Zero-Click is the internal name of a product decision we made early at Laudos.AI: the radiologist should be able to produce, review and sign a report without clicking buttons that don't add clinical value. It is not aesthetic minimalism. It is the consequence of mapping, exam by exam, where each click goes — and discovering that most of them serve systems, not physicians.
Anatomy of a traditional report — clicks mapped.
Before removing any button, we mapped an average chest-CT report on a conventional editor. The result was instructive. In one exam, the radiologist usually: clicks to open the queue; clicks to open the exam; clicks to open the template; clicks to navigate between sections (technique, findings, impression); clicks to insert a macro; clicks to confirm the macro; clicks to correct a pre-generated sentence; clicks to confirm the correction; clicks to exit the macro; clicks to open the measurement module; clicks to confirm the measurement; clicks to insert a comparison; clicks to open the prior exam; clicks to close the prior exam; clicks to review; clicks to sign; clicks to confirm the signature; clicks to next exam. At volume, that is hundreds of clicks per shift.
Clicks have three costs: time (the obvious one), cognitive (each interface decision drains focus from clinical reasoning) and physical (repetitive-strain syndromes are real in radiology). Reducing clicks is not cosmetic; it is clinical capacity released.
Clinical-first UX principles.
- Anything that can be inferred should be inferred. Modality, body part, priority, patient — should come from exam context, not clicks.
- Natural voice as default input. No dictated punctuation, no artificial commands ("new line," "comma," "open parenthesis"). Speak like you think.
- Structure as output. Text comes out organized by modality (technique/findings/impression), with consistent terminology, without the radiologist navigating sections manually.
- Review always visible, never friction. The physician sees what the AI proposed, edits, and continues. There is no separate "review mode."
- Signature as intentional action. One large isolated button, with confirmation proportional to the clinical risk of the exam.
- Shortcuts for what matters — push-to-talk, next exam, mark critical finding — not shortcuts to fake speed.
What disappeared (and why).
Explicit macros disappeared. Instead of "insert macro X, configure parameters, confirm," the radiologist describes the finding in natural voice and AI applies the right template for the modality — visible in the editor, editable at any moment. The macro still exists (per-modality template structure); it stopped being a button.
Formal voice commands disappeared. "New line," "paragraph," "comma," "open quote" — radiologists take months to internalize those, and lose months when switching vendors. Speech-to-report (not speech-to-text) understands the text as clinical discourse and formats it automatically.
Manual navigation between sections disappeared. When the radiologist says "adequate multiphase technique, no biliary dilation," the sentence routes itself to the right section.
The standalone measurement module disappeared as separate context. Spoken measurements ("eight by six millimeter nodule in the posterior segment of the right upper lobe") enter the description in a consistent format, with unit preserved and rounding per service protocol.
A button that exists because of software, not because of medicine, should disappear.
What looked like it disappeared but only became implicit.
Structure: still there. The report comes out with technique, findings and impression organized. The radiologist can edit the structure, add service-specific sections (incidental findings, recommendations), reorder it. Structure did not disappear — it became a visible starting point instead of a clicks layer.
Standardization: also there. Institutional services can define templates (modality, protocol, client), vocabulary, signature phrasing. The individual radiologist sees the service starting point and can personalize within the limits institutional governance allows. Standardization was not removed — it was made negotiable where that made sense and fixed where it did not.
Governance: same. CRIT (critical-finding communication), audit trail, model/version logging, AI-use identification — all still there. They don't appear as visible friction to the radiologist; they become report metadata, available for audit and for patients exercising LGPD rights.
Zero-click risks — and how we mitigate.
The biggest risk of less-click is too-fast-click: signing without reading. We mitigate with three layers. First, mandatory pre-signature review — the report passes through a visual review screen that cannot be skipped by configuration; it can be quick (seconds for a simple report) but it exists. Second, risk-proportional confirmation — reports with marked critical findings trigger an explicit additional confirmation. Third, a signed per-exam audit hash: each digital signature carries a hash of the content at the moment of signing, bound to the log of everything the AI suggested and everything the physician edited. Any divergence becomes detectable.
The second risk is homogenization — every report looking the same. We mitigate with explicit personalization: the radiologist's style is learned (preferred phrasing, description pattern) and respected within institutional template limits. The service can choose between high standardization (100% institutional language) and individual style preserved within the approved structure.
The third risk is dependence: the radiologist losing the ability to report without assistance. This risk is real and old (it already existed with traditional voice and macros). The mitigation is training — residents need to learn structure before delegating to AI, and services need to maintain contingency scenarios (manual reporting when the system goes down).
Comparison with close neighbors.
Voxel: strong on assembly via selectors and checkboxes. In highly parameterizable flows (obstetric US, thyroid ultrasound, simple X-rays), that accelerates a lot. In complex CT and MRI, the selection tree can interfere with clinical reasoning — because the radiologist thinks in findings, not click sequences.
Laudite: historical strength in Brazilian medical voice recognition with masks and autotexts. For radiologists who want to type less without changing the macro structure, it's the obvious alternative. For those who want to change the structure — speak like you think and receive a structured report instead of transcribed text — it's a different choice.
LeoRad: strong bet on structured-report AI and mobile. For some profiles (individual teleradiology), it works well. The differential sits in PACS/RIS integration depth and institutional governance granularity.
Laudos.AI: native speech-to-report + serious integration + documented governance + CRIT + public telemetry. Optimized for operations that cannot afford ambiguity — where reporting must be fast AND auditable.
What the radiologist notices on day one and in the first month.
On day one, two feelings compete: "this is much more fluid" and "what if the AI writes something wrong." The first comes from the relief of not navigating menus. The second is healthy prudence — a good radiologist is skeptical. We answer the second by leaving what the AI proposed visible in the editor, with an implicit diff between proposed and final versions. Within a few hours, the radiologist sees the pattern (the AI is consistent, edits infrequently, fails predictably when it fails) and settles into the right level of review.
In the first month, three things consolidate. Time per report drops measurably (Laudos.AI Labs publishes production telemetry: median TAT of 52s on the rolling 30-day window). Rework drops (fewer report kickbacks for missing findings or language inconsistency). Inter-radiologist variability in the same service drops — the standardized starting point forces convergence without trapping individual style.
Zero-Click is not eliminating interaction. It is eliminating interaction that doesn't serve medicine. The radiologist still decides. The physician still signs. AI is still an auxiliary tool. What changes is the amount of friction between clinical reasoning and the delivered report — and that difference, in production, becomes clinical capacity released.