Which Fintech Brands Win in AI Search? We Analyzed 68,000 Answers to Find Out
Brand AI visibility is not one market-wide rank: teams need category, engine, and source-mix views before deciding where to invest.
68,334Industry
Last added 28 Sep 2026newest paper 22 Sep 2026 · curated, each linked to the original
Research
· AI search research, with a 30-second read of each paperpapers on how AI finds, ranks and cites the web. 75 came out in 2026.
50 of the 79 papers came out since June
Papers in the library by month published. Each block is one paper.
Preprint 45Peer-reviewed 19Industry 14Trakkr 1
| Month | Papers |
|---|---|
| Before 2026 | 4 |
| Jan | 4 |
| Feb | 3 |
| Mar | 6 |
| Apr | 7 |
| May | 5 |
| Jun | 10 |
| Jul | 16 |
| Aug | 10 |
| Sep | 14 |
Newest papers
Click a paper for its 30-second read. Key stat on the right.
Newest first · 79 papers
GroundedGEO: Auditing the Evidence Gap in Generative Search Rankings
Product and service pages need verifiable support for ranking-driving claims, because answer engines can reward detailed claims even when evidence is missing.
0.65 to 0.43Preprint
Auditing Source Exposure in Baidu and Google AI Search
International AI visibility needs engine- and language-specific source audits, because similar answer text can expose very different source domains.
0.008-0.034Peer-reviewed
Characterizing Web Search by Conversational LLM Agents: From Search Decisions and Strategies to Results and Responses
Source visibility depends on the whole agentic search stack, because assistants can differ on whether they search, how they query, which domains they see, and which sources they cite.
14% vs 83%Preprint
"If I Had to Buy Just ONE: Galaxy S26 Ultra": Auditing AI-Generated Product Recommendations
Commerce audits should use repeated consumer-interface checks, because APIs and single answers can miss the advice framing and source set shoppers actually see.
5.4%Preprint
Understanding AI Provider Recommendations in Local Service Markets
Local and professional-service visibility depends on whether the assistant can retrieve current web evidence, because fluent no-search referrals can be fabricated or skewed.
64-71%Preprint
Whom Do AI Agents Work For? Role Assignment Induces Sponsorship Bias in LLM Recommenders
Paid placement in AI recommendations needs direct measurement and governance, because a disclosure label seen by an agent may not protect the shopper who receives the final answer.
50.2 ppPreprint
Evaluating Brand Retrieval and Ranking in Large Language Model Recommendations
Brand audits need an external competitive set and repeated prompts, because an AI shortlist can hide relevant brands unless the user wording activates the right associations.
2,400Preprint
Agentic Share-of-Search: A Multi-Agent AI System for Competitive Decision-Making in LLM-Mediated E-Commerce
Commerce teams need engine-level share-of-search checks, because AI shopping assistants can redirect discovery before a shopper reaches a product page.
39%Peer-reviewed
Purchase Advice and Observable Buyer Responses in Real AI Conversations
AI visibility measurement should not treat a recommendation as a conversion, because the answer is often observable while the buyer decision is not.
26.9%Preprint
Following the Preference, Missing the Optimum: Compliance Without Optimization in AI Housing Recommendation
AI recommendation quality should be judged against the full available inventory, because an answer can obey stated preferences while still missing a plainly better option.
39.0%Preprint
Repeated Queries Exhaust an LLM's Brand Recommendations but Not Its Sources
Brand visibility audits should not treat one prompt run or one fixed roster as complete, because repeated answers can reveal brands and source domains that a single pass misses.
86-92%Preprint
Inventory-Grounded Policy-Level Optimization for Training-Free AI Search
For commerce and marketplace brands, AI search needs live product facts and availability signals, because static prompts or stale catalog knowledge can select the wrong item.
-38.9%Peer-reviewed
The Dice Roll Method: A Standardized Protocol for Repeated-Query Auditing of Large Language Model Brand Recommendations
AI visibility measurement should publish its sampling depth and uncertainty, because a brand win or loss from too few runs can be a sampling artifact.
15 runsPreprint
Sources of Truth: A Multi-Platform, Multilingual Audit of Citations in AI Mental Health Information Queries
Consumer AI search can concentrate trust in a small set of sources, so health, wellness, and regulated brands need citation monitoring by platform and language.
43.6%Preprint
CHASE: How Content Ecosystems Are Reshaped When Ranking Is the Only Target
GEO work should protect useful evidence and editorial quality, because chasing a narrow ranking signal can reshape a category into content that is more optimized but less useful.
