Part 2: How to Write Chapters 4 and 5
Part 1 is done. Now the practical part starts. Chapter 4 is the proposal — what you plan to do and why. Chapter 5 is the implementation — documenting exactly how you did it, with screenshots and step-by-step tool walkthroughs. Here’s how to approach both correctly, in line with the thesis structure and your research questions.
Chapters 1 through 3 established the theoretical ground — the technology, the literature, the case studies. Chapters 4 and 5 are where you show what you actually did. That shift in voice matters. You’re no longer reporting what others found. You’re designing and executing your own research. The writing style changes, the content changes, and the documentation standards change. Here’s how to approach each chapter correctly.
What This Guide Covers
Where You Are in the Thesis — and Why It Changes How You Write
Your thesis has eight chapters. Chapters 1–3 are theoretical. Chapter 3 ended with a cross-case analysis of three organisations using AI video. That analysis connects directly to Chapters 4 and 5 — you’ve now seen what others did, and you’re going to do something similar yourself and document it rigorously.
In Chapter 2, you cited sources for every claim. In Chapter 3, you described what others did. In Chapter 4, you switch to first-person (or the institutional first person) and write prospectively: “This chapter proposes…” / “The practical research will involve…” In Chapter 5, you write retrospectively: “The process began with…” / “Figure 4 shows the prompt interface after entering the commercial brief.” Getting this voice right signals to your examiner that you understand the difference between secondary and primary research.
Chapter 4 — What It Must Do
Chapter 4 is the proposal for your primary research. Think of it as a contract between you and the examiner: “Here is exactly what I am going to do in Chapter 5, here is why I chose to do it this way, and here is how I will know if it worked.” Three things. All three must be present.
A common mistake is to jump straight into tool descriptions without first stating the research goal clearly. Don’t do that. The examiner needs to understand what question you are answering before they care which tools you picked to answer it.
Goal → Tool Justification → Evaluation Design
These three are not optional — they’re the structure the chapter’s marking criteria are built around. Each one directly responds to one or more of your four research questions. Miss one and the chapter feels incomplete, regardless of how detailed the other two are.
Element 1 — Research goal: What commercial will you produce? For which audience? With what objective? What success looks like needs to be defined before you can evaluate whether you achieved it. A single specific brief — for example, “a 30-second AI-generated commercial for a fictitious sustainable fashion SME targeting 18–35 year olds on Instagram and YouTube” — is more useful than a vague statement about “exploring AI video tools.”Element 2 — Tool selection and justification: Which tools will you use and why? This must connect back to your Chapter 2 platform comparison. You already reviewed RunwayML, Synthesia, HeyGen, Pika Labs, and Sora. Now explain which subset you’ll use for the practical work and what each tool contributes to the production workflow. Each choice needs at least one paragraph of justification — citing the academic or platform documentation you already used in Chapter 2.
Element 3 — Evaluation approach: How will you assess the output? Since this is a qualitative thesis (not a quantitative survey), evaluation doesn’t mean collecting statistics. It means defining what criteria you’ll use to judge the commercial — professional standard, brand alignment, production cost, time-to-completion, and comparison against the traditional production cost benchmarks cited in Chapter 2. State these criteria here so Chapter 6 can refer back to them.
How to Write the Research Goal for Chapter 4
Your thesis title is “Using AI Video to Create Impressive Commercials at a Low Cost.” That’s your anchor. The research goal for Chapter 4 needs to operationalise that title for a specific, concrete production exercise. It can’t stay abstract.
Specific Brief + Research Rationale + Connection to RQ4
RQ4 asks: “What strategic principles should guide the integration of AI video into commercial advertising workflows?” Your practical production is the evidence base for answering that question. So the goal statement needs to make that connection explicit: you’re not just making a commercial for fun — you’re producing it in order to test the strategic and practical claims made in Chapters 2 and 3.
What to include in the goal statement paragraph: The commercial brief (product/service, target audience, platform, duration). The rationale for that brief — why this type of commercial allows you to test the research questions (e.g. an SME context tests democratisation, as in RQ2; a low-budget constraint tests cost claims from Section 2.6). The intended outcome — what “success” looks like, in non-quantitative terms. And an explicit statement that this practical research responds to the gap identified in Chapter 2’s critical synthesis: academic research into the practical implementation of AI video for commercial advertising is still limited.Selecting and Justifying Your Tools
You’ve already done the hard work here in Chapter 2. The platform comparison in Section 2.5 reviewed five tools. For Chapter 4, you’re not re-describing those tools — you’re making and defending a selection from them based on your specific brief.
