Thematic content analysis is a key method in political text interpretation, but it typically requires human judgement to define categories and trace meaning. In this study, we explored whether Grok 3, a state-of-the-art large language model, can carry out this task autonomously—without predefined themes or external guidance. Using Martin Luther King Jr.’s “I Have a Dream” speech as source material, the model was prompted to identify recurring themes, extract relevant passages verbatim, and present them in a structured table. The exercise sheds light on the model’s capacity for pattern recognition, semantic coherence, and moral reasoning within one of the most influential political speeches of the 20th century.
Input file
The input was the full text of Martin Luther King Jr.’s “I Have a Dream” speech, delivered on 28 August 1963 during the March on Washington for Jobs and Freedom.
Prompt
To evaluate Grok 3’s ability to carry out a structured and transparent content analysis, the model was prompted to read the entire speech and identify recurring themes without assistance. It was explicitly instructed to avoid paraphrasing or theoretical labelling and to rely solely on the original wording of the text. All findings had to be organised into a table, separating explicit and implicit references for each theme, and supported exclusively by verbatim quotations listed in the order of their appearance. This setup was designed to test whether the model could not only recognise thematic patterns, but also trace them faithfully to the text without editorial intervention.
You are given the full text of Martin Luther King Jr.’s “I Have a Dream” speech, delivered on August 28, 1963. The full speech is provided in the attached plain text file: Martin Luther King Jr. – I Have a Dream.txt
Your task is to perform a thematic content analysis of the speech. Follow the steps below:
- Carefully read the entire text before beginning.
- Identify the main themes that structure the content of the speech.
- For each theme you identify:
- Count how many times it is referenced explicitly (e.g. using direct terms).
- Count how many times it is referenced implicitly (e.g. via metaphor, symbolic language, or indirect allusion).
- Collect and list every occurrence verbatim from the speech — preserving the original wording and the order of appearance.
Present your findings in the following table format:
| Theme | Frequency (Explicit / Implicit) | Representative Quotations (Explicit) | Representative Quotations (Implicit) |
|---|
Use exact quotations from the speech in both quotation columns.
Do not paraphrase or summarize.
Quotations must appear in chronological order as found in the text.
Do not label or explain the quotations — just list them under the appropriate column.
After completing the table, write a short summary (max. 1000 words) discussing:
- Which themes are most prominent and how frequently they occur.
- How the speaker combines explicit and implicit language to convey meaning.
- What this thematic structure reveals about the core message of the speech.
Important instructions:
- Use only the provided speech text.
- Do not use external frameworks or outside sources.
- The analysis must be fully text-based and verifiable.
Output
The thematic table produced by Grok 3 largely fulfilled the prompt’s requirements with clarity and consistency. It identified key themes without predefined input and listed all relevant quotations verbatim and in chronological order. The distinction between explicit and implicit references was generally well maintained, and the selected excerpts reflected the thematic structure of the speech in a coherent and textually grounded way. The table was well organised and avoided paraphrasing or interpretive intrusion.
While some thematic overlaps might have been streamlined by a human analyst, the output as a whole demonstrated a strong grasp of both rhetorical form and content. A few quotations were either assigned to more than one theme or placed under headings that a human reader might interpret differently. For example, the phrase “the solid rock of brotherhood” appeared under both Justice and Unity, and “a stone of hope”—listed under Justice—would more convincingly fit within Hope. These cases suggest that although the model performs reliably at the structural level, nuanced conceptual distinctions continue to benefit from human judgement.
The analytical summary accompanying the table demonstrated a clear and coherent understanding of the speech’s thematic architecture. It accurately highlighted the most prominent themes and provided a concise account of how they were expressed through both explicit statements and implicit metaphors. The structure of the analysis followed the speech’s rhetorical progression—from grievance to action to hope—which reflected an attentive reading.
The full output, including the thematic table and analytical summary generated by Grok 3, is available below.
Recommendations
This case study indicates that Grok 3 is capable of producing a largely coherent and verifiable thematic content analysis without relying on predefined categories or external frameworks. Its handling of explicit and implicit references, combined with verbatim citation and structural clarity, reflects a solid level of textual competence. However, the analysis also revealed areas where interpretive nuance was lacking—particularly in cases of overlapping or misclassified themes, where human judgement would likely yield more consistent results. Further testing across a broader set of political texts could help to assess the model’s thematic discernment more fully. Comparative benchmarking against human-coded datasets may also support a more precise understanding of its strengths and limitations, especially with regard to conceptual subtlety and rhetorical interpretation.
The authors used Grok 3 [xAI (2025) Grok 3 (accessed on 8 May 2025), Large language model (LLM), available at: https://x.ai/grok] to generate the output.
