Source Criticism
Federal Council and Council of Europe 1949
This exercise was translated automatically from the German original and can contain errors. Consult the original version in case of doubt.
Overview and Didactic Goal
This exercise teaches transferable competencies in source criticism (external and internal criticism) along one concrete source from the Diplomatic Documents of Switzerland (Dodis). The central question is how you treat a source as an object of historical evidence: not as a “store of facts”, but as a situated artifact with a context of creation and a history of transmission.
Case study (primary source): Dodis: Switzerland and the Council of Europe
The source that we use here exists in an edited digital edition. For source criticism, the object of study therefore holds not only the text, but also the edition layer (metadata, archive reference, links, editorial guidelines). (Fickers and Tatarinov 2022)
At a Glance
| Item | Value |
|---|---|
| Duration | about 180 minutes |
| Level | Intermediate |
| Core path | 1 → 2 → 3 → 5 → 6 |
| Optional deepening | 4 (contextualization and counter-check) |
| Hand-in (minimum) | Source analysis log on one source. AI log: 6 entries. |
The core path leads you to a complete source criticism in about 180 minutes. Step 4 costs about 90 minutes more, because you search for further sources.
The two logs have different tasks. The source analysis log describes the source. The AI log describes your AI use. Keep the two separately.
What You Need
Subject knowledge
- A basic understanding of historical research methods
- Basic knowledge of work with generative AI, above all prompting
If prompting is new to you, we recommend that you work through the exercise Prompt Engineering first.
Capabilities of the AI system
The system must be able to do the three following things:
- Read a file that you upload, and quote from this file. You need this from step 1.
- Deliver an answer in a format that you prescribe, for example as a table with a claim and its evidence.
- Search the web and support each finding with an address. You need this only in step 4.
Other tools and access
- The source as a file. You download the document Dodis 5020 freely, and you need no institutional access for it.
- One file for your source analysis log and one file for your AI log.
You can do every exercise with AI systems from different makers. What a system can do depends on the product and on your account. Settle the five following points before you start.
Cost
A free account is enough for short text tasks. For some capabilities you need a paid account. This mostly concerns the search on the web, the reading of long files and the built-in reasoning mode. Check first whether your university provides access.
Tools
Check which tools your account unlocks. Can the system search the web? Can it read a file that you upload? Can it call external services? Each exercise names the capabilities that it requires.
The section Check and classify capabilities explains five newer capabilities. It also describes what each capability means for your work with sources.
Memory
Some systems carry over statements from earlier sessions or from a project context. Check the settings for memory, projects and chat history. A new session therefore guarantees no empty context.
Account: institutional or private
Check which account you work with. The rules differ in three points.
| Point | Account of your institution | Private account |
|---|---|---|
| Contract | Your university concludes it and shares the liability. | You conclude it yourself and carry the liability alone. |
| Training | A contract often excludes the use of your inputs. | Many providers use inputs for training. |
| Permitted material | The university rules which data you can enter. | You have no such rule and carry the risk yourself. |
Ask the IT office or the library of your university when you cannot find the rules. Record the answer in your AI log.
Do not upload material that you have no right to pass on. An upload leaves your computer and you cannot take it back.
These materials stay outside the AI system:
- Personal data of living persons, for example from interviews, from private papers or from holdings under a protection period.
- Digitized items whose terms of use forbid a disclosure.
- Unpublished texts of other persons, for example manuscripts or reviews.
- Licensed full texts, when the license of your library forbids a disclosure.
In case of doubt, work with a short excerpt or with an anonymized version. You can also use a local model, because the material then stays on your computer.
Law: data protection and disclosure
Two areas of law concern your work directly. This section gives an orientation and no legal advice.
Data protection. In the EU the General Data Protection Regulation applies. In Switzerland the revised Data Protection Act applies (Verordnung (EU) 2016/679 Zum Schutz Natürlicher Personen Bei Der Verarbeitung Personenbezogener Daten (Datenschutz-Grundverordnung) 2016, Bundesgesetz Über Den Datenschutz (DSG) 2020). Both laws protect personal data of living persons. The text of the law itself is binding. Article 6 GDPR requires a legal basis for every processing operation, and an upload is a processing operation.