| Tool | Role in Your Production | Why It Fits the Brief | Your Chapter 2 Reference |
|---|---|---|---|
| RunwayML Gen-2 | Generative scene footage — text-to-video and image-to-video clips | Versatility, compatibility with post-production, inpainting/editing tools. Established in Sections 2.5.1 and 3.2. | Sections 2.5.1, 2.6, 3.2 |
| Synthesia or HeyGen | AI avatar spokesperson narration | If commercial includes a human narrator, avatar tools remove talent costs entirely. Both reviewed in Chapter 2; choose based on brief specifics. | Sections 2.5.2, 2.5.4, 3.3 |
| Pika Labs | Short dynamic clips for social media assets | Motion control and style consistency for Instagram-format content. Four-second clip limit is a constraint to acknowledge. | Section 2.5.3 |
| ElevenLabs / Murf AI | Voice synthesis and narration audio | Professional voiceover without studio costs. Referenced briefly in Section 2.6 in the cost breakdown ($100–$1,000 total production range). | Section 2.6, Murf AI (2024) |
Chapter 4 is not a second platform survey. Sora isn’t in your toolkit — it was restricted to limited access at the time of the thesis and wasn’t available for practical implementation. Say that explicitly. It shows methodological awareness. The tools you select must be genuinely accessible on a low budget (student plan or free tier), which is itself evidence supporting the democratisation argument in RQ2.
Designing the Evaluation Approach
Since this is a qualitative thesis, your evaluation criteria can’t be “did engagement go up by X%.” You don’t have a live campaign. What you can do is define a set of qualitative benchmarks and assess your output against them systematically.
Suggested Evaluation Criteria
- Visual quality: Does the output meet a minimum broadcast-acceptable standard? Compare against the Chapter 2 definition of “professional-grade” commercial video.
- Brand alignment: Does the generated content reflect the brief’s brand guidelines (style, colour, tone)?
- Production cost: What was the actual cost (subscription fees + time)? How does it compare against the $25,000–$150,000 traditional benchmark from Section 2.6?
- Time-to-completion: How many hours from brief to final export? Compare to the 6–12 week traditional timeline.
- Technical limitations encountered: Document AI artefacts, consistency failures, or clip-length constraints that required workarounds.
How to Present Evaluation in Chapter 4
State each criterion as a named category. Give one sentence explaining what it measures and why it matters for answering the research questions. This becomes the evaluation framework Chapter 6 will use to discuss findings. If you define five criteria here, Chapter 6 should have five corresponding sections — that structural consistency is what makes a well-organised thesis.
- Visual quality ← RQ1
- Cost effectiveness ← RQ2
- Workflow efficiency ← RQ4
- Ethical/transparency considerations ← RQ3
- Technical constraints encountered ← RQ1, RQ4
Chapter 5 — What It Must Do
Chapter 5 is the implementation chapter. It’s the largest single chapter in the practical section — around 20% of total length. Its job is to document the production process in enough detail that a reader could replicate it. Every tool, every step, every prompt, every design decision needs to be traceable.
This is where the screenshots go. And screenshots have rules.
Pre-Production → Tool-by-Tool Production → Assembly → Output Assessment
Mirror the traditional production phases from Section 2.4 of your own thesis — pre-production, production, post-production — but document the AI equivalent of each phase. This creates a direct structural connection to your earlier theoretical work, which examiners appreciate.
Pre-production (AI equivalent): Writing the brief. Creating the script. Generating the storyboard (you can use AI image generators for this, or describe it textually). Selecting style references for prompt engineering. Documenting the initial prompts you developed.Production — tool by tool: One section per tool. For RunwayML: what prompts were used, what settings were applied, what the raw outputs looked like, which clips were kept and which were discarded and why. For Synthesia/HeyGen: avatar selection, script input, language settings, output. For voice tools: voice selection, script input, export format. Each section should have screenshots at each stage.
Post-production: How were clips assembled? What editing was applied? What final format was exported for which platform (Instagram 9:16, YouTube 16:9)? Document the tools used for assembly — even if it was Adobe Premiere or CapCut — and show screenshots of the timeline/editing interface.
Output section: Present the final commercial as a described output (since the thesis is a written document, you describe and screenshot it rather than embed a video). State the total cost, total time, and flag any AI artefacts or limitations in the final output.
How to Use Screenshots Correctly
Screenshots are evidence. They are not decoration. Every screenshot needs three things: a figure number, a caption, and a reference in the surrounding text. If none of those three are present, the screenshot weakens the chapter rather than strengthening it.
