Free tools Windows power users keep installed
One-click scans. No signup required.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
What if a detector could flag a document because its ideas came from AI—even when a person wrote every sentence? That is the problem IdeaLens is designed to study. Unlike a prose detector, which looks for signs in the wording, IdeaLens estimates whether the ideas behind a long-form document came from a human or an AI system. Its score is an estimate, not proof of who originated an idea or how a writer worked.
What does it mean to detect AI ideas?
AI-writing detectors and idea-provenance detectors ask different questions. A prose detector estimates who wrote the words. An idea-provenance detector estimates where the document’s underlying concepts or plan came from, regardless of who composed the final wording.
That distinction matters in both directions: a person can write original prose from an AI-suggested plan, while an AI system can generate prose from a plan supplied by a human. A tool that only examines surface wording may not distinguish those cases.
Recommended Free Tools
The authors of IdeaLens frame the issue this way: “While modern AI detectors identify who wrote the words, emerging policies on AI use increasingly hinge on a different question: who came up with the ideas?” Their paper, “IdeaLens: Detecting AI Ideas in Long-form Writing”, was submitted to arXiv on October 5, 2026. It is a preprint, so its reported findings should be read as the authors’ results rather than independently replicated conclusions.
#1 Best Overall
How IdeaLens analyzes a document
Instead of classifying a document solely from its original prose, IdeaLens represents it as an outline. Each outline item pairs a discourse role with a brief paraphrase of the content. The goal is to retain the conceptual structure while reducing overlap with the document’s exact wording.
The authors describe the method as using “lists of items that each pair a discourse role with a brief, paraphrased description of the content, minimizing word-level overlap with the raw text.” In practical terms, the system tries to examine the document’s conceptual plan rather than treating its sentences as the main evidence.
Rank #2
That representation is then classified to estimate idea provenance. It does not record a writer’s brainstorming session, identify a particular prompt, or observe who first conceived a specific idea. The output is a model’s inference from the document.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →What the reported results show
The paper reports several tests intended to separate idea provenance from prose provenance. The figures below are results reported by Rajendhran and coauthors for their study conditions; they are not guarantees for other writing, detectors, or settings.
When AI writes from increasingly detailed human plans
In a controlled study, the AI-flag rate for IdeaLens fell from 95% to 7% as AI models wrote from increasingly detailed plans supplied by humans. Pangram 4, a prose-provenance detector, still flagged 92% in the reported condition. This pattern is consistent with the systems responding to different signals: IdeaLens is aimed at the origin of ideas, while Pangram examines prose.
When people write from AI-generated plans
For AI-derived plans, IdeaLens remained above 96% in the reported tests. In a separate set of 50 stories written by humans from AI-generated plans, IdeaLens flagged 68% as AI, compared with 8% for Pangram 4. The result illustrates why prose authorship and idea origin should not be treated as interchangeable. It also comes from a small study set, not a universal estimate of performance on human writing.
Benchmarks and broader coverage
The paper says IdeaLens was evaluated on 19 existing detection benchmarks, with tests across domains, formats, and languages. Its authors also say they released models and labeled datasets for future research. A separate Unite.AI account of the work reports 95.3% accuracy for shared idea-and-prose provenance and 81.3% for mixed provenance, as well as a 24-language test. Those specific figures are attributed here to that article’s summary; they should not be compared with other headline accuracy numbers without matching the benchmark, dataset, and threshold.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteWhy an IdeaLens score is not proof
A central limitation is how the model was trained. The authors used one million FineWeb documents with “silver” labels generated by Pangram, a prose-provenance detector. Those labels are proxies; they do not directly document the actual origin of every idea in each training text. If the proxy labels are imperfect or capture prose characteristics rather than idea history, that can affect what the model learns.
The reported rates and benchmark results therefore describe performance on the paper’s datasets and evaluation conditions. They do not establish that the tool can reliably determine the origin of ideas in every real-world document. A flag cannot, by itself, establish a writer’s intent, show that a particular person used AI, or reconstruct the precise history of an idea.
The paper’s arXiv page links a demo, code, and model and data resources, but their privacy terms and commercial status have not been established here. Anyone considering a live document tool should check its current terms and handling of submitted text before sharing sensitive writing.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to evaluate idea-provenance detectors
Do not rank tools by headline accuracy alone. A meaningful comparison needs to account for what each system is classifying, how it represents documents, and whether the test resembles the case at hand.
- Target: Does the system estimate who wrote the wording, who supplied the ideas, or a combination?
- Input representation: Does it classify full prose, or a transformed representation such as IdeaLens’s role-labeled paraphrase outline?
- Provenance setup: Were the ideas and the prose produced by the same source, or was one human-supplied and the other AI-generated?
- Evaluation conditions: What datasets, domains, formats, and languages were tested, and how were the labels established?
- Threshold and errors: What score threshold triggers a flag, and what false-positive and false-negative rates were measured at that threshold?
- Ground truth: Are labels based on directly observed writing and ideation histories, or on proxies such as another detector’s classifications?
These distinctions are especially important when a result could affect a student, employee, or author. A tool designed to estimate idea provenance can contribute one signal to a careful review, but the study does not establish that its output should be used as a standalone judgment about an individual.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
