AI blood test analyzer glossary

• A
• B
• C
• D
• E
• F
• G
• H
• L
• M
• O
• P
• R
• S
• T
• W

A

AI blood test analyzer

Software that reads a blood test report and explains the results in plain language using artificial intelligence. On BloodGPT.net the term covers both dedicated, health-trained analyzers and general-purpose chatbots used for the same job, all scored on the same criteria. An analyzer interprets results you already have; it does not perform the test and does not replace a clinician.

See also: Dedicated health-trained model, General-purpose chatbot, Structured report

B

Biomarker

A measurable substance or characteristic in the body — such as hemoglobin, glucose, LDL cholesterol or TSH — that provides information about health. Each row on a lab report is usually one biomarker, shown with a value, a unit and a reference range. Clinicians rarely rely on a single biomarker; they interpret results together and in context.

See also: Reference range, SI and conventional units

C

CMP and BMP (metabolic panels)

The basic metabolic panel (BMP) measures glucose, calcium, electrolytes (sodium, potassium, chloride and bicarbonate) and kidney markers (BUN and creatinine). The comprehensive metabolic panel (CMP) adds liver-related tests: albumin, total protein, ALP, ALT, AST and bilirubin. Both are among the most frequently ordered blood panels.

See also: Complete blood count (CBC), Biomarker

Complete blood count (CBC)

A panel that measures the cells in your blood — red blood cells, white blood cells and platelets — along with hemoglobin, hematocrit and red cell indices such as MCV. A “CBC with differential” also breaks white blood cells into types: neutrophils, lymphocytes, monocytes, eosinophils and basophils. It is one of the most common blood tests worldwide.

See also: CMP and BMP (metabolic panels), Biomarker

Context window

The maximum amount of text — and, for multimodal models, images or documents — that a large language model can consider at once, measured in tokens. Long multi-page reports, earlier chat history and instructions all share this space. When a conversation outgrows it, earlier details may be dropped or compressed, so a chatbot can lose track of values or context you gave it earlier.

See also: Large language model (LLM), Prompt

D

Data retention

How long a service keeps your uploads, chat history and account data, and what happens after you delete them. A clear retention policy states a period, explains whether backups or safety logs persist after deletion and says who can access stored data. For health documents, shorter and clearly stated retention is better.

See also: Model training opt-out, De-identification and redaction

De-identification and redaction

Redaction permanently removes identifying details — name, date of birth, record numbers — from a document so they cannot be recovered or extracted. De-identification is the broader process of making data no longer reasonably linkable to a person; HIPAA's safe harbor method, for example, lists 18 identifier types to remove. Drawing a shape over text in a PDF is not true redaction, because the underlying text can often still be selected.

See also: Protected health information (PHI), HIPAA

Dedicated health-trained model

An AI model built and tuned specifically for a health task — here, interpreting blood test results — rather than a general model prompted to discuss health. Dedicated models are designed around lab-report formats, reference ranges and structured output. On BloodGPT.net this is assessed under the AI technology criterion, the most heavily weighted in our methodology.

See also: General-purpose chatbot, Weighted criteria

E

Editorial score

The single 0–10 score BloodGPT.net assigns each platform, combining 10 weighted criteria and rounded to one decimal place. It reflects the editorial team's assessment of public information under a published methodology; it is not a clinical accuracy measurement. Current scores: Kantesti 9.4, ChatGPT 7.1, Gemini 6.9, Claude 6.8 and Perplexity 6.7.

See also: Weighted criteria

F

False reassurance

When an AI explanation suggests a result is normal or unimportant when it may need attention — for example, because the tool used the wrong unit, applied a generic reference range or missed a lab flag. False reassurance can delay a conversation with a clinician, which is why AI output should always be checked against the original report. Its opposite, a false alarm, causes needless worry.

See also: Hallucination, Lab flag (H/L)

FTC Health Breach Notification Rule

A US Federal Trade Commission rule requiring vendors of personal health records and related entities not covered by HIPAA — including many health apps — to notify users, the FTC and sometimes the media after a breach of unsecured health information. Updates in 2024 clarified that unauthorized sharing of data, not only hacking, can count as a breach. The rule governs notification, not what data a company may collect.

See also: HIPAA, Data retention

G

GDPR special-category data

Under Article 9 of the EU General Data Protection Regulation, certain personal data — including data concerning health, genetic data and biometric data used for identification — is “special category” data. Processing it is prohibited unless a specific condition applies, such as the person's explicit consent. The UK GDPR has an equivalent provision, so blood test results uploaded by EU and UK users receive heightened protection.

See also: Protected health information (PHI), Data retention

General-purpose chatbot

An AI assistant built to handle almost any topic, such as ChatGPT, Gemini, Claude or Perplexity. These tools can read lab reports and explain biomarkers well, but they are not dedicated blood test analyzers: they have no built-in lab reference ranges, answer in free-form chat and can vary between sessions. Perplexity adds web citations, while Claude is notably cautious and defers to clinicians.

See also: Large language model (LLM), Dedicated health-trained model, Repeatability

H

Hallucination

Output from an AI model that sounds plausible and confident but is not supported by the input or by fact. On lab reports, hallucinations include invented values, markers that were never tested and made-up reference ranges. They are hard to detect without comparing the output against the original document.