Peer-reviewed
ChatGPT Referral Traffic Near Triples Overnight
AI traffic is not only a volume story: when assistants expose brand links, discovery can shift toward homepage-level visits that look more like brand demand than classic keyword traffic.
+354.7%Industry
AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence
Publishers and brands should measure answer exposure separately from referral traffic, because a more conversational search interface can keep attention inside the search product.
-18.8 ppPreprint
GEO-Flag: Detecting and Measuring GEO-Optimized Web Content
Content teams should pair GEO work with source-quality checks, because optimized pages can win AI visibility while making citations harder for users to verify.
8.90%Preprint
Whose doctor does the AI recommend? An algorithm audit of reputation and demographic signals in large language model-assisted physician choice
For local and professional-service brands, AI recommendations can create invisible visibility shifts that only repeated behavioral audits will catch.
+31.4 ppPreprint
AI and organic search are doing different jobs: What Shopify's data shows
Commerce brands should measure AI referrals beside organic search and make product data machine-readable, because AI traffic is smaller but often arrives with clearer buying intent.
+197%Industry
Invisible to the Machine: Auditing AI Restaurant, Cafe, and Bar Recommendation Against a Complete Market Census
Local AI visibility depends on whether a venue enters the answer set at all, and documentation signals like websites, price information, and third-party mentions may matter before ratings can help rank it.
85.6%Preprint
SAGEO Arena: A Realistic Environment for Evaluating Search-Augmented Generative Engine Optimization
Content teams should not treat GEO as one rewrite step, because a change that looks good to a generator can still hurt whether the page is retrieved or ranked for the answer.
Peer-reviewed
Investigating Click Behaviors On Google Search Result Pages That Produce an AI Overview
Publishers and brands should measure answer exposure separately from referral traffic, because being cited in an AI Overview may not translate into a site visit.
~1%Preprint
Language Models Agree With Each Other, Not With Readers
Content teams should not treat LLM salience as human salience, because a passage can look important to models while missing what real readers mark.
2,523Preprint
Evaluating and Pricing Advertisements in AI-Generated Responses
If paid placements move into AI answers, brands will need to measure user utility, placement quality, and likely engagement, not only whether an answer names them.
86%Preprint
Learning Dynamics of Strategic Publishers in Generative AI Ecosystems
Publisher AI-search strategy should expect platform incentives to shape source diversity and content drift, not just reward the best page.
Preprint
The Prompt Is Not the Query: How Request State Evolves Across Multi-Turn AI Conversations
AI search measurement should treat the session as the unit, because purchase constraints, evidence requests, and brand alternatives can appear several turns before the final answer.
50.3%Preprint
SIREN (Luring LLMs onto the Rocks): PAIR-Driven Preference Manipulation in Web-RAG Recommenders
AI recommendation visibility can be shaped by source-page claims after retrieval, so brands need to monitor ranked answers and the sources that may steer them.
62/124Preprint
What Gets Cited: Competitive GEO in AI Answer Engines
Content teams should treat AI citation work as evidence and source-selection work, not a cosmetic rewrite of headings or keywords.
252KPeer-reviewed
Answer First, Evidence Second? Uncovering Hidden Risks in Well-Structured AI Search Summaries
Visible citations can make an AI answer look trustworthy, so brands and publishers need to check whether cited pages actually support the answer.
32.31%Peer-reviewed
A Comparative Study of Users' Information-Seeking Practices Across Search Engines and Generative AI Chatbots
Brands should expect some customers to stop at the AI answer unless the system makes sources visible and easy to verify.
Peer-reviewed
The AI Traffic Report Q2 2026: Agentic Traffic Surged 45%, With Meta Taking the Lead
Publishers and brands should separate crawl volume from business value, because the bots consuming infrastructure are not always the engines sending visitors.
17.7BIndustry
Answering Without Referring: How AI Search Rewrites the Web's Economic Bargain
Publishers and brands should measure visibility inside AI answers as well as referral traffic, because discovery can now happen without a visit to the source website.
5.2%Preprint
Large language models create an uneven informational layer over cities
Local businesses can remain invisible even when AI systems know they exist, so visibility measurement must test recommendations across models, locations, and customer profiles.
47.5%Preprint
Curated retrieval versus open web search in public AI information services: a coverage-trust trade-off
Publishers and brands cannot assume authority guarantees selection, and answer fluency alone does not show whether an AI system chose trustworthy sources.
35%Preprint
Think Before Writing: Feature-Level Multi-Objective Optimization for Generative Citation Visibility
Brands should improve the page-level evidence, structure, and clarity that answer engines can cite, instead of relying on keyword-style rewrites.
