# US 12,602,436 B2 · System for real-time fact-checking of multimedia content > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,602,436 B2 - Document ID: US12602436B2 - Status: Granted - Kind: B2 - Application number: 19/011,265 - Filed: 2025-01-06 - Published: 2025-07-17 - Granted: 2026-04-14 - Earliest claimed: 2024-01-17 - Assignees: AI Seer Pte. Ltd. - Applicants: AI Seer Pte. Ltd. - Inventors: Dennis B. Yap; Shi Ting Wong; A M Shahruj Rashid; Ngok Jeun Wan; Ying Chong Sim; Jiamin Bai; Anmol Dua; Mark Claude Frankel; Sriram Srikanth; Eshaan Govil; Aarush P. Goradia - Web page: https://patentgrep.com/patent/US12602436B2 ## CPC - G06F 16/90335: {Query processing} - G06F 16/24575: {using context} - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract Using a machine learning model query on received content, one or more statements in the content that meet one or more candidate criteria for correctness verification are automatically identified. For at least one statement of the identified one or more statements, a machine learning model is used to assess correctness of the at least one statement including by searching resources associated with the at least one statement to determine a search result and evaluating the search result. A revision of at least one statement assessed to be incorrect in the identified one or more statements is automatically determined. In some embodiments, a bias report is generated, wherein the bias report compares two or more resources identified by the search result and used to assess the correctness of the at least one statement. ## Claims ### Claim 1 A method, comprising: receiving content; using a machine learning model query to automatically identify one or more statements in the content that meet one or more candidate criteria for correctness verification, wherein identifying one or more statements that meet the one or more candidate criteria include distinguishing purported facts from subjective opinion statements based on machine linguistic analysis; for at least one statement of the identified one or more statements, using a machine learning model to perform a first assessment of correctness of the at least one statement including by searching resources associated with the at least one statement to determine a search result and automatically evaluating the search result as an input to the machine learning model using retrieval-augmented generation to determine whether the at least one statement is verifiable, wherein the first assessment includes categorizing the at least one statement as missing context when an ambiguity associated with the at least one statement prevents a definitive assessment of correctness; in response to categorizing the at least one statement as missing context, requesting a clarification on the ambiguity; receiving a user feedback response clarifying the ambiguity; performing a second assessment of the correctness of the at least one statement based on the user feedback response; and automatically determining based on the second assessment, a machine learning model generated revision of at least one statement assessed to be incorrect in the identified one or more statements. ### Claim 2 (depends on 1) The method of claim 1, wherein the search result includes one or more resources for supporting or refuting the assessed correctness of the at least one statement. ### Claim 3 (depends on 2) The method of claim 2, wherein the one or more resources for supporting or refuting the assessed correctness of the at least one statement are ranked based on one or more metrics. ### Claim 4 (depends on 3) The method of claim 3, wherein the one or more metrics are based on one or more of the following associated with the statement: reliability, accuracy, bias, political leaning, endorsements, purpose, intention, reputation, credibility, transparency, language, currency, peer review, or objectivity. ### Claim 5 (depends on 1) The method of claim 1, wherein using the machine learning model to assess the correctness of the at least one statement includes determining a confidence score associated with the assessed correctness. ### Claim 6 (depends on 5) The method of claim 5, wherein the confidence score is based on a ranking of at least one resource identified by the search result, and wherein the at least one resource is used to support or refute the assessed correctness of the at least one statement. ### Claim 7 (depends on 1) The method of claim 1, further comprising categorizing the assessed correctness of the at least one statement as belonging to one of the following categories: true, false, unverifiable, or missing context. ### Claim 8 (depends on 1) The method of claim 1, further comprising generating a summary of the assessed correctness of the at least one statement, wherein the summary identifies resources of the search result that are in agreement or are in disagreement. ### Claim 9 (depends on 1) The method of claim 1, further comprising generating a bias report, wherein the bias report compares two or more resources identified by the search result and used to assess the correctness of the at least one statement. ### Claim 10 (depends on 9) The method of claim 9, wherein the bias report includes a visualization of the two or more resources identified by the search result and used to assess the correctness of the at least one statement. ### Claim 11 (depends on 1) The method of claim 1, wherein the content is received from an extension of a client application, and the revision of the at least one statement accessed to be incorrect is provided to the client application via the extension. ### Claim 12 A system, comprising: one or more processors; and a memory coupled to the one or more processors, wherein the memory is configured to provide the one or more processors with instructions which when executed cause the one or more processors to: receive content; using a machine learning model query, automatically identify one or more statements in the content that meet one or more candidate criteria for correctness verification, wherein identifying one or more statements that meet the one or more candidate criteria include