The role of the provider decides the classification. A provider can act as a processor and process your data only on your behalf. Article 4 point 10 GDPR expressly excludes a processor from the third parties. This role needs a contract under Article 28 GDPR. Without such a contract, a provider uses your inputs for its own purposes. An upload is therefore not always a disclosure to a third party, but without a contract you must assume that it is. Check the contract that applies to your account.
Research is subject to safeguards, but it gets no exemption. Article 89(1) GDPR requires appropriate safeguards, for example data minimisation and pseudonymisation. Article 89(2) GDPR permits the legislator to derogate from single rights of the data subject. The information of the data subjects can also fall away, namely under Article 14(5)(b) GDPR. That applies only when the information proves impossible or would involve a disproportionate effort. Research therefore does not release you from the information duty in general. Settle your individual case with the data protection office of your university.
The European Data Protection Board presented a guideline on research in 2026 (European Data Protection Board 2026). This guideline was in public consultation until June 2026. It is therefore not applicable law, but an aid to interpretation. For deceased persons, the protection periods and the rules of use of the archive apply.
Disclosure. The AI Act of the EU governs the transparency of generated content (Verordnung (EU) 2024/1689 Zur Festlegung Harmonisierter Vorschriften Für Künstliche Intelligenz (Verordnung Über Künstliche Intelligenz) 2024). Article 50 applies from 2 August 2026. It separates two addressees: the provider of an AI system and the deployer. You are a deployer as soon as you use an AI system professionally. For purely private use you are not a deployer, under Article 3 point 4.
As a deployer you carry two separate duties from Article 50(4).
- Deep fake. If an AI system generates or manipulates an image, an audio recording or a video, and the content is a deep fake, then you disclose that. Under Article 3 point 60, a deep fake resembles existing persons, places or events and falsely appears authentic. A generated image without this resemblance to reality does not fall under this duty.
- Text on public matters. If you publish generated text to inform the public on a matter of public interest, then you disclose that. The duty falls away when a person reviewed the text and holds the editorial responsibility for the publication.
Article 50(4) therefore requires no disclosure for every generated image and for every generated text. Article 50(2) addresses the provider instead. The provider must mark generated content in a machine-readable format. This duty does not lie with you, and it does not replace your own disclosure.
The legal disclosure and the academic disclosure are two separate duties. Your university requires a declaration even when the law requires no disclosure. The exercise Citing shows how you declare your AI use.
Record your answers to these five points. They belong to the description of your work environment.
Learning Objectives
After the exercise you can:
- apply external and internal source criticism systematically,
- check the authenticity, the provenance and the transmission of a digitally edited memorandum,
- reconstruct authorship, horizon and standpoint, intention and bias hermeneutically,
- tell witting and unwitting testimony apart,
- determine the evidence value of the source in relation to a research question precisely,
- operationalize the “veto power of the source” (what does it exclude? what does it leave open?),
- use AI systems critically and under methodological control for source criticism and for context work.
Structure of the Exercise
Each step holds:
- an objective,
- concrete tasks, including optional AI prompts,
- a work assignment and a reflection assignment.
- Orientation: source type, minimal paraphrase, first research heuristic
- External criticism: provenance, authenticity, transmission, edition
- Internal criticism: speaker position, intention, argument logic, bias
- Contextualization and counter-check: triangulation with further sources
- Evidence value and “veto power”: what holds, and what does not?
- Documentation and reflection: reproducibility, AI log
1. Orientation: What is This Source (and What Can It Stand For)?
Goal
A first controlled classification of the source as an object: type, situation, constellation of actors, thematic focus.
Task 1a: Minimal Paraphrase Without AI
Formulate in at most 4 sentences:
- Who speaks or acts?
- What is the occasion?
- What is the central statement or decision?
- Which consequence is indicated?
Task 1b: The Same Minimal Paraphrase with AI
Analyse the attached memorandum (Bern, 7.7.1949, Petitpierre/Hansen, Council of Europe) and answer the following questions:
Who speaks or acts?
What is the occasion?
What is the central statement or decision?
Which consequence is indicated?Check before the upload whether you have the right to pass the material on. The section “What you need” names the materials that stay outside the AI system. Free access alone permits no upload: a freely accessible source can still forbid a disclosure. Check therefore the terms of use of Dodis before you upload the digitized item.
Upload the document as a PDF file together with the prompt. That is the normal case. Make sure that the AI also has access to the metadata. Retrieve the metadata online, or state it yourself.
