INTERNATIONAL RESEARCH / ACADEMIC PUBLICATION
Does greater visibility in AI bring more customers?
A "40% increase" in a study does not mean that sales rose by 40%. We explain which results to examine when deciding whether investment in your website is paying off.
GEO: Generative Engine Optimization
ResearchersPranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande
KDD 2024 · conference paper · Source language: English
Open the original publicationWe explain the paper in plain English, discuss what it may mean for organisations and identify our own examples separately.
English review: 9 September 2026 · The source date appears in the citation above.
The source has been checked; the review still carries its original editorial status. DijitalPi did not conduct a customer experiment for this paper.
LET'S READ THE RESEARCH TOGETHER
We first explain the researchers' question, method and findings. We then discuss how to interpret the results, clearly separating DijitalPi's commentary from the source.
01 / WHAT DID THE RESEARCHERS WANT TO UNDERSTAND?
How can a web page appear in an AI-generated answer?
Aggarwal and colleagues investigated whether changes in web content affected source visibility in generated answers.
Explaining the context · DijitalPi commentary
As a result of a search, the user chooses between links. Artificial intelligence, on the other hand, can draw on several sources and prepare a single explanation. Even if your page is among the sources, the part you describe may not be included in the final answer. In order to understand the question of the research, it is necessary to consider accessing a document and benefiting from that document separately.
Consider a simple example: you give an editor five files on the same subject. You cannot expect each file to receive equal space in the final article. The question is whether changing how one file is written changes how much of it appears in the final text.
02 / HOW WAS THE RESEARCH CONDUCTED?
How did they test the question?
GEO-bench contains 10,000 queries. The main experiment used 1,000 test queries, the top five sources from Google, and GPT-3.5-turbo. Nine content editing methods were compared to the initial text.
Understanding the method · DijitalPi commentary
In such comparisons, the original text provides the baseline before any change. The edited text's score alone is not enough; we also need the comparison point. This tells us whether an apparently strong result is actually different from the original.
Visibility is a measurement definition here. It is checked where and how much content from the source is included in the answer. Without knowing this definition, it is easy to confuse “more visible” with completely different results such as sales, site visits or brand preference.
03 / RESEARCH FINDINGS
Adding quotes increased the visibility score.
In Table 1, the visibility score weighted by location is 19.3 in the beginning and 27.2 in the citation insertion method. 17.7 for keyword stuffing. Therefore, not every content arrangement produces the same results.
- Start text
- 19.3Average visibility score before editing.
- Quoted text
- 27.2Comparison score on the same criterion; not the percentage of customers.
Source: Method and findings of the original study, part 3 and table 1 ↗
04 / DIJITALPI'S EXPLANATION
How should we interpret these findings?
Let's calculate the difference in the table together: Subtracting 19.3 from 27.2 leaves 7.9 points. When we divide this difference by the initial 19.3, we find a relative increase of approximately 41%. In other words, “increased by 7.9 points” and “increased by approximately 41%” express the same comparison differently. Even though the second expression appears larger, what is measured does not change.
Several steps come before a customer receives a service: seeing the answer, noticing the brand, opening the link, finding a suitable offer and getting in touch. This chain is our business interpretation. The visibility score alone does not show how many of these steps occurred.
So in a content study report, look for the name of the measure first. If it says “increase,” ask what is increasing, what starting value it is calculated from, and on what date it is measured. The same habit also prevents accidentally juxtaposing percentages from different studies.
05 / CONCLUSION AND OPEN QUESTIONS
What did we learn, and what do we still not know?
The study shows that the presentation of source text can change its visibility in the experiment. It does not measure sales or search-ranking impact, and the results are limited to the system and queries used.
Does the same effect appear today in answers shown to someone searching for a service in Turkish? Finding out requires our own queries and a separate measurement under current conditions. The proposed experiment below follows from this open question; it is not presented as completed work.
Return to the researchers' original publication
Method and findings of the original study, part 3 and table 1 Limitations of the study, chapter 9Source and ownership
KDD 2024 · conference paper. Open the original publication
The academic work belongs to the researchers named above. This page contains DijitalPi's explanatory review and original business examples; it is not a full translation of the paper.
← Return to all research articles
Cookie preferences