Caption Format, Numbering, Source Attribution, and In-Text Reference
Screenshots from third-party tools are not your own work — they require source attribution. Caption format: “Figure 5.1: RunwayML Gen-2 interface showing prompt input for Scene 1 of the commercial brief. Source: RunwayML (2024).” The figure number follows the chapter numbering (Chapter 5, Figure 1 = 5.1). The in-text reference comes before the screenshot appears: “As shown in Figure 5.1, the prompt was entered using…” Not after. Not instead of. Before — so the reader knows what to look for before they look.
What to screenshot and when: The interface before you enter a prompt (establishing the tool’s baseline state). The prompt entered (so the exact wording is documented). The raw output generated. Any editing steps applied to the raw output. The final clip as it will appear in the assembled commercial. Five screenshots per major tool is a reasonable baseline — more if you had multiple iterations, which you should document as evidence of the prompt engineering process.Prompt input: “Slow-motion close-up of sustainable fabric in natural light, earthy tones, editorial fashion aesthetic, 4K, cinematic”
Settings: 16:9 ratio · 768px width · Motion: Medium · Style: Cinematic
The caption isn’t just a label. It’s an opportunity to connect the screenshot to the theoretical content from earlier chapters. If the prompt you’re showing uses the style-descriptor technique described in the EcoWear case study (Section 3.2), say so. If the interface shows a pricing tier that matches the cost range cited in Section 2.6, say so. Each caption that makes a cross-reference shows the examiner that your practical work is grounded in the theoretical framework — that’s exactly what they’re looking for in a good master’s thesis.
Documenting Each Tool’s Workflow Step by Step
Here’s how to approach the documentation for each major tool in your production toolkit. These are not instructions for how to use the tools — they’re suggestions for what to document and how to structure each section.
RunwayML Gen-2 — Generative Scene Production
Start with the brief for this scene. What visual were you trying to create? What was the first prompt you tried and what did it produce? Show the raw output (screenshot or still frame description). Then show the refined prompt and compare. Document how many iterations it took to get commercially usable footage. Note: the EcoWear case in Section 3.2 found that generic prompts required significant prompt engineering before producing brand-aligned results — your documentation should either confirm or challenge that finding, which feeds directly into your Chapter 6 analysis.
Synthesia or HeyGen — Avatar Spokesperson
Document avatar selection (why this avatar? does it match the brand brief?). Show the script input interface. Show the avatar preview. Document any adjustments to speech speed, tone, or background. If you used a custom voice or translated the script, document that step separately — it directly addresses the multilingual capability discussed in Section 2.5.2 and the cost implications in Section 2.6. HeyGen’s plans ($29–$89/month) should be cited in your cost tracking for this chapter.
Pika Labs — Short Dynamic Clips
Pika’s four-second clip limit is a documented constraint (Section 2.5.3). Show how you worked within it — multiple four-second clips spliced together, or using Pika for specific hero moments only while RunwayML handles longer sequences. Document the motion control settings. Note any style consistency issues between clips generated in separate sessions — this is a known limitation of current AI video tools and your documentation of encountering it is valuable empirical evidence.
Voice Synthesis — ElevenLabs or Murf AI
Voice selection rationale. Script input. Output format (MP3 at what sample rate). Any adjustments to tone or pacing. The final audio file’s role in the assembly. Cost: what subscription tier was used and what did it cost? This feeds into the total cost calculation that closes Chapter 5. Remember that Murf AI (2024) is already in your reference list from Chapter 2 — use the same citation, don’t create a new entry.
Assembly and Final Export
Which editing tool did you use to assemble the clips? Screenshot the timeline with all clips visible and labelled. Document the export settings (resolution, frame rate, format) for each platform version (YouTube 16:9, Instagram 9:16). State the total file size and duration of the final commercial. This is where you close the production log and set up the evaluation section — the next thing the examiner should read is how the final output performed against the criteria defined in Chapter 4.
Create a cost summary table at the end of Chapter 5. List every tool, the plan used (free/student/paid), the cost for the production period, and the unit produced. Total it. Compare it to the Section 2.6 benchmark ($25,000–$150,000 for traditional production). This table is one of the most important pieces of evidence in your entire thesis — it’s the direct empirical test of the “low cost” claim in your title. A table that shows total production cost of $47 or $180 against a $50,000 traditional benchmark is powerful, concrete evidence that addresses RQ2 directly.
Example format: RunwayML (Basic plan, $15/month) — $15. Synthesia (Free trial, educational) — $0. ElevenLabs (Starter, $5/month) — $5. Time cost (your hours at zero monetary cost, but document number of hours). Total cash spend: $20. Total hours: 14. Equivalent traditional production: $50,000–$150,000.
Mistakes That Weaken These Chapters
Writing Chapter 4 Like a Tool Manual
Spending most of Chapter 4 re-describing RunwayML, Synthesia and HeyGen in detail that already appeared in Chapter 2. Chapter 4 is about your decisions — not another tool overview. Cross-reference Chapter 2 with “as discussed in Section 2.5.1” and move on to justification.
