The AUTOMAT Framework

The AUTOMAT Framework is a robust methodology for functional task execution with AI. It provides a structured approach to crafting high-fidelity scientific prompts.

Framework Components

AUTOMAT is an acronym for:

Component	What It Means	Why It Matters
A	Audience	Who will read/use this output?
U	User Persona	Who is the AI acting as?
T	Task	What specific action is required?
O	Output	What format/structure for results?
M	Method	How should the task be approached?
A	Assumptions	What constraints or boundaries apply?
T	Tone	What voice/style is appropriate?

Component Breakdown

A – Audience

Define who will consume this output.

This shapes: - Technical depth - Terminology level - Required disclaimers - Level of detail

Examples:

Audience	Implication
IP Legal Team	Formal, cite sources, flag novelty
R&D Colleagues	Technical depth, assume domain knowledge
Management	High-level, business impact focus
External Partners	Balanced detail, avoid proprietary info

U – User Persona

Define the AI's role/expertise.

This influences: - Knowledge domain - Perspective - Depth of analysis

Examples:

Act as a Senior Polymer Chemist with expertise in electrospinning
Act as a Materials Science Patent Analyst
Act as a Laboratory Safety Officer reviewing protocols
Act as a Data Scientist specialising in mechanical testing analysis

Why it matters: Without a persona, the AI defaults to "helpful generalist"—too broad for scientific work.

T – Task

Specify the precise action required.

Be explicit:

❌ "Summarise this paper"
✅ "Extract synthesis parameters from Section 3.2 of this paper"

❌ "Analyse this data"
✅ "Calculate tensile strength and Young's modulus from these 7 stress-strain curves, excluding outliers >2 standard deviations"

O – Output

Define the exact format and structure.

Common scientific formats:

Markdown table
CSV data
JSON structure
Bullet points with citations
Structured report (sections defined)
Python code (commented)

Example specification:

Output format: Markdown table with columns:
- Polymer Type
- Synthesis Temperature (°C)
- Precursor Materials
- Yield (%)
- Source (DOI)

Include 10-15 entries, sorted by yield (descending).

M – Method

Specify the approach or methodology.

Examples:

"Use systematic literature review principles"
"Apply ISO 527 tensile testing standards"
"Follow IUPAC nomenclature"
"Compare results against baseline (pure PLA)"
"Calculate statistics using standard error of the mean"

A – Assumptions (Constraints)

Critical for scientific accuracy.

Define:

What to include (whitelisting)
What to exclude (blacklisting)
How to handle uncertainty

Examples:

Constraints:
- Focus exclusively on non-oxide ceramics
- Exclude papers before 2020
- If data is missing, mark as "Not reported"—do not estimate
- Flag any mentions of electrospinning
- Only include peer-reviewed sources (no preprints)

Why it matters:

Without constraints, the AI may: - Include irrelevant material - Speculate on missing data (hallucination risk) - Mix incompatible datasets

T – Tone

Set the voice and style.

Common tones for scientific work:

Technical/Objective: For peer review, reports
Formal: For legal, regulatory documents
Conversational: For internal brainstorming
Cautious: For safety-critical procedures

Example:

Tone: Technical and objective, suitable for submission to 
Materials Science & Engineering: A. Avoid speculation or 
promotional language.

Complete AUTOMAT Example

Scenario: Literature Review for Patent Application

Prompt:

[U] Act as a Materials Science Patent Analyst with expertise in 
polymer composites and prior art searches.

[A] Audience: AmaDema IP Legal Team preparing a patent application 
for a novel PLA/graphene nanocomposite synthesis method.

[T] Task: Review the following 12 papers and extract information 
relevant to prior art assessment for our patent application focusing 
on electrospinning-based PLA/graphene synthesis.

[O] Output: Markdown table with columns:
   - Paper DOI
   - Year
   - Synthesis Method
   - Graphene Content (wt%)
   - Key Innovation
   - Overlap with Our Method (High/Medium/Low/None)
   - Notes for Legal Team

[M] Method: 
   - Apply patent prior art search principles
   - Focus on claims that would challenge patentability
   - Highlight any methods using electrospinning + in-situ reduction

[A] Assumptions/Constraints:
   - Only include peer-reviewed journal articles
   - Exclude review papers and patents
   - If graphene content not specified, mark "Not reported"
   - Flag any papers from competing companies (Nanotech Solutions, 
     PolyMat GmbH)
   - Do not speculate on methods not explicitly described

[T] Tone: Formal, objective, suitable for legal review. Err on 
side of caution when assessing overlap.