Only in the exceptional case, when the upload does not work, copy the content of the document to the start of the prompt. Separate it visually, for example with “““. Example:
"""
DOCUMENT CONTENT
"""
"""
DOCUMENT METADATA (incl. URL)
"""
YOUR PROMPTThis exercise uses an edited version with a checked text. For a scan or for a handwritten original, the situation differs. The system then reads an image and produces text from it.
A general AI system does not stop at an illegible passage. It inserts a plausible word and marks nothing. The error rate of handwritten text recognition depends strongly on the script, the layout and the language (Romein et al. 2025; Díez García et al. 2025). Such errors change every later analysis (Strien et al. 2020).
Compare therefore every citation with the digitized item. Note in the source analysis log whether a machine recognized the text. The section Check and classify capabilities explains this point in detail.
Work Assignment (Reflection)
Compare the answer of the AI with your own. Which differences do you find?
Do you see systematic biases, in your answers and in the AI answers?
Task 2a: First Research Heuristic Without AI
Note three possible research questions that you could in principle work on with this source, still without a delimitation. Examples of question types:
- conceptual history (neutrality as an argument)
- institutional history (decision paths in foreign policy)
- history of knowledge (what counts as “compatible” or “incompatible”?)
- political and cultural history (expectations towards referendums)
Task 2b: The Same Research Heuristic with AI
Give me three possible historical research questions that follow from the attached memorandum (Bern, 7.7.1949, Petitpierre/Hansen, Council of Europe).
Formulate each question so that it is (a) limited in time, (b) verifiable against sources and (c) analytical.
For each question, name the additional source types that I need for the counter-check.Work Assignment (Reflection)
- Which of your three questions depends most strongly on this specific source (high binding to the source)?
- Which question depends least on the source (you would formulate it “similarly” with other sources)?
- Which terms already carry an interpretation (for example “neutrality”, “Europe”, “compatibility”)?
2. External Source Criticism: Authenticity, Provenance, Transmission, Edition
Goal
Clarify what this document is as an object, how it came down to us, and which edition layer lies between the original and you.
Task 1: Provenance Check (Metadata and Archive Reference)
Collect from the Dodis page and record:
- the document type or genre (memorandum, minute, report, letter)
- the date, the time and the place
- the author or signatory (here: Max Petitpierre)
- the addressee or addressees, explicit or implicit
- the archive reference (for example “E 2800(-)…”)
- the institutional context (department, commissions, participating organizations)
- the tags, persons, organizations and geography (as edition metadata)
Result: a small table and 5 to 8 lines of “edition notes” (What does the edition deliver? What stays open?).
Task 2: Authenticity as a Question Form
Formulate three verifiable questions on authenticity, for example:
- Is this an original, a copy or an edited version?
- Are there signs of later editing (cuts, translations, editorial normalization)?
- Which parts are document text, and which are editorial additions (footnotes, cross-references)?
Task 3: Transmission as an Epistemic Problem
Reconstruct a plausible chain of transmission:
Creation → filing in the administrative process → archiving → cataloguing → digitization and edition → display on the web
Note at each station:
- the possible losses and filters (logic of selection)
- the possible points of transformation (editing, classification, assignment of metadata)
- the possible losses of context (missing enclosures, missing prior history)
Work Assignment (Reflection)
- Which information comes from the document text (primary text), and which comes from the edition (metadata, links)?
- Which parts of your classification would be impossible without the Dodis edition?
- Where is the edition layer itself a source (for example categorization, tags, selection)?
Task 4: External Criticism with AI
Analyse the digitally edited memorandum (Dodis 5020, Bern 7.7.1949) only from the viewpoint of external source criticism.
1) Identify on the basis of the metadata:
- the document type or genre
- the date and the place
- the author or signatory
- the institutional context
- the archive reference and the form of transmission
2) Separate strictly between:
a) information from the document text
b) information from the edition and metadata layer (Dodis)
3) Formulate three verifiable questions on authenticity and transmission.
Important:
- Make no historical evaluations.
- Mark each statement explicitly as:
[document text] / [edition or metadata] / [conclusion].Reflection Assignment
- Where does the AI add systematically, and where does it hallucinate assumptions about the transmission?
- Which metadata does it adopt correctly, and which stay vague or wrong?
