KHB Research
The independent research arm of KHB — we publish quarterly AI influence benchmarks, cross-model consistency studies, and citation economy whitepapers that make AI's impact on brands measurable, comparable, and improvable.
Making AI influence measurable, comparable, improvable
A research mandate built on independence, longitudinal data, and open methodology — not on selling software.
AI influence has become a material business variable: it shapes which brands get recommended, which sources get cited, and which narratives get amplified. Yet until 2025 there was no shared, reproducible way to measure it. KHB Research was founded to close that gap. Our mission is to build the longitudinal datasets, transparent metrics, and peer-reviewable methodology that let enterprises, media, and regulators treat AI influence with the same rigor as search ranking or brand sentiment.
To protect objectivity, the research division operates with a structural separation from KHB's commercial advisory and platform teams. Researchers do not earn commission tied to client contracts, published rankings cannot be edited by the sales organization, and every dataset is reviewed by an internal ethics committee before release. This separation is what allows our benchmarks to be cited by third parties without conflicts of interest — and it is the reason our numbers are trusted even by brands that are not our clients.
Our method is longitudinal rather than snapshot-driven. We continuously track more than 1,000 brands across 8+ mainstream AI models, refreshing over 1,200 brand-specific queries every month. The result is a quarterly flagship benchmark, monthly vertical insight briefs, and on-demand anomaly alerts — a publishing cadence designed to give decision-makers signal, not noise, and to make every score reproducible by anyone who downloads our methodology notes.
Six programs tracking how AI reshapes brand influence
Each program runs on the same longitudinal panel, so findings can be cross-referenced rather than siloed.
AI Visibility Benchmarks
Cross-industry panels that measure how often a brand is named in AI responses to high-intent buyer queries. We compute a Visibility Rate per brand, per model, per vertical, then publish quarterly league tables so enterprises can compare themselves to direct competitors and to the vertical mean. The benchmark methodology is open and any analyst can re-run it on our published query library.
Citation Source Analysis
Which publishers, domains, and document types do AI models actually cite when they explain an answer? This program maps the citation graph at scale and quantifies the power-law distribution of referenced sources. For brands and PR teams, it reveals which media investments compound into AI authority — and which publications, despite high human traffic, are effectively invisible to models.
Recommendation Patterns
What makes an AI model move from mentioning a brand to actively recommending it? We decompose recommendation decisions into feature signals — recency, authority, specificity, sentiment, and structural placement — and run controlled experiments to isolate which factors carry the most weight in each model. The output is a Recommendation Score and a ranked list of intervention levers per brand.
Multilingual AI Influence
AI models do not behave identically across languages. This program compares English-language models (ChatGPT, Claude, Gemini, Perplexity) with Chinese-language models (DeepSeek, ERNIE, Tongyi) on identical brand query sets. Findings expose structural divergences in citation sources, entity recognition, and recommendation logic — critical intelligence for cross-border brands operating across Greater China and global markets.
AI Hallucination & Brand Risk
When a model describes a brand with the wrong founder, the wrong product, or the wrong headquarters, the cost is real and silent. This program quantifies hallucination rates by brand, by category, and by model, and tracks how long erroneous descriptions persist after a correction is published. We pair the dataset with a remediation playbook co-developed with our advisory team.
Cross-Model Consistency
If eight AI models give eight different answers about the same brand, which one should a buyer trust? Our consistency index measures answer divergence across the full model panel for every tracked brand, flagging entities where models disagree on facts, ranking, or recommendation. High divergence is a leading indicator of brand risk and an early-warning signal for our advisory clients.
Recent research releases
Executive summaries are free; full datasets are available to subscribers and accredited media on request.
Q1 2026 AI Influence Benchmark Report
A cross-vertical benchmark covering 480 brands across 12 industries — financial services, consumer tech, healthcare, manufacturing, luxury, retail, education, logistics, legal, real estate, travel, and FMCG. The headline finding: financial services posted the highest mean visibility at 72/100, while manufacturing lagged at 31/100 — a 41-point gap driven by under-investment in structured knowledge assets. The report also debuts our first longitudinal delta, comparing Q1 2026 against Q4 2025 baselines.
The Citation Economy Whitepaper
A first-of-its-kind map of where AI models actually source their claims. Across 1.2 million sampled citations, we document a steep power-law distribution: the top 50 publishers contribute roughly 80% of all citations, while the long tail of 14,000+ domains shares the remaining 20%. The whitepaper names the top cited domains per vertical, explains why legacy SEO authority does not map onto AI authority, and outlines the structural implications for media business models.
