Your page answers the customer’s question, but your brand is absent from the AI answer. Is the gap in the content, source selection or final response? A study of Aiso data finds a strong association between an own-domain citation and a brand mention, while also observing mentions without such citations. DijitalPi commentary: Examine question fit, source visibility and brand mentions separately instead of relying on a single content score. This observational preprint uses commercial platform data; it does not establish hidden engine mechanisms or sales effects. The author is Aiso’s founder and CEO and explicitly declares a financial conflict of interest.
ACADEMIC WORK REVIEWED
From Prompt to Recommendation: A Fitted Stage Model of Brand Visibility in AI Search
ResearchersBenjamin Tannenbaum.
arXiv · 19 September 2026 · v1 · Preprint · Commercial Aiso data · Source language: English
English review: 27 September 2026 · The source date appears in the citation above.
Editorial review draft
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?
Why separate page fit, source citations and brand mentions?
Benjamin Tannenbaum develops a measurement model that separates stages of AI brand visibility instead of reducing it to a page score. It examines question fit, visible source selection, final brand inclusion and the tendency to appear without visible supporting sources.
Explaining the context · DijitalPi commentary
DijitalPi commentary: A page can answer a customer question well and still miss the source list. A cited page need not lead to a brand mention, and a brand can appear without its own site being cited. Combining these events into one success score obscures which problem the content team should investigate.
02 / HOW WAS THE RESEARCH CONDUCTED?
How did they test the question?
The study combines Aiso datasets collected for different purposes with a large monitoring panel from June through September 2026. The large panel contains 34,960 GPT/Gemini prompt-engine-run observations from 75 anonymized projects. Records with the target brand in the user’s prompt are excluded from this panel analysis. These are not customer or purchase counts.
Understanding the method · DijitalPi commentary
A stored source URL belonging to the brand’s own domain serves as an indicator of visible evidence exposure. Fan-out means the additional search queries an engine derives from the user’s question; the presence of the target brand in those queries is recorded separately. The outcome is literal inclusion of the brand in the answer, which does not necessarily mean a positive recommendation.
The fitted model combines earlier mention history with sources and additional queries observed during the current response. Those sources are contemporaneous, so the model does not know everything before generation begins. Repeated answers to the same prompt and evaluation on later records help examine associations; they do not causally reveal proprietary engine mechanisms.
03 / RESEARCH FINDINGS
Source visibility and brand mentions are associated, but distinct.
Within large-panel records without branded fan-out, GPT mentions the brand in 438 of 15,524 observations with neither own-domain exposure nor branded fan-out (2.8%). With an own-domain source, it mentions the brand in 866 of 1,769 observations (49.0%). The corresponding Gemini counts are 524 of 13,801 (3.8%) and 1,995 of 3,415 (58.4%). Each denominator is the set of answer observations for that engine and stated evidence condition. These are different groups, not before-and-after measurements; the differences do not establish a causal effect of adding a citation.
DijitalPi commentary: Test the claim that a page is ready for AI against specific questions. Assess question fit, visible sources, brand mentions and positive recommendations separately. Keep subsequent website visits, qualified enquiries and sales in independent records. The paper’s visibility rates do not measure those business outcomes.
A brand mention without an own-domain citation does not prove the information came from training data. Third-party information or invisible retrieval could also explain it. The association with page fit also varies by engine; there is no basis here for a fixed success coefficient across all engines.
05 / CONCLUSION AND OPEN QUESTIONS
What did we learn, and what do we still not know?
This single-author preprint is observational and diagnostic. Benjamin Tannenbaum is the founder and CEO of Aiso Boost Ltd., which develops commercial AI-search measurement software; the paper explicitly declares a financial conflict of interest. The data come from Aiso’s proprietary research and monitoring systems. Without raw conversations and client records, the released aggregates do not support independent model refitting.
Our open question at DijitalPi: for a consistent set of Turkish customer questions, which observable stage accompanies your brand’s absence? The framework generates diagnostic questions; adding a source link does not guarantee a recommendation or sale.
We created these scenarios to make the topic concrete. They are not cases from the paper or measured client results.
EXAMPLE 01
Ecommerce: a generic page for a specific need
An illustrative retailer wants a generic category page to answer questions about products for confined spaces.
Does the page address the customer’s actual constraints?
Open the recommendation for example 01+
Review content against a specific customer question.
Check dimensions, use cases and limitations against verified product information. Record page fit and engine citation as separate outcomes. Repeat observations under comparable conditions before attributing visibility changes to an edit.
EXAMPLE 02
Local services: a citation without a recommendation
An illustrative maintenance firm treats its presence in a source list as an endorsement.
Does the answer recommend the service or simply use information?
Open the recommendation for example 02+
Read the citation and brand sentence separately.
Record service area, source URL and the wording around the brand. Have a person distinguish recommendations, comparisons and incidental mentions. A source link does not establish that anyone called or booked.
EXAMPLE 03
B2B: the same question across different engines
An illustrative software company presents visibility on one engine as its performance across AI search.
Does the same stage explain absence on every engine?
Open the recommendation for example 03+
Keep separate observations for each engine.
Record the customer question and necessary conversation context. Track engine, model version, date, sources and brand mentions together. Do not treat a single absence as permanent exclusion or repeated outputs as independent customers.
TRY IT WITH YOUR TEAM
Add stage distinctions to your visibility report.
Create a record for customer question, relevant page, visible source, brand sentence and repeat conditions. Measure later visits and enquiries separately. This DijitalPi proposal does not claim that we implemented the fitted model or achieved a visibility gain.
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.
Cookie preferences