- How far does the explicit marking of the levels force methodological clarity?
3. Internal Source Criticism: Intent, Standpoint, Argument Logic, Bias
Goal
Analyse how the text produces meaning: through selection, wording, implicit premises and the institutional perspective.
Task 1: Speech Act and Function Analysis
Determine the function of the memorandum in the administrative context, with a plausible argument:
- Does it record an event “for the file”?
- Does it justify a position towards third parties?
- Does it steer the internal formation of opinion?
- Does it mark a red line (neutrality as a stop signal)?
Result: 1 paragraph (6 to 10 sentences) that argues the function and the addressing, with evidence from the text.
Task 2: Argument Mapping (Neutrality as the Key Operator)
Extract all passages on neutrality, compatibility or decision processes, and break them into:
- claim
- warrant
- assumption
- implication (so what)
Format suggestion (Markdown):
- Claim:
- Warrant:
- Assumption:
- Implication:
- Evidence in the text (quotation or paraphrase):Task 3: Witting Against Unwitting Testimony
Mark two examples for each of:
- witting testimony (deliberately addressed communications, statements of a position)
- unwitting testimony (things taken for granted, implicit norms, institutional routines)
Guiding questions:
- Which self-images of Switzerland does the text presuppose (for example “a more absolute neutrality”)?
- Which political mechanisms count as expected (parliament, the people, the cantons)?
- Which comparison does the text use (Sweden), and for what?
Task 4: Bias as a Structure, Not as an “Error”
Analyse the limits of the perspective:
- Which interests and constraints does the author carry as the head of a department?
- Which aspects are missing systematically (for example legal dimensions, international reactions, domestic political camps)?
- Which alternative readings would be plausible from other positions (for example party politics, the economy, the press)?
Work Assignment (Reflection)
- Which of your readings are close to the text (with evidence), and which are hypotheses that depend on context?
- Which terms work as a hinge between the text and the context (neutrality, compatibility, referendum)?
Task 5: Internal Criticism with AI
Carry out an internal source criticism of the attached memorandum (Bern, 7.7.1949, Petitpierre/Hansen).
Analyse separately:
1) the speaker position and the institutional standpoint
2) the function and the intention of the memorandum in the administrative context
3) the argument logic around neutrality and decision processes
4) witting against unwitting testimony
Rules:
- Every analytical statement must carry concrete evidence in the text
(a short quotation or a precise paraphrase).
- If evidence is missing, mark the statement expressly as a "hypothesis".
- Use no context information that the text does not carry.Reflection Assignment
- Which readings of the AI are close to the text, and which are speculative?
- Where does the AI level institutional perspectives into general statements?
- Which hermeneutic steps stay impossible to automate here as well?
4. Contextualization and Counter-Check: Triangulation Instead of “Context as Décor”
Goal
Check the force of the memorandum through a comparative embedding: what is specific? what is routine? what is strategic?
Task 1: Intertextual Search for Traces (Laid Out in the Document)
The memorandum points indirectly to further contexts, for example commissions, private federalist networks and the comparison with Sweden.
Build a research matrix with at least 6 rows:
| Trace in the text | What exactly to search? | Expected source type | Where to search? | Purpose of the counter-check |
|---|
Examples for “Where to search?”, as categories and not as a complete list:
- further Dodis documents (cross-references, dossiers)
- official publications (messages, reports)
- parliamentary minutes and commission minutes
- newspapers (reception, controversy)
- records of private committees and associations
Task 2: Counterfactual Counter-Question (Control Instrument)
Formulate two counter-questions that challenge the memorandum:
- What would I have to find so that the thesis “neutrality prevented the accession” becomes too crude?
- What would I have to find so that the statement on the referendum looks like strategic rhetoric and not like a forecast?
Work Assignment (Reflection)
- Which kind of context is necessary to check claims, and which one only makes a claim plausible?
- Where does “context” risk becoming a later smoothing?
Task 3: Planning the Search and the Counter-Check with AI
Derive a systematic research plan from the memorandum (Bern, 7.7.1949, Petitpierre/Hansen).
1) Name at least:
- 5 additional primary sources (different genres)
- 3 types of secondary literature
2) Assign each source to a concrete purpose of verification
(for example confirmation, contextualization, refutation).
3) Formulate one guiding question per source.