See also: False reassurance, Large language model (LLM)

HIPAA

The US Health Insurance Portability and Accountability Act, whose Privacy and Security Rules protect health information held by covered entities — health plans, health care clearinghouses and health care providers that conduct certain electronic transactions — and their business associates. HIPAA generally does not apply to a consumer app you choose to use on your own, even if you upload medical records to it. A tool provided through your clinician or insurer under a business associate agreement may be covered.

See also: Protected health information (PHI), FTC Health Breach Notification Rule

L

Lab flag (H/L)

A marker printed by the laboratory next to a result outside its reference range: H for high, L for low, and often HH, LL, an asterisk or “critical” for values needing prompt attention. Flags reflect the lab's own range and methods, which makes them one of the most reliable signals on a report. AI tools sometimes drop flags during extraction, so compare them with the tool's output.

See also: Reference range, Optical character recognition (OCR)

Large language model (LLM)

An AI model trained on very large amounts of text to predict and generate language. LLMs power general-purpose chatbots and explain complex topics fluently, but they generate probable text rather than looking up verified facts, which is why they can hallucinate. Many current LLMs are also multimodal.

See also: Multimodal model, Hallucination, Context window

Lipid panel

A blood test measuring fats in the blood: total cholesterol, LDL cholesterol, HDL cholesterol and triglycerides, sometimes with non-HDL cholesterol or calculated ratios. Results are reported in mg/dL in the US and in mmol/L in most other countries. Fasting status can affect triglycerides, so it matters when reading the panel.

See also: SI and conventional units, Biomarker

M

Model training opt-out

A setting or request that stops an AI provider from using your conversations and uploads to train or improve its models. Availability, defaults and naming vary by provider and plan, and they change over time. Check it before uploading health documents, alongside retention settings.

See also: Data retention

Multimodal model

An AI model that can process more than one kind of input — typically text plus images, and often PDFs or audio. Multimodal chatbots can read a photo or PDF of a lab report, but accuracy depends on image quality and layout. Extraction errors pass silently into the explanation unless you check them.

See also: Optical character recognition (OCR), Large language model (LLM)

O

Optical character recognition (OCR)

Technology that converts images of text — photos, scans or image-based PDFs — into machine-readable text. On lab reports, OCR must also keep table columns aligned so each value stays with its marker, unit and range. Blurry photos, skew, faint decimal points and multi-column layouts are common sources of OCR errors.

See also: Multimodal model, Lab flag (H/L)

P

Prompt

The instruction or question you give an AI model, along with any attached files. Clear prompts — for example, asking for an extraction table before any explanation, or telling the model to use only the reference ranges on your report — reduce errors. No prompt can give a general model built-in lab reference ranges it does not have.

See also: Context window, Large language model (LLM)

Protected health information (PHI)

Under HIPAA, individually identifiable health information held or transmitted by a covered entity or its business associate — such as lab results linked to your name or record number. Once you download your own results and share them with a consumer app, they remain sensitive, but HIPAA protections usually no longer apply to that copy.

See also: HIPAA, De-identification and redaction

R

Reference range

The range of values for a biomarker that a laboratory considers typical, usually set so that about 95% of a healthy reference population falls within it. Ranges vary between laboratories, methods and populations, and can differ by age, sex and pregnancy. A result just outside the range is not automatically a problem, and one inside it is not automatically fine — context matters.

See also: Lab flag (H/L), Biomarker, SI and conventional units

Repeatability

The degree to which a tool produces the same output from the same input. A repeatable AI blood test report shows the same values, flags and explanations each time the same report is analyzed, so changes over time reflect your results rather than the model. General chatbots generate text probabilistically, so their answers can vary between sessions.

See also: Structured report, General-purpose chatbot

S

SI and conventional units

Two systems for reporting lab results. Conventional units such as mg/dL are standard in the US, while SI units such as mmol/L and µmol/L are used in the UK, Europe and most other countries. Each analyte has its own conversion factor — glucose in mmol/L × 18 ≈ mg/dL, while creatinine in mg/dL × 88.4 = µmol/L — so mixing the systems can badly distort interpretation.

See also: Reference range, Lipid panel

Software as a medical device (SaMD)

Software intended for one or more medical purposes — such as diagnosing, monitoring or treating disease — that performs those purposes without being part of a hardware medical device. Regulators including the US FDA have specific frameworks for SaMD and AI-enabled medical software. Whether a product is regulated depends on its intended use and claims, so check any regulatory claim directly in the regulator's public records.

See also: AI blood test analyzer

Structured report

An output format that presents results in consistent fields — typically biomarker, value, unit, reference range, status and a plain-language explanation — organized by panel. Structured reports are easier to verify, compare over time and share with a clinician than free-form chat. Report depth and structure is one of the criteria in BloodGPT.net's methodology.

See also: Repeatability, Dedicated health-trained model

T

Thyroid panel

A group of blood tests that assess thyroid function, most often TSH (thyroid-stimulating hormone) and free T4, sometimes with free or total T3 and thyroid antibodies. TSH is often the first test ordered. Results can be affected by pregnancy, illness, some medications and high-dose biotin supplements, so context is important.

See also: Biomarker, Reference range

W

Weighted criteria

A scoring approach in which each criterion contributes to the total in proportion to its assigned weight. BloodGPT.net scores every platform 0–10 on 10 criteria whose weights add up to 100% — from AI technology at 18% to report depth, medical transparency and validation evidence at 12% each — and combines them into one editorial score.

See also: Editorial score, Dedicated health-trained model

Missing a term? Email editorial@bloodgpt.net. See also the methodology criteria glossary and our guides.