Cross-Border AI Influence Study
A side-by-side comparison of US-based English AI models (ChatGPT, Claude, Gemini, Perplexity) and China-based Chinese AI models (DeepSeek, ERNIE, Tongyi) on 240 cross-border brands. We quantify divergence in brand recognition, recommendation logic, and citation sources, and document three archetypes of cross-border influence failure. The study is co-released with two academic partners and includes a methodology appendix for replication.
AI Recommendation Engineering Framework
The public release of the KAIF™ methodology whitepaper. It formalizes the five engineering layers — Knowledge, Authority, Influence, Footprint — into a measurable framework, defines the inputs and outputs of each layer, and publishes the scoring rubric used by our advisory team. The framework is released under a Creative Commons license to invite peer review and external contribution from the broader AI-influence research community.
Built for reproducibility, not for headlines
Every score we publish can be re-derived by anyone who downloads our methodology notes and query library.
Data collection runs on a fixed panel of eight mainstream AI models — ChatGPT, Claude, Gemini, Perplexity, DeepSeek, Copilot, Baidu ERNIE (Wenxin), and Alibaba Tongyi Qianwen. For each tracked brand we execute more than 1,200 brand-specific queries per month, drawn from a documented library spanning discovery, comparison, recommendation, and troubleshooting intent. Queries are templated to control for phrasing variance, run from neutral accounts with no prior brand interaction, and re-executed on a rolling schedule to capture model drift without contaminating the panel.
From the raw responses we compute three first-class metrics. Visibility Rate measures how often a brand is named in a relevant response. Citation Index measures how often the model references a source the brand controls or has earned. Recommendation Score measures how often the brand is the model's primary recommended option. Each metric is normalized per vertical so cross-industry comparisons remain meaningful, and each is published with its raw distribution, not just its mean, so outliers are visible.
Transparency and privacy are non-negotiable. Our full methodology notes, query templates, and scoring rubrics are released alongside every report and can be downloaded from this page. We do not store user-submitted query content, we do not ingest private client data into the public benchmark panel, and we publish a methodology changelog whenever a metric definition, model panel, or query library is updated — so historical comparisons stay reproducible across releases.
Three ways to work with KHB Research
We collaborate with enterprises, media, and academic institutions — each under a clearly scoped engagement model.
Enterprises can commission confidential, single-industry benchmark studies and competitive visibility audits. All commissioned studies are clearly labeled as sponsored, and the underlying methodology remains identical to our public benchmarks — sponsorship buys scale and confidentiality, never a different result. Media partners can co-release industry reports under a joint byline, gaining early access to embargoed datasets in exchange for independent editorial review. Academic researchers can request complimentary access to full datasets for peer-reviewed work, with methodology support from our research staff.
Whatever the engagement, the same rules apply: methodology is published, data privacy is preserved, and conflicts of interest are disclosed. To scope a project, request a media partnership, or submit an academic research proposal, reach our research team at research@khb.com (a demonstration address for this site). We respond to qualified inquiries within three business days.
If you would prefer a structured conversation rather than email, our advisory team can also walk you through current benchmark findings and help you interpret where your brand sits in the latest release.
Frequently asked questions about KHB Research
Short answers on independence, access, custom studies, model coverage, and update cadence.
Is KHB Research independent from KHB's commercial services?
Yes. KHB Research operates with a structural separation from KHB's commercial advisory and platform teams. Research staff do not receive commission tied to client contracts, methodology and datasets are reviewed by an internal ethics committee, and published rankings cannot be edited by the sales organization. This separation is what allows our benchmarks to be cited by media, academia, and regulators without conflicts of interest.
How can I access full research reports?
Executive summaries and methodology notes are free to download from this page. Full benchmark datasets, brand-level scorecards, and the underlying query libraries are available to enterprise subscribers and accredited media on request. Academic researchers can apply for complimentary access by submitting a brief research proposal to research@khb.com.
Can our company commission a custom research study?
Yes. Enterprises can commission confidential, single-industry benchmark studies, competitive visibility audits, and KAIF-aligned methodology reviews. All commissioned studies are clearly labeled as sponsored, and the underlying methodology remains identical to our public benchmarks. Contact research@khb.com to scope a project.
What AI models does KHB Research track?
Our active panel covers eight mainstream models: ChatGPT, Claude, Gemini, Perplexity, DeepSeek, Copilot, Baidu ERNIE (Wenxin), and Alibaba Tongyi Qianwen. We refresh the panel annually and publish any additions or retirements in our methodology changelog so historical comparisons remain reproducible.
How often are benchmarks updated?
Flagship cross-industry benchmarks are published quarterly. Vertical-specific insight briefs ship monthly, and real-time anomaly alerts are pushed to subscribers whenever a tracked brand's visibility moves more than two standard deviations from its trailing 90-day baseline.