Signalment drafts.
You are the doctor.

01 Signalment is a documentation tool

Signalment takes exam notes you dictated or typed and drafts a clinical record from them: a SOAP note, ranked differentials, a tiered plan, a gap check, and a plain-language letter for the client. It reorganises and expresses what you told it. It does not examine the patient, order diagnostics, weigh what you saw across the table, or decide anything.

The model has no access to the animal. It has no access to your practice's records, your imaging, your in-house analyser, or anything you did not put in the box. Every clinical judgement in the output is a restatement or an extrapolation of your input, and it inherits every error and omission in it.

Signalment does not provide veterinary medical advice, diagnosis, or treatment, and using it does not create a veterinarian-client-patient relationship between Signalment and anyone.

02 Who it is for

Signalment is sold to, and intended only for, licensed veterinarians and the credentialed veterinary staff working under their supervision. It assumes a reader who can tell a good differential from a bad one at a glance, and it is unsafe in the hands of a reader who cannot.

It is not for pet owners, not for animal caretakers making treatment decisions without a veterinarian, and not for students using it in place of learning the medicine.

03 Nothing enters a record until you review and sign it

You are solely responsible for every word that reaches a patient's medical record, a client, or a colleague. Before you use any part of a Signalment draft you must read it in full and satisfy yourself that it is accurate, complete, and appropriate for the patient in front of you.

That responsibility is not shared, delegated, or reduced by anything on this site. Your professional judgement, your licence, your state practice act, and your standard of care all govern; Signalment does not displace any of them. If the draft and your judgement disagree, your judgement is right and the draft is wrong.

Copying a draft into a medical record without reading it is a misuse of this product.

04 Differentials are ranked hypotheses, not diagnoses

The differential list is the model's ordering of possibilities consistent with the text you gave it. The ranking reflects how the presentation reads on paper, not prevalence in your region, in your caseload, or in this patient's breed and history beyond what you wrote down.

A serious diagnosis can be absent from the list entirely. Its absence is not evidence against it. The list is a prompt for your own thinking, and it is at its most dangerous when it is plausible and incomplete at the same time.

05 Doses, label claims and reference ranges must be checked

Drug names, doses, routes, frequencies, durations, label claims, contraindications, withdrawal times and laboratory reference ranges in a Signalment draft are generated by a language model. They are not currently verified against a formulary or a reference table. The application says so on every draft.

Check every number against your own formulary, the product label, and the patient's weight, species, breed, age, and organ function before it reaches a prescription, a label, or a client. Extra-label use, species sensitivities such as ivermectin in MDR1 breeds or acetaminophen in cats, and compounding all remain entirely your call.

We intend to move numbers onto verified tables rather than model output. Until this page says they are verified, they are not.

06 The gap check is a prompt, not a standard of care

The gap check names things the visit appears to have called for that the notes do not show. It compares your text against a general expectation of what a workup of that kind usually documents. It is not an audit, not a compliance review, and not a statement of the standard of care in your jurisdiction.

It can flag something you deliberately and correctly chose not to do. It can also stay silent about a real omission, because it only ever sees what you wrote. A clean gap check is not a finding that the visit was complete.

07 Every citation comes from PubMed

When Signalment shows literature for a record, the model's only job is to turn your assessment into a search query. Every title, journal, year, author and link comes back from the National Library of Medicine's PubMed database. We do not let the model write citations, because a model asked for citations will produce convincing ones that do not exist.

What we cannot do is judge the papers for you. A returned abstract may be old, underpowered, retracted, about another species, or simply wrong. Relevance is ranked by a search engine, not by a clinician. Read the paper before you rely on it or cite it to a client, and note that we show abstracts as PubMed returns them — we are not affiliated with, and not endorsed by, the NLM or the NIH.

08 Transcription gets things wrong

Recorded visits are transcribed automatically by a third-party speech model. In a real exam room — with a barking dog, a clipper running, two people talking at once, an accent the model has not heard much of, or a drug name that sounds like another drug name — it will make mistakes. Numbers and units are especially fragile: a misheard dose or a dropped decimal reads perfectly fluently.

The transcript is a starting point for the note, never a record of what was said. Read it against your own memory of the visit before you draft from it, and never rely on it as evidence of what you or the client actually said.

09 The client letter goes out under your name

The plain-language letter is drafted for the owner but addressed by you. Read it as the client will read it. Check that it matches what you actually said in the room, that it does not promise a prognosis you did not give, that it does not soften a warning you meant to land, and that it names the right animal.

You are the sender. Anything in a letter you forward becomes a communication from your practice.

10 The specific ways AI fails

Signalment is built on large language models. They fail in characteristic ways:

• Fluent invention. A model will state a dose, a mechanism, or a finding that is simply false, in exactly the same confident register it uses for things that are true. Fluency carries no information about accuracy.
• Agreement. If your notes imply a diagnosis, the draft will tend to build the case for it rather than against it. It is a poor devil's advocate and should not be used as a second opinion.
• Silent omission. What is missing from a draft is invisible. A complete-looking SOAP can be missing the finding that mattered.
• Confabulated specificity. Asked for detail it does not have, a model supplies plausible detail — a weight, a duration, a value you never gave it. Check that every specific in the draft came from you.
• Instability. The same notes drafted twice can produce different differentials, different rankings, and different plans. Neither run is authoritative.
• Stale knowledge. Model training data has a cutoff. Recent label changes, recalls, guideline revisions and newly described conditions may be absent or wrong.

We route harder cases to more capable models and give them more room to reason, which reduces these failures. It does not eliminate any of them.

11 Regulatory status

Signalment is not a medical device. It has not been reviewed, cleared, approved or certified by the U.S. Food and Drug Administration, the FDA Center for Veterinary Medicine, the USDA, or any other regulator, and we make no claim that it has been. No output should be read as a regulatory clearance, an endorsement, or a validated clinical decision support finding.

Signalment is not certified to any veterinary practice-management or records standard, and it does not determine or attest to compliance with your state practice act, your board's recordkeeping rules, DEA recordkeeping requirements, or any accreditation programme. Meeting those obligations remains yours.

HIPAA does not apply to veterinary medicine — animal health records are not protected health information — and we make no HIPAA claim. What we do with your data is set out in the Privacy Policy.

12 If you are a pet owner

Signalment is a tool used by veterinarians to write up visits. It is not for you, and nothing on this site is advice about your animal.

If a veterinarian has sent you a letter that Signalment helped draft, they wrote it, they reviewed it, and they are the person to ask about it. If your animal is unwell, contact your veterinarian or your nearest emergency veterinary hospital. If you think your animal has been poisoned, in the United States the ASPCA Animal Poison Control Center is available around the clock on (888) 426-4435 and the Pet Poison Helpline on (855) 764-7661; both charge a consultation fee and neither is affiliated with Signalment.