The Impact of Artificial Intelligence on Organic Search: Results from an Applied Study in Search Marketing

The relationship between artificial intelligence, digital writing and organic positioning is no longer a distant prediction. It is now an operational issue for marketing teams, journalists, content managers, programmers, data specialists and brands that depend on visibility in search engines. In this context, José Santiago Encina Jara, a former student of Luis Cordeiro, Managing Partner & CEO of BYDAS and University Professor at Católica Porto Business School as an Industry Fellow in the Search Marketing course of the Master’s in Marketing, developed an applied research project on the impact of AI on writing and publishing on the internet.

After completing his master’s programme, José was invited by BYDAS to develop a research project in Search Marketing, under the cooperation protocol between BYDAS and Católica Porto Business School and under the supervision of academic advisor Prof. Ricardo Alexandre Morais. The challenge was clear: to understand whether articles written by humans would have greater organic relevance than articles created by AI and, consequently, to assess whether ranking algorithms would prioritise human content, AI-generated content or, most likely, quality signals independent of authorship.

The scientific problem: does AI replace the writer or expand their capabilities?

The central question of the study was not simply whether AI can write. That question already has a practical answer: it can produce text, structure articles, propose titles, generate descriptions, suggest keywords and adapt styles. The more demanding question is: does this content perform organically? And, if it does, under what conditions does it outperform or fall short of a text produced by a human writer with journalistic method, editorial judgement and cultural sensitivity?

The study starts from a tension that defines contemporary digital marketing. On the one hand, tools such as ChatGPT make it possible to accelerate tasks, synthesise information and create versions of content in seconds. On the other hand, content published online does not compete merely by existing. It competes through trust, search intent, clarity, authority, reading experience, click-through rate, retention, answerability and semantic coherence. This is where SEO, GEO and AEO intersect.

SEO remains the discipline that seeks to improve organic visibility in search engines. GEO, or Generative Engine Optimization, adds a new layer: preparing content to be understood, cited, summarised or retrieved by generative engines and AI-based assistants. AEO, or Answer Engine Optimization, focuses on direct, useful and verifiable answers, whether on a results page, in a featured snippet, in voice search or in a conversational interface. BYDAS’ research is relevant because it anticipates this convergence.

Odd Takes: an editorial laboratory created by BYDAS

To turn the question into an observable experiment, BYDAS created a fictional project called Odd Takes. The concept was simple, yet specific enough to allow analysis: a blog dedicated to strange, disruptive, unusual and curious products found on the internet. Instead of competing in saturated areas such as generic technology, fashion or lifestyle, the project chose a more clearly defined editorial niche, with potential for long-tail searches (long tail keywords).

José, with a background in Journalism, took on the role of human author. Every week, he produced curated articles in review format, applying journalistic technique, angle selection, narrative structure and critical product assessment. In parallel, BYDAS prepared an AI agent with specific instructions to carry out the same task: research topics, structure articles, create titles, generate descriptions, produce text in an optimised format and publish content regularly.

The result was a blog fed by two editorial sources: half of the articles written by a human and half by AI. This division made it possible to compare organic performance signals without turning the study into an abstract discussion. The focus shifted to data: impressions, clicks, CTR, average position and page behaviour on monitoring platforms.

Methodology: content, data and interviews

The research combined a qualitative and quantitative approach. On the qualitative side, interviews were conducted with members of the BYDAS team, including leadership, content creation and web development. These interviews made it possible to understand how digital marketing, programming and management professionals view AI in their daily work, which tools they use most, which risks they identify and which tasks they consider more or less replaceable.

On the quantitative side, the study tracked the performance of Odd Takes using tools such as Google Search Console and Google Analytics. The articles were published in English to maximise international reach and were distributed across categories such as technology, appliances, crafts and gifts. The experimental period ran between the final quarter of 2024 and January 2025, with registration on search engines from October onwards.

To reduce bias, topics were randomly assigned between the human author and the AI author. Paid promotion, when used, was directed to the homepage rather than to specific articles, which avoided directly favouring either profile. The goal was not to create an advertising campaign, but to observe how different content types competed in an organic context.

• Human author: José Santiago Encina Jara, with a journalistic background and his own editorial responsibility.
• AI author: a customised ChatGPT profile, prepared to write formal, structured and SEO-oriented articles.
• Indicators assessed: impressions, clicks, CTR and average position on results pages.
• Object of study: articles and reviews about unusual products published on Odd Takes.

