Project journal

I brought my mentor 10 marketing tricks. The tool is now a re-ranker: take public ratings, account for tricks, give shoppers a transparent ranking. Fakespot is the case study. Yang et al. (2027) survey nearly 2 decades of detection research. A classmate suggested Ott et al.'s (2011) labeled corpus. Classmates asked how I confirm fakes and measure false positives.

What changed: the tool became a re-ranker for marketing tricks, and transparency replaced deception.

Roughly 30% of reviews are fake, costing $770.7 billion in 2025. Luca and Zervas: Yelp filtered about 16% of restaurant reviews. He et al. found fake Amazon reviews sold on Facebook. The FTC's Consumer Review Rule allows $53,088 per violation. 10 companies got warning letters. My project: a consumer tool flagging products ranking on tricks, not quality. Likely 1 category.

What changed: the idea has numbers, plus a warning: Fakespot did something similar and shut down in 2025.

A professor and I discussed an Amazon browser extension counting fake-looking reviews, telling human from AI writing, and flagging marketing tricks. My mentor said to limit scope to 1 category and pointed to cold-start and content-based recommendation. Surveys need IRB training. Working definition: a fake review is fabricated by the company itself. Homework: 10 tricks so far, with detection methods.

What changed: the tool became a consumer-facing browser extension, and the scope shrank to 1 product category.

My focus is clearer: how 1 variable influences LLM recommendations, LLM citations, and search rankings. Candidates are reviews, backlinks, or social media footprint. Very likely reviews, testimonials, and case studies. Audience: online shoppers and SEO or GEO marketers. Zhang et al. (2023) found fake reviewers measurably hurt platform trust, which favors reviews. 2 professors helped me narrow it down.

What changed: a field became 1 variable. Classmates pushed for 1 input, 1 outcome, 1 system, 1 LLM.

People ask AI instead of visiting websites. Cited sites get seen. AI-answer ranking factors: recency, reputable backlinks, reviews on Google, Trustpilot, G2, or Capterra, and social media presence. 2 ideas: a legitimacy checker for websites AI cites, and a consumer tool summarizing product reviews with AI. Homework: more ideas, plus a look at Perplexity, NotebookLM, and OpenAI Deep Research.

What changed: the problem moved from marketing in general to how AI answers pick which websites to cite.

My topic is marketing and technology. Marketing is now about human behavior. I have about 5 years in corporate marketing, years running my own businesses, and am Director of Marketing at a tech company. Affected: anyone buying goods or services. My only source is Chen et al. (2025) on generative engine optimization: marketers influencing purchasing decisions at their core.

What changed: a field and 1 source, no problem yet, but GEO gave me a thread to pull.

Kickoff with my mentor for the applied project. I came in wide: marketing and technology. We discussed marketing as a science: understanding human behavior, then putting content, links, and offers in front of people, matched to their needs. AI moves fast and belongs in the picture. Homework: a couple of ideas connecting marketing and data science, to brainstorm together.

What changed: a whole field became a homework task: a couple of ideas due next week.

The first 2 semesters of the doctorate were coursework: methods, systems, and a lot of reading that had nothing to do with product rankings yet. No research question, no mentor meetings about it, no journal. The thinking got serious in the fall, and that is where the entries above begin.

What changed: the ground got laid. The research direction did not exist yet.