Units
1. Introduction to Artificial Intelligence in Chemistry: Roles, Boundaries, Validation and Ethics 2. Molecular Representation and Chemical Structure: Validation with SMILES, InChI and RDKit 3. Reaction Prediction and Mechanism: Product, Condition and Adverse Reaction Prediction 4. Spectrum Interpretation Support: NMR, IR and Mass Spectrometry 5. Literature Review and Synthesis Planning: Source, Procedure and Roadmap 6. Chemical Data Analysis and Python: pandas, Stoichiometry and Calibration 7. Molecular Property Prediction and QSAR: Resolution, pKa and Model Limits 8. Laboratory Documentation: Experiment Notebook, ELN and Reproducibility 9. Safety and Hazard Assessment: GHS, SDS and the Limits of Artificial Intelligence 10. Confirmation, Hallucination, and Scientific Integrity 11. Human in Experimentation: Ethics, Privacy, Responsibility and End-to-End Workflow
Unit 8 / 11

Laboratory Documentation: Experiment Notebook, ELN and Reproducibility

Gains:

  • Understand the components of a complete, honest, and repeatable experimental record and be able to use AI as a shaping and control tool
  • Ability to record deviations from the plan completely, without having to adapt the observations to artificial intelligence
  • 'Can someone else repeat the recording?' Ability to have artificial intelligence inspect through your eyes and fill in the gaps with real information

In chemistry, a result acquires scientific value only when someone else can repeat it. The key to this is documentation: a complete record of what you did, under what circumstances, in what quantity, with what result. The paper experiment notebook or electronic notebook (ELN - Electronic Lab Notebook) is the carrier of this record. AI dramatically speeds up documentation: transcribes the procedure, reminds you of missing fields, drafts a security note, tabulates data. But the truth and honesty of records belong to man; AI doesn't know what happened, you have to write down what happened. In this unit we will learn about complete, honest and repeatable documentation with AI.

What should be in a good experimental record?

  • Purpose: What are you trying to do, what hypothesis?
  • Materials: Each reagent name, source/lot number, amount (mass, mol, volume), purity.
  • Procedure: What was done step by step; temperature, time, addition order.
  • Observations: Color change, outgassing, temperature increase—everything expected and unexpected.
  • Results: Yield, spectrum data, purity.
  • Deviations: What was done differently from the plan? (This is the most skipped but most valuable part.)
  • Date, signature, circumstances: Who, when, on what device.

AI is very good at generating this template and guiding its filling. But only you have the knowledge of "what is observed"; AI cannot make up observations; if it did, it would be scientific fraud.

Caution: Do not tell the AI ​​to "write down the observations." AI doesn't know what it is; It can produce observations that seem plausible but do not occur. AI formats and reminds, you provide the observation.

Step by step: documentation with AI

  1. Generate template: Ask AI for an experiment notebook template suitable for your project.
  2. Give raw notes: Give the AI ​​the scattered notes you keep in the lab.
  3. Edit: AI puts the notes into the template, fill in the missing fields with "?" Mark it with .
  4. You fill in the gaps: Fill in the gaps marked by AI with real information.
  5. Reproducibility check: "Can someone else repeat the experiment with this recording?" Have the AI ​​check it.
  6. Version and signature: Add date, version, signature; Store it in an unalterable format.

Four copyable templates

1) Producing an experiment notebook template:

Task: Produce an experimental notebook template for an organic synthesis experiment. Sections: Purpose, Materials (name, lot, amount mol/g, purity), Procedure (step), Observations, Results (yield, characterization), Deviations from Plan, Safety notes, Waste disposal, Date/Signature. Show the fields to be filled under each section with square brackets.

2) Streamlining raw notes:

Below are the loose notes I kept in the lab:[RAW NOTES]Task: Place them in the experimental notebook template.Mark any fields where information is missing as [MISSING - fill in].FINISH: do not add any observations or numbers that are not in my notes.

3) Repeatability inspection:

Below is a recording of an experiment:[RECORDING]Task: Can another chemist read this recording and repeat the experiment exactly? List the missing information (amount, temperature, time, addition order, purification detail) one by one. Show only the missing information; filling it yourself.