Constraints:
- No statements about the actual content of these sources.
- No narrative syntheses.
- Focus only on verification steps and counter-checks.Reflection Assignment
- Which contexts does the AI prioritize systematically (politics, institutions)?
- Which contexts are conspicuously missing (for example media, law, economy)?
- Where does the AI tend to use “context” as an explanation instead of as a means of verification?
5. Evidence Value and “Veto Power”: What Does the Source Support, and What Not?
Goal
Determine the evidence value strictly relationally: source × research question.
Task 1: Evidence Profile (Claim Scope)
Build a list in two parts:
A. Well supported (high certainty): statements that the memorandum carries directly, with evidence in the text.
B. Not supported or open (veto): statements that the memorandum suggests, but does not support, or even excludes.
Guiding questions:
- What can I say about the position of Petitpierre, and what not?
- What can I say about the Danish position, and what stays unclear (for example motives, internal debates)?
- What does the source say about “Switzerland”, and where is that an impermissible generalization?
Task 2: Micro-Hypotheses with a Path to Falsification
Formulate two micro-hypotheses of one sentence each, for example:
- “In 1949, neutrality works as an institutional gatekeeper term that blocks a rapprochement with the Council of Europe politically.”
- “The reference to the referendum serves as an argument of legitimation towards foreign interlocutors.”
For each hypothesis:
- Which additional source could support it?
- Which source or observation would refute it?
Task 3: Evidence Profile with AI
Build two separate lists on the basis of the memorandum:
A) statements that the text supports directly and unambiguously.
B) statements that the text does not support, only suggests, or excludes implicitly.
For each statement:
- short evidence in the text, or a paraphrase
- rating: certain / probable / speculative
- reason for the rating
Addition:
For two statements from list B, formulate one
concrete path to falsification each (which source would have to show what?).Reflection Assignment
- Where is the AI surprisingly restrictive, and where is it too generous?
- Which statements look “plausible”, but get a correct veto?
- How does the explicit split into A and B change your own urge to interpret?
Work Assignment (Reflection)
- Which statement “would you like to make”, although the source does not permit it?
- Where is the border between interpretation and claim?
6. Documentation and Reflection: Reproducibility, AI Log
Goal
Source criticism as a documented method: traceable, verifiable, reusable.
Task 1: Source Analysis Log (Minimum Standard)
The source analysis log records the source, and not your AI use. You record your AI use separately in the AI log.
Set up a short log:
- the full citation or permalink of the source
- the date of access
- the version or display form (HTML, PDF, export)
- your research question (working version)
- the external criticism (bullet points)
- the internal criticism (bullet points)
- the evidence profile (list A and list B)
- the open questions and the next search steps
Task 2: AI Log (At Least 6 Entries)
Note:
- Which prompts did you use (verbatim)?
- Which answers helped you, and why?
- Where did you find shortenings or overgeneralizations?
- Which steps stayed unavoidably human decisions (for example the choice of the research question, the weighting of evidence, decisions about context)?
A good source criticism is not the one that “explains everything”. It is the one that makes limits and verification steps visible: what has support? what is hypothetical? what would you have to check next?
Learning Outcome
At the end you hold:
- a reasoned external and internal source criticism of Dodis 5020,
- an evidence profile, with the veto points, in relation to a research question,
- a research plan for the context and the counter-check,
- a reproducible source analysis log, with a structured extraction of claims,
- an AI log with at least 6 entries,
- a considered judgement on where AI supports exploration and where you must limit it methodologically.
Further Resources
- Ranke.2: Wikipedia as Historical Source – Provenance, revision histories and digital source criticism.
- Knowing Machines: Models All the Way Down – Critique of training data and its parallels to source criticism.
- Stark (2023) – Reads machine learning as a conjectural science that infers from traces; discusses the epistemological risks of automated pattern recognition.
- Lang and Suárez Cronauer (2026) – Ethical data work in the (digital) humanities; relevant to data provenance and dataset documentation.
Bibliography
Citation
@inreference{mähr2025,
author = {Mähr, Moritz},
title = {Source {Criticism}},
booktitle = {Critical AI Literacy for Historians},
date = {2025-12-29},
url = {https://maehr.github.io/critical-ai-literacy-for-historians/en/exercises/source-criticism-federal-council-council-of-europe-1949.html},
langid = {en}
}