distinguishing purported facts from subjective opinion statements based on machine linguistic analysis; for at least one statement of the identified one or more statements, using a machine learning model, perform a first assessment of correctness of the at least one statement including by searching resources associated with the at least one statement to determine a search result and automatically evaluating the search result as an input to the machine learning model using retrieval-augmented generation to determine whether the at least one statement is verifiable, wherein the first assessment includes categorizing the at least one statement as missing context when an ambiguity associated with the at least one statement prevents a definitive assessment of correctness; in response to categorizing the at least one statement as missing context, request a clarification on the ambiguity; receive a user feedback response clarifying the ambiguity; perform a second assessment of the correctness of the at least one statement based on the user feedback response; and automatically determine based on the second assessment, a machine learning model generated revision of at least one statement assessed to be incorrect in the identified one or more statements. ### Claim 13 (depends on 12) The system of claim 12, wherein the search result includes one or more resources for supporting or refuting the assessed correctness of the at least one statement. ### Claim 14 (depends on 12) The system of claim 12, wherein the memory is further configured to provide the one or more processors with instructions which when executed cause the one or more processors to: generate a summary of the assessed correctness of the at least one statement, wherein the summary identifies resources of the search result that are in agreement or are in disagreement. ### Claim 15 (depends on 12) The system of claim 12, wherein the memory is further configured to provide the one or more processors with instructions which when executed cause the one or more processors to: generate a bias report, wherein the bias report compares two or more resources identified by the search result and used to assess the correctness of the at least one statement. ### Claim 16 (depends on 12) The system of claim 12, wherein the content is received from an extension of a client application, and the revision of the at least one statement accessed to be incorrect is provided to the client application via the extension. ### Claim 17 A method, comprising: receiving a selection of content; generating a content fact-checking request, wherein the content fact-checking request includes the selected content; providing the content fact-checking request to a content fact-checking service that performs machine linguistic analysis for distinguishing a purported fact from a subjective opinion statement for the selected content; and in response to providing the content fact-checking request to the content fact-checking service, receiving a content fact-checking result, wherein the content fact-checking result is generated by the content fact-checking service searching resources associated with the selected content to determine a search result and evaluating the search result as an input to the machine learning model using retrieval-augmented generation to determine whether the selected content is verifiable including by: categorizing at least a portion of the selected content as missing context due to an ambiguity, requesting a clarification on the ambiguity, receiving a user feedback response clarifying the ambiguity, performing an updated assessment of correctness of the selected content based on the user feedback response, and automatically determining based on the updated assessment, a machine learning model generated revision of the selected content; wherein the content fact-checking result includes: one or more statements of the content identified as meeting one or more candidate criteria for correctness verification, and a correctness assessment of each of the one or more identified statements. ### Claim 18 (depends on 17) The method of claim 17, wherein a correctness assessment of at least one of the one or more identified statements is categorized as belonging to one of the following categories: true, partly true, inconclusive, partly false, or false. ### Claim 19 (depends on 17) The method of claim 17, wherein the content fact-checking result includes a revision of at least one of the one or more identified statements. ### Claim 20 (depends on 19) The method of claim 19, wherein the revision includes a citation to a resource in support of a factual claim of the revision. ### Claim 21 A method, comprising: receiving content; using a machine learning model query to automatically identify one or more statements in the content that meet one or more candidate criteria for correctness verification, wherein identifying one or more statements that meet the one or more candidate criteria include distinguishing purported facts from subjective opinion statements based on machine linguistic analysis; for at least one statement of the identified one or more statements, using a machine learning model to identify an ambiguity associated with the at least one statement that prevents a definitive assessment of correctness and categorize the at least one statement as missing context including by searching resources associated with the at least one statement to determine a search result and evaluating the search result as an input to the machine learning model using retrieval-augmented generation to determine whether the at least one statement is verifiable; in response to categorizing the at least one statement as missing context, request a clarification on the ambiguity; receiving a user feedback response to clarify the ambiguity of the at least one statement; perform a second assessment of the correctness of the at least one statement based on the user feedback response; and automatically determine based on the second assessment, a machine learning model generated revision of at least one statement assessed to be incorrect in the identified one or more statements.