Justify the Selection, Don’t Repeat the Description
One sentence naming the tool. One sentence referencing where you reviewed it. Then three to four sentences on why it’s the right fit for your specific commercial brief and how it addresses one of your four research questions. That’s the Chapter 4 formula for each tool.
Dropping Screenshots Into the Text Without Introduction
A screenshot that appears before any explanatory text is confusing. A screenshot that appears with only a one-word caption (“RunwayML”) is nearly useless. And a screenshot that isn’t mentioned anywhere in the surrounding paragraphs is just a placeholder that found its way into a final document.
Introduce, Show, Explain — In That Order
Text before the screenshot explains what you were trying to do. The screenshot shows what happened. The caption names it, sources it, and connects it to prior work. Text after the screenshot (if needed) interprets what’s visible. Three parts. Every time.
No Mention of What Didn’t Work
A Chapter 5 that only shows successful outputs looks like a sales brochure, not a thesis. Every AI video project encounters artefacts, inconsistencies, clip-length failures, or avatar uncanny valley moments. Documenting these isn’t a weakness — it’s evidence that directly addresses RQ1 (capabilities and constraints) and RQ4 (strategic principles).
Document Failures Explicitly
Screenshot the first prompt that didn’t work. Caption it: “Figure 5.2: Initial RunwayML output from generic prompt, showing style inconsistency with brand guidelines. Source: RunwayML (2024).” Then show the refined prompt that produced a better result. The iteration is the research. The examiner wants to see your problem-solving, not just your final product.
Using Wikipedia or Platforms’ Own Marketing Pages as the Only References
Chapter 5 still requires citations when you describe tool capabilities. “RunwayML allows inpainting and outpainting” needs a citation — RunwayML (2024) from your existing reference list. Relying only on the platform’s marketing copy without cross-referencing the academic sources already in your bibliography weakens the scholarly quality.
Use Your Existing Reference List
You already have citations for every tool in your Harvard reference list — RunwayML (2024), Synthesia (2024), HeyGen (2024), Pika Labs (2024), Murf AI (2024). Use them consistently in Chapter 5 when you describe tool outputs. Cross-reference academic sources from Chapter 2 (Rubin, 2023; Davenport et al., 2020) when you make analytical claims about what the implementation demonstrates. Your bibliography is a resource — use it.
No Connection Back to Research Questions in Chapter 5
A Chapter 5 that reads as a pure production log — “then I did this, then I did that” — without connecting the steps to the research questions, misses the point of a thesis. You’re not writing a how-to guide. You’re conducting research.
Signal RQ Connections in Sub-Headings or Paragraph Openers
Add one sentence at the start of each tool section: “This stage of the implementation directly addresses RQ1 by demonstrating the practical constraints of RunwayML’s clip-length limitation in a commercial advertising context.” That sentence takes 10 seconds to write and immediately lifts the analytical quality of the entire section.
The practical thesis structure used in these chapters — research proposal followed by implementation documentation — follows the framework described in Yin, R.K. (2018) Case Study Research and Applications, 6th edn., Sage, which is already in your reference list from Chapter 3. Yin’s guidance on documenting implementation processes with evidence (Chapter 4 of his text) directly supports the screenshot and process-log approach described here. For Harvard citation format guidance applicable to your tool citations, see Cite This For Me — Harvard Referencing Guide as a free cross-check tool.
Frequently Asked Questions
Need Help Writing Your Thesis Practical Chapters?
Chapter 4 proposals, Chapter 5 implementation write-ups, screenshot documentation, cost analysis, evaluation frameworks — our dissertation team works with master’s students across all disciplines.
Dissertation Writing Service Get StartedBefore You Start Writing Chapter 4
Re-read your research questions. All four of them. Print them out and tape them to your monitor. Every paragraph you write in Chapters 4 and 5 should serve at least one of those four questions. If a paragraph doesn’t, either cut it or add a sentence that makes the connection explicit.
Then write the commercial brief before anything else. One paragraph. Specific. Give the fictional brand a name, a product, a target audience, a platform, a duration. Once you have that brief, everything else in Chapter 4 flows from it: the tools you select are justified by the brief, the evaluation criteria are defined by the brief, and the success benchmarks are measurable against the brief.
Chapter 5 practically writes itself if you document as you go. Don’t produce the commercial first and then try to reconstruct the process from memory. Keep a production log — a simple document where you record what you tried, what setting you used, what it produced, and what you changed next. That log is Chapter 5. The screenshots are the evidence. The analysis is the connection back to your research questions.