Results: AI appears, the human persuades

The results are particularly interesting because they do not give an absolute victory to either side. In the queries analysed, the AI profile obtained more impressions and more clicks than the human profile. This suggests that AI-generated content managed to appear more frequently in certain searches and capture initial visibility signals.

However, when the analysis shifts to pages, the human profile shows greater ability to convert into clicks. In the combined calculation, both profiles achieved the same total number of clicks. The difference in impressions was minimal. In other words, the experiment does not demonstrate a clear superiority of AI in effective traffic.

The difference emerges in the combined CTR. The human profile recorded an average of 33.88%, while the AI profile stood at 27.49%. Since CTR measures the relationship between impressions and clicks, this indicator suggests that the titles, descriptions or editorial angles of the human author were more effective at turning visibility into action. AI managed to appear well; the human managed to persuade better.

Meanwhile, the average SERP position slightly favoured AI. The AI profile achieved a combined average position of 13.31, while the human profile recorded 15.45. Since, in this indicator, a lower value represents a better position, the result suggests that AI-generated content may have been more easily structured for algorithmic reading. Even so, the difference is limited and should be interpreted with caution, given the size of the sample.

What do these data say about ranking algorithms?

The most balanced scientific interpretation is this: the study does not prove that algorithms systematically favour AI or humans. What it shows is more subtle and more useful for digital marketing teams. Authorship, in itself, appears to matter less than the combination of structure, relevance, search intent, editorial quality and the ability to generate clicks.

An article created by AI may be well organised, contain keywords, answer objectively and present a predictable structure for search engines. This may support crawling, indexing and semantic framing. However, a human article may find a more appealing angle, choose a title with legitimate curiosity, avoid mechanical repetition and build a more memorable reading experience.

In terms of technical SEO, this distinction is essential. Search engines need to understand the page: title, meta description, HTML hierarchy, structured data, speed, indexability, internal links, intent and thematic consistency. But users need a reason to click and keep reading. SEO is not just about being found. It is about being chosen.

Technical SEO: where BYDAS adds method

BYDAS starts from a practical view: AI should perform analytical and repetitive tasks, but it should not replace strategic judgement. In a real project, content production does not begin with the text. It begins with information architecture, technical auditing, intent analysis, opportunity selection, entity definition, code quality, performance and the ability to measure results.

This is where expertise in technical SEO becomes decisive. An excellent article may fail if the page is not indexable, if loading is slow, if the heading structure is confusing, if canonicalisation is incorrect, if duplicate content disperses relevance or if internal linking does not help search engines understand the relative importance of the page.

At the same time, AI can support several technical tasks: suggesting semantic markup, reviewing meta information, identifying patterns across large volumes of URLs, grouping searches by intent, proposing frequently asked questions, detecting content inconsistencies and accelerating documentation. However, it needs supervision. A mistake produced quickly is still a mistake. In SEO, speed without validation can multiply problems.

GEO: optimising for generative engines

The concept of GEO, or Generative Engine Optimization, is gaining importance because search is no longer limited to a list of blue links. Generative interfaces summarise, compare, suggest and present composite answers from multiple sources. In this scenario, content needs to be understandable both to traditional engines and to language systems that look for entities, relationships, definitions, evidence and consistency.

The Odd Takes experience helps explain this point. An AI text may structure information well, but it risks sounding generic. A human text may be more original, but it needs enough structure to be interpreted by automated systems. The competitive advantage lies in the combination: content with a human voice, editorial rigour and semantic organisation prepared for algorithmic reading.

For GEO, repeating keywords is no longer enough. It is necessary to build topical authority. This means answering related questions, explaining concepts, using examples, organising subtopics, creating relationships between entities and presenting verifiable information. Content prepared for GEO should be easy to summarise without losing what matters. If a generative assistant has to choose between several sources, it will tend to value clarity, completeness, coherence and trust.

AEO: answering better, not merely appearing better

AEO, or Answer Engine Optimization, shifts the focus from the page to the answer. The goal is for content to be able to solve a question in a direct, contextual and useful way. This is decisive in informational searches, voice searches, assistant queries, answer boxes and rich results.

In the case of Odd Takes, many pieces of content were reviews. This favours questions such as: is this product worth buying? What is it for? What are the advantages? What limitations does it have? An article with good AEO anticipates these doubts and answers them without forcing the reader to decipher long blocks of text. Editorial structure becomes a user experience tool.

