AI Interview Prep in 15 Minutes – A businessman with a laptop, surrounded by floating digital documents
|

How I Prepared for a PhD Interview in 15 Minutes — With AI

ByViktor Szekeres

I received seven interview questions from a PhD researcher and used artificial intelligence to prepare for the interview on the stock market presence of Hungarian SMEs. In 15 minutes, the AI compiled a 15-page, academic-level preparatory document—drawing from 16 public media sources, 11 podcasts, 6 internal email threads, and the company’s SharePoint documents. I’ll show you how I did it.

I’m not the type of person who sits down and spends hours searching through my own past interviews. But when a PhD researcher sends me seven complex questions about IPOs, SME competitiveness, and ESG, I need to know what I’ve said about these topics before—so I don’t contradict myself and can remain consistent.

This task used to take days. Now it only took 15 minutes.

What the AI Found — By the Numbers

Before I get into the details, let’s take a look at the results. In 15 minutes, the AI found and processed the following sources for the 7 interview questions:

  • 16 public media sources: Portfolio.hu, Telex.hu, Világgazdaság, Economx.hu, CMBP.hu, Fintech.hu, Novekedes.hu, KKV Magazin, ITBusiness.hu, G7, and other articles
  • 11 podcast episodes: Company Builders (MiniCRM), VG Arbitrage, VG Exclusive, Moneyvlog — available on Spotify, Apple Podcasts, and YouTube
  • 6 internal email threads from my Outlook inbox—including an email about AI strategy sent to a management group
  • SharePoint internal documents: sustainability report, ESG strategy, investor presentations, BSE ESG Pilot materials
  • LinkedIn posts and BSE stock exchange announcements

He compiled a coherent document backed up by citations from a total of 40+ sources—in just 15 minutes.

What tools did I use?

The entire workflow took place within a single tool: Claude AI(Anthropic, Opus 4.6 model). But not on its own—rather, in conjunction with the ecosystem of plugins I’ve built over the past few months.

The AI utilized the following capabilities simultaneously:

  • Web search (deep research): Real-time scanning of 16 public media sources and 11 podcasts, content extraction from full-text articles
  • Microsoft 365 Connector: Search Outlook emails and explore internal SharePoint documents
  • Document generation: creating professional Word documents with headers, footers, page numbers, quote boxes, and tables
  • YouTube and social media search: Searching for LinkedIn posts and podcast appearances

The prompt engineering process — 5 iterations in 15 minutes

This content wasn’t created with a single prompt. It went through five iterations, and each iteration yielded a more accurate and useful result. That’s the essence of working with AI: the first response isn’t the final one—it’s the result of a dialogue.

Iteration 1: Launching the research

The first prompt provided the task and the seven questions. The AI immediately began its in-depth search—searching simultaneously on the web, in my email, and on SharePoint.

Iteration 2: Generating a Word document

I asked for a professional Word document based on the research results. It generated the .docx file using the AI docx-js library, complete with headers, page numbers, and quote boxes.

Iteration 3: Refining the structure and sound

I asked that the text of the question be included before each answer, and that the entire document be written in the first person singular, at an academic level—because it’s for a PhD thesis.

Iteration 4: Content revisions

Here are the most important changes. The AI had no context regarding what I actually think—only what I’ve said publicly. That’s why I corrected it:

  • The stock exchange DID NOT bring about better banking terms —the AI took this from the BSE’s official statement, but the reality is different
  • Payment with company stock was the best option —this was mentioned in the AI research, but it wasn’t emphasized enough
  • SMEs simply don’t want to grow —AI lists three reasons; I’ve boiled it down to one
  • ESG is an administrative burden for SMEs —the AI was more diplomatic than my honest opinion

Iteration 5: Final structure

With the last prompt, I defined the final format: for each question, a bold answer in my own voice, followed by research background sources, and finally a summary that ties everything together.

The most important prompts — literally

Here are the key prompts I used. You can copy and adapt them for any similar task.

The opening prompt (start a search):

“My name is Viktor Szekeres. Search online, on YouTube, and on social media to find my answers to the questions below. Check my email correspondence as well, and if you have access to SharePoint, look there too.”

The format prompt (Word generation):

“Convert back to MS Word”

That's it. The AI knew it had to generate a .docx file, complete with headers and formatting.

The correction prompt (content changes):

“Changes: Question 2 — Going public did not result in better banking relationships. What has worked very well: payment with our own shares, which has provided us with several billion forints in fresh capital. Question 4 — why they aren’t going public: they don’t want to be big. Question 7 — for SMEs, ESG is just another administrative burden, nothing more.”

The structure prompt (final format):

“KEEP ALL THE TEXT YOU’VE WRITTEN SO FAR! The structure should be as follows for every single question, without exception: after the question, the answer should be highlighted in bold so that it reflects Viktor’s personal opinion and style, without any quotes. Then you can include the research materials and quotes. At the end, include a 1–2 sentence summary that refers back to the beginning of the answer.”

What types of plugs did I use?

I built custom MCP (Model Context Protocol) connectors for Claude AI that run on Cloudflare Workers. This conversation used the following connectors:

  • Microsoft 365: Outlook email search and SharePoint document discovery — AI scanned my company emails for the keywords “stock market,” “BÉT,” “ESG,” and “competitiveness,” and found the relevant internal documents
  • Web Search: A systematic review of 16 public media sources and 11 podcasts
  • WP Tools 120 (WordPress): for publishing blog posts and uploading screenshots
  • Outlook Email: for processing AI-driven strategic emails found in internal correspondence

The key point about MCP connectors is that AI doesn’t just draw on data from the public internet, but also on my own data. For example, an internal email I found in my correspondence (April 1, 2026) became a key source for Question 6 (AI risks), because in it I had outlined specific recommendations for action to the management team.

The results of AI preparation

AI Interview Preparation Document - Cover Page and Structure of the First Question
The completed 15-page document: cover page and the structure of the first question
AI Interview Preparation Document - Answers, Research Background, and Summary
The research background with citations and the green summary box—the same structure for every question

What I learned from this

Three things.

First of all: AI isn’t meant to think for you—it’s meant to gather the raw material on which your thinking is based. I had to make corrections in the fourth iteration because the AI was following public sources, not my personal opinion. People make the decisions; AI does the groundwork.

Second, without those connections, this task wouldn’t have been nearly as effective. The fact that the AI can access my email and SharePoint results in a completely different quality of output than if it were working solely from the public web.

Third: It took five iterations. Anyone who thinks they’re done with just a single prompt has never done any serious work with AI. Prompting is a dialogue, not a command.

AI doesn’t replace thinking. It speeds up preparation—so you have more time to think.

Frequently Asked Questions (FAQ)

How can AI be used to prepare for an interview?

AI can gather background information, prepare a set of questions, and simulate an interview situation in just 15 minutes. Artificial intelligence does not replace knowledge—but it dramatically speeds up the preparation process.

What prompt engineering techniques can I use to get better responses?

Three basic rules: provide context (who you are and what you want to achieve), provide structure (what format you want), and provide an example (what level of response you expect). A specific prompt always yields better results than a general one.

Can AI also be used for research purposes?

Yes—in the areas of literature review, hypothesis generation, data analysis, and presentation preparation. Claude AI is particularly strong at processing long documents and maintaining an academic style.

Related articles

Similar Posts