Materials Science Examples

Example 1: Tensile Testing Analysis

[U] Act as a Materials Testing Engineer with expertise in ISO 527 
tensile testing standards.

[A] Audience: R&D team conducting quality control on nanofiber mats.

[T] Task: Analyse these 7 stress-strain curves and calculate tensile 
strength, Young's modulus, and elongation at break for each sample.

[O] Output: 
   1. Summary table (Sample ID, Tensile Strength, Young's Modulus, 
      Elongation)
   2. Identify outliers (>2 SD from mean)
   3. Calculate average ± standard deviation for passing samples
   4. Python code used for calculations (commented)

[M] Method: Follow ISO 527-1:2019 standards. Use linear regression 
for modulus (strain 0.05-0.25%).

[A] Constraints:
   - Exclude any samples with irregular load curves (noise, slippage)
   - Young's modulus calculated from linear region only
   - Report all values to 2 decimal places
   - Flag any samples with premature failure (<50% expected strain)

[T] Tone: Technical, objective, suitable for QC documentation.

Example 2: SEM Image Analysis

[U] Act as a Materials Characterisation Specialist with expertise 
in SEM image analysis of electrospun fibers.

[A] Audience: R&D scientists preparing a manuscript for 
Polymer journal.

[T] Task: Analyse this SEM image (attached) and measure fiber 
diameter distribution and calculate porosity.

[O] Output:
   1. Average fiber diameter ± standard deviation (20 measurements)
   2. Diameter distribution histogram (bin size: 50 nm)
   3. Calculated porosity (% area not covered by fibers)
   4. Assessment of fiber uniformity (coefficient of variation)
   5. Brief interpretation (2-3 sentences)

[M] Method:
   - Use scale bar for calibration
   - Measure 20 representative fibers randomly across image
   - Calculate porosity via threshold segmentation
   - Compare to typical electrospun PLA (literature benchmarks)

[A] Constraints:
   - Exclude fibers at image edges (edge effects)
   - Exclude beads or defects from diameter measurements
   - If scale bar unclear, request clarification—do not estimate
   - Compare only to electrospun materials (not melt-spun)

[T] Tone: Scientific, suitable for peer-reviewed publication. 
Emphasise reproducibility of measurements.

Common Mistakes to Avoid

❌ Mistake 1: Vague Task Definition

Bad:

Analyse this data

Good:

Calculate mean tensile strength and standard deviation from these 
7 datasets, excluding outliers beyond 2 SD

❌ Mistake 2: Missing Output Format

Bad:

Summarise these polymer papers

Good:

Create a markdown table with columns: DOI, Polymer Type, Synthesis 
Method, Key Finding. Include 10-15 entries.

❌ Mistake 3: No Constraints on Hallucination

Bad:

What is the melting point of this polymer?

Good:

If the melting point is reported in the paper, extract it with page 
number citation. If not reported, state "Not reported"—do not 
estimate or use literature values.

Exercise: Build Your Own AUTOMAT Prompt

Challenge

Scenario: You need to format 5 lab notebook entries (handwritten synthesis protocols) into a standardised digital template.

Your task: Construct a complete AUTOMAT prompt for this task.

Requirements: - Define appropriate user persona - Specify output format (what fields?) - Include constraints (what to do with missing info?) - Set appropriate tone

Test your prompt in the sandbox and refine based on output quality.

When to Use AUTOMAT

Best for:

✅ Functional tasks with clear outputs (data extraction, formatting)
✅ Scientific workflows requiring precision
✅ Tasks where IP protection is critical
✅ Situations requiring reproducible results

Not ideal for:

❌ Open-ended brainstorming
❌ Conversational learning (use CO-STAR instead)
❌ Simple, one-sentence tasks

Next: CO-STAR Framework – For context-heavy communication →