze controls: your own documents are picked up first by Analyze Next Batch, or select them and click Analyze Selected.
This lets you benchmark your own findings — or documents from another database — against the same feature analysis as the Agent’s own results.
Tips for best results
Choose the cut-off date deliberately.
The priority date is the standard cut-off and the default. Switch to the application date when the priority claim is likely to be contested — the broader window can surface additional prior art — or type a custom date when your case requires one.
Remember that the Agent takes the priority date as given; assessing the priority claim itself is up to you.
Focus the attack when you can.
Targeting Specific claims instead of all claims concentrates the feature analysis on the claims that matter for your case and makes the Feature Chart and report easier to digest.
Use the feature review for precision.
Turn on Review features before agent runs and edit the features so they say exactly what needs to be found in the prior art. Well-formulated features are the single biggest lever on the quality of the analysis.
Review in batches — stop when you have enough.
The first 100 documents are often enough to find strong invalidity positions. Check the tiers, then decide whether to analyze the next batch. You can generate the Written Report at any point and regenerate it later.
Mind the Vs. Target Priority Date column.
Documents marked After or Unknown have limited or uncertain usability as prior art — verify their dates before building arguments on them.