Peer-reviewed
Making AI search smarter
For publishers and brands, AI visibility may increasingly depend on machine-readable freshness and access signals, not just whether a crawler can fetch a page.
50%+Industry
Top AI Chatbots and Assistants by Market Share July 2026
AI visibility planning should not treat ChatGPT as the whole market, because Gemini and Claude now carry enough consumer attention to warrant separate measurement.
10.40BIndustry
An Eye Tracking Study: Are AI Overviews Changing Search Behavior?
When an AI answer occupies the top of the page, brand visibility inside that answer can matter before a user ever evaluates blue links.
Peer-reviewed
AI Visibility Index 2026 Study
Brand visibility now needs competitive benchmarking by model and category, because one engine’s answer market is not a proxy for another.
126MIndustry
How Large Language Models Source Brand Reputation Across Languages and Markets
For brand visibility, earned media and local reference sources may matter more than owned pages when AI systems form reputation answers.
85.7%Preprint
What's making 4 in 5 Americans doubt AI?
Brands need source-visible content and attribution checks, because trust in an AI answer depends on whether users can see where the answer came from.
86%Industry
Generative Engine Optimization at Scale: Measuring Brand Visibility Across AI Search Engines
AI visibility is measurable, but smaller brands should not expect one prompt run or one citation count to reflect their real answer-market position.
11%Preprint
Why Calling Yourself the 'Best' Could Be Helping Your Competitors Win in AI Search
Brands should separate citation tracking from recommendation tracking, because a cited page can still help competitors win the answer.
69%Industry
Americans and AI 2026: Chatbots, Smart Devices and Views on Impact
AI answers are now a mainstream discovery surface, so brand measurement needs to include exposure inside summaries and chatbots, not only clicks.
60%Industry
7 LLMs.txt myths we should clear up
Brands should treat LLMs.txt as an agent-readable source map that may help discovery, not as a replacement for crawlable pages and source-worthy content.
Industry
Quantifying Uncertainty in AI Visibility: A Statistical Framework for Generative Search Measurement
AI visibility reports should show uncertainty around citation share and rank, because a single prompt run can make brand/source performance look more precise than it is.
Preprint
Explicit Evidence Grounding via Structured Inline Citation Generation
AI visibility should measure whether a cited page actually supports the answer, because URL-level citation alone can hide weak grounding.
Preprint
Rethinking Search as Code Generation
Brand visibility is moving beyond blue-link rankings: AI assistants may find sources through many hidden sub-searches before a user sees the final answer.
Industry
The Impact of AI Search on the Online Content Ecosystem: Evidence from Google and Reddit
AI search can push attention toward cited communities, but interface changes can quickly rewrite the benefit.
+12.0%Preprint
LLMs applied to web scraping and web crawling: a systematic review
AI crawling is becoming semantic and agentic, which changes what crawlable content means for AI visibility.
91Peer-reviewed
Measuring Google AI Overviews: Activation, Source Quality, Claim Fidelity, and Publisher Impact
Google AI visibility needs citation tracking and claim checks, not just rank tracking, because cited pages and supported claims are separate outcomes.
11.0%Preprint
Benchmarking Retrieval Strategies for Biomedical Retrieval-Augmented Generation: A Controlled Empirical Study
Query expansion can add noise; retrieval strategy should be tested before assuming more searches improve AI answers.
0.852Preprint
FT-RAG: A Fine-grained Retrieval-Augmented Generation Framework for Complex Table Reasoning
For brands with comparison tables, specs, pricing, or datasets, AI retrieval needs structured facts at the cell and entity level.
+59.2%Preprint
How Generative AI Disrupts Search: An Empirical Study of Google Search, Gemini, and AI Overviews
Brands cannot assume Google rankings, Gemini answers, and AI Overview citations are the same visibility surface.
51.5%Peer-reviewed
From Citation Selection to Citation Absorption: A Measurement Framework for Generative Engine Optimization Across AI Search Platforms
Citation count alone can mislead; a cited page only matters if its evidence actually shapes the answer.
21,143Preprint
Faithfulness-QA: A Counterfactual Entity Substitution Dataset for Training Context-Faithful RAG Models
AI engines can cite a page but still answer from memory; visibility work needs context-faithfulness tests.
99,094Preprint
The Half-Life of AI Citations
AI visibility decays fast. Getting cited once isn’t the finish line; staying cited is the real work - which is exactly what continuous tracking is for.
6.8 daysTrakkr
From Experience to Skill: Multi-Agent Generative Engine Optimization via Reusable Strategy Learning
This is useful for Trakkr because it treats AI visibility and citation accuracy as a joint measurement problem.