AI is useful for mapping frequently asked questions and intent variations. However, the final answer requires judgement. A good answer is not merely grammatically correct. It must be honest, proportionate, useful and adapted to the reader’s level of knowledge. This is where the role of the writer remains strategic: interpreting the real intent behind the search.

The role of the writer in the future of digital publishing

The study suggests that the writer does not disappear; their role changes. They cease to be only someone who writes from scratch and begin to act as editor, researcher, strategist, verifier, information architect and curator of perspectives. AI can accelerate production, but the human writer remains responsible for intent, ethics, differentiation and credibility.

In José’s case, his background in Journalism was relevant because journalistic writing is not limited to formal accuracy. It involves selecting information, establishing hierarchy, applying critical thinking, connecting with the reader and turning data into narrative. These elements help explain why the human profile achieved a better combined CTR: there was an editorial attraction component that AI, even when well configured, did not fully replicate.

This does not mean that AI failed. On the contrary, it showed the ability to compete. The strongest conclusion is precisely this: AI is already competent enough to enter the race, but it still depends on human direction to produce distinctive value. The question is no longer “AI or human?” but “which part of the process should belong to each one?”.

Implications for brands and marketing teams

For brands that publish regularly, the study offers a practical lesson. Producing more content does not guarantee better results. Producing with method does. AI can help scale research, structure and variations, but the editorial strategy must define priorities: which topics deserve investment, which searches have commercial value, which pages should be updated and which content requires human authority.

1. Use AI to accelerate research: collect ideas, group intents and identify recurring questions.
2. Maintain editorial supervision: review facts, remove generalities and strengthen original arguments.
3. Apply technical SEO from the outset: ensure clean HTML, appropriate metadata, correct indexation and performance.
4. Write for people and systems: combine human clarity with semantic structure for GEO and AEO.
5. Measure continuously: compare impressions, clicks, CTR, position and traffic quality before drawing conclusions.

This approach avoids two common mistakes: rejecting AI out of fear or accepting it without control. Maturity lies in the middle. AI should be a tool serving a strategy guided by specialists, with clear metrics, human validation and an integrated vision of SEO, content and technology.

Study limitations and scientific value

Like any applied research, the study has limitations. The traffic sample was small, the experimental period was relatively short and the editorial niche chosen had specific characteristics. In addition, the performance of a human author depends on their skills, just as the performance of an AI agent depends on the quality of the prompt, configuration, instructions and subsequent review.

These limitations do not reduce the project’s value. On the contrary, they make the analysis more honest. The study does not attempt to present a universal truth; it offers a controlled, documented and useful experiment to formulate new questions. Future studies may extend the analysis period, test other sectors, compare different AI models, measure reading time, scroll depth, backlinks, trust signals and impact on commercial conversions.

The main scientific contribution lies in bringing academia and the market closer together. Instead of discussing AI and SEO only in theory, BYDAS created a real publishing environment, with real data, real content and real limitations. This connection between research and practice is essential in a field that changes quickly and where many decisions are still made based on perceptions, fears or excessive enthusiasm.

Conclusion: the future is hybrid, technical and editorial

The Odd Takes study shows that AI already deeply influences digital writing and organic optimisation. It can generate structure, volume and useful technical signals. It can reach competitive positions and support tasks that used to consume time. But human performance remains relevant, especially when the goal is to generate clicks, trust and differentiation.

The answer to the initial challenge is therefore balanced: AI should not be seen only as a threat, nor as an automatic solution. It should be integrated into a workflow that combines technical SEO, GEO, AEO, data analysis and editorial competence. The future of digital publishing will be hybrid: machines accelerating processes, humans defining meaning.

At BYDAS, this research reinforces a practical conviction: winning content is content that combines technology, strategy and creativity. AI can help us get there faster, but it is method, validation and human vision that transform pages into results.

BYDAS applies this knowledge in technical SEO, GEO and AEO projects, combining analysis, web development, content and performance for brands that want to grow in organic search with method, rigour and a future-oriented vision.

URL for the study: https://repositorio.ucp.pt/entities/publication/efbbf235-079e-4527-a212-56e0c4a620c9

Supervisor: Ricardo Alexandre Morais

Credits: José Santiago Encina Jara