4) Characterization summary (for the article):

Below are characterization data for a compound (NMR, IR, MS, yield):[DATA]Task: Arrange them in the "experimental part" format of a journal article (standard notation: yield, 1H NMR (solvent, frequency): ...,HRMS: ...). Changing my data; just format it.

Weak prompt / Strong prompt

Weak:

Write a report for this experiment: I synthesized aspirin.

AI makes up the gaps: quantities, observations, efficiency all become illusions; This is a fake record.

Strong:

Edit the experiment record with my ACTUAL data below:- 2.00 g salicylic acid, 5 mL acetic anhydride, 5 drops H2SO4.- 85 °C water bath, 15 min. It crystallized when it cooled. - It was filtered and washed with cold water. Dry product: 2.15 g. mp 134-136 °C.Task: Arrange in experimental notebook format; If there is a missing field, mark [MISSING]. Do not change or add my data.

Difference: real data given, AI only formatted, fabrication banned, omissions flagged.

Paper notebook, ELN and AI-powered workflow

Size

paper notebook

ELN

AI support

Searchability

low

high

Structures the text

constancy

physical signature

Version/timestamp

(Does not keep records)

speed

slow

medium

High (formatting)

Risk of error

illegibility

copy-paste

Risk of fabrication

best role

instant observation

permanent archive

Regulation/audit

mini cases

Case 1 — Missed deviation proved costly. A student accidentally ran a reaction at 80 °C instead of the planned 60 °C but did not record this. The product turned out unexpectedly; When it reoccurred three weeks later, no one knew the temperature difference. If there was a "plan deviation" field in the AI-supported ledger and it was filled in, the problem would be seen immediately. Lesson: record deviations; The most valuable information is there.

Case 2 — The danger of fabricated observation. One user told the AI ​​to "write the report observations as well"; YZ wrote "the solution was clear and colorless" but in reality the solution was yellow (a sign of a byproduct). User noticed and fixed it. Lesson: never make the AI ​​match the observation, it hides critical clues.

Case 3 — Repeatability audit won. A researcher asks his recording to the AI ​​"can someone else repeat it?" He had it inspected; YZ found that the order of addition (acid first or anhydride first) was unclear. This detail was important for efficiency. User added. Lesson: AI is invaluable as an under-detection checker.

Common mistakes

  • Making the observation fit the AI. This is scientific fraud and destroys critical clues.
  • Not recording deviations. Everything that is done differently than the plan is the most important information that explains the result.
  • Writing incomplete amounts. Phrases such as “somewhat,” “enough” destroy repeatability; Give mass/mole/volume.
  • Skipping lots/resources. The lot number of the reagent allows tracking problems caused by impurities.
  • "Beautifying" the recording afterwards. Correcting/polishing data after the fact is data fraud.
  • Not storing raw data. Raw records such as spectrum files and weighing slips should be archived.
Tip: A good rule of thumb: "The experimental notebook should be detailed enough that an outsider who doesn't remember anything six months later can repeat the experiment." Have AI check your registration with this eye.

In summary

  • Documentation is the foundation of repeatability; must be complete and honest.
  • AI formats, recalls and audits documentation; but you provide the observation and facts.
  • Never make the AI ​​adapt the observation; Be sure to record any deviations.
  • Use AI as a “missing checker”: is recording repeatable?
  • Store raw data; Beautifying the record later is forgery.

Application task

Take raw notes of an experiment you actually did (or know in detail about). Ask the AI ​​to generate an experiment notebook template, then insert your notes into it and mark any missing items (no faking). Then ask "can someone else repeat the recording?" Have the AI ​​check it and fill in the gaps found with real information. Try to catch at least three omissions/deviations. Complete the final record with date and signature.

checklist

  • [ ] My experiment record contains purpose, material, procedure, observation, result, deviation.
  • [ ] I write the observations myself, I do not have them adapted by AI.
  • [ ] I record every deviation from the plan.
  • [ ] I write quantities as mass/mole/volume, lot/source.
  • [ ] I have the AI ​​check the recording for repeatability.
  • [ ] I keep the raw data and do not beautify the recording later.
Previous Next