Peer-reviewed
Self-Correcting RAG: Enhancing Faithfulness via MMKP Context Selection and NLI-Guided MCTS
Citation quality depends on both what evidence enters the context and whether the answer is checked against it.
Preprint
Google AI Overviews and Publisher Traffic: Evidence from a Field Experiment
This is one of the most useful study shapes for Trakkr because AI visibility only matters commercially when it changes downstream attention.
Preprint
Signaling the Shift to Generative Engine Optimization (GEO)
Large brands are experimenting with GEO signals, but most still have shallow AI-specific technical coverage.
7.4%Industry
Hypothesis-Conditioned Query Rewriting for Decision-Useful Retrieval
AI engines often answer from hidden rewritten searches, so the phrasing of your content needs to satisfy multiple evidence paths.
+5.9 ptsPreprint
Structured Linked Data as a Memory Layer for Agent-Orchestrated Retrieval
Structured data may matter most when it helps agents navigate entities, not when it is treated as a standalone ranking trick.
+29.6%Preprint
Diagnosing and Repairing Citation Failures in Generative Engine Optimization
The practical takeaway is sharp: diagnose the citation failure before rewriting the page.
+40%Preprint
Reason and Verify: A Framework for Faithful Retrieval-Augmented Generation
The useful pattern is not biomedical-specific: grounded sub-claims make citation failures easier to diagnose.
89.1%Peer-reviewed
AgenticGEO: A Self-Evolving Agentic System for Generative Engine Optimization
GEO is moving from fixed page-writing recipes toward engine-aware, page-specific optimization.
Preprint
State of AI Agent Policies - 2026
Most of the web has not made a clear machine-readable choice about AI crawling, which makes crawler monitoring a real governance issue.
90.1%Industry
The Rise of AI Search: Implications for Information Markets and Human Judgement at Scale
AI search may concentrate attention, so brands in the long tail need visibility measurement before they assume search demand still reaches them.
243Preprint
Generative Engine Optimization: A VLM and Agent Framework for Pinterest Acquisition Growth
Visual and marketplace brands should think in AI-retrievable collections, not only individual asset captions.
+20%Preprint
Evaluating the Utility of Grounding Documents with Reference-Free LLM-based Metrics
Not every relevant page is useful to every model; content utility should be measured model by model.
+18.2 ptsPreprint
PruneRAG: Confidence-Guided Query Decomposition Trees for Efficient Retrieval-Augmented Generation
For AI visibility, being retrieved is only the first gate; the model still has to use the evidence.
Preprint
Over-Searching in Search-Augmented Large Language Models
More retrieval is not always better; AI visibility systems should measure retrieval quality, negative evidence, and abstention behavior.
Peer-reviewed
When Attention Becomes Exposure in Generative Search
AI answers may reward existing attention, so brand visibility work needs to track third-party voices, not just owned pages.
Preprint
GEO: Generative Engine Optimization
This is the founding paper of the field Trakkr is built around. It shows the things you put on a page measurably change whether AI quotes you - visibility is optimisable, not luck.
up to 40%Peer-reviewed
Lost in the Middle: How Language Models Use Long Contexts
Being retrieved isn’t enough - position inside what the model reads matters. Being the clear, early, self-contained answer beats being one mention buried among many.
U-shapedPeer-reviewed
Evaluating Verifiability in Generative Search Engines
AI cites loosely. If you’re the source, being the unambiguous support for a clear, checkable claim is how you earn a citation that holds up - not just a passing mention.
51.5%Peer-reviewed
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
This is the architecture behind ChatGPT search, Perplexity and Google’s AI answers. It’s why “what’s retrievable about you” quietly decides what AI is able to say about you at all.
Peer-reviewed
Almost every paper measures something
Papers per topic. A paper can carry up to three.
75 papers point to data you can check here
Papers per Trakkr dataset that tests the same question.
4 papers have no matching dataset yet.
How we choose
We read peer-reviewed papers, preprints and primary research from AI labs, universities and publishers on how AI search finds, reads, ranks and cites the web. Each paper gets a 30-second read in our words (what they did, what they found, why it matters) and a link to the original.
Key stats are quoted from the paper. Where a Trakkr dataset tests the same question, the paper links to it. Industry studies carry an Industry label.
What is the Trakkr Data research library?
A curated, plain-language index of the research on how AI search discovers, reads, ranks and cites the web. Each paper gets a short read in our words and, where a Trakkr dataset tests the same question, a link to it so the finding sits next to the data.
Where do the papers come from?
Peer-reviewed papers, preprints (mostly arXiv), industry studies and our own research. Every entry links out to the original source, so the library is a map into the literature, not a replacement for it.