Every week, thousands of businesses ask why AI recommends one company instead of another.
Most assume the answer is rankings.
Increasingly, it isn't.
The next generation of search is driven by understanding.
And that raises a new question:
How well do AI systems actually understand a business?
That question led to the Illuminate Research Initiative — an ongoing editorial research program led by Firefly Web Labs and supported by research performed using Illuminate, the AI visibility analysis platform developed by Faeth. This report introduces the initiative: what it studies, how it is conducted, and what it has found so far.
Why This Is A Research Initiative, Not A Partnership Announcement
It would be easy to describe this as a partnership between two companies. That description would also undersell it. A partnership is an arrangement. A research initiative is a body of work — one that accumulates findings, refines its own methodology, and produces evidence that holds up over time.
The Illuminate Research Initiative is the name for that body of work. Firefly Web Labs created and leads the initiative, drawing on Illuminate, the AI visibility analysis platform Faeth has developed, as one of its primary research inputs. Illuminate generates initial findings on how AI systems interpret a given business; Firefly designs the research program, validates those findings against real businesses, and publishes what is learned. Every report that follows this one — case studies, industry analyses, methodology updates — belongs to the same initiative, numbered and indexed as part of a single, continuous research record.
Editorial Standards
A research initiative is only as credible as the discipline behind it. The Illuminate Research Initiative publishes observations only after they have met the following standards:
- Observed across multiple businesses, not a single example
- Manually reviewed against the live website
- Independently validated by Firefly's editorial team
- Reproduced under varying conditions
- Documented with enough specificity to be tested by others
- Speculative conclusions presented as findings
- Fabricated or invented statistics
- Generalized claims drawn from isolated examples
- Predictions framed as documented behavior
Research Methodology
Research is only as credible as the conditions it was tested under. Firefly did not validate Illuminate's findings against a narrow or convenient sample. The validation set was deliberately built for variation — across industry, platform, market, and authority level — because a finding that only holds for one type of business isn't a finding about AI understanding. It's a finding about that business.
Contractors, attorneys, accountants, manufacturers, healthcare providers, nonprofits, consultants, home service companies, sports organizations, hospitality businesses, and other professional services.
Single-page sites through large multi-location architectures, built on WordPress, Wix, Squarespace, Shopify, and custom platforms.
Single-location local businesses, regional multi-location operators, and national service organizations.
Businesses ranging from newly established entities with minimal external signal presence to decades-old organizations with substantial — but not always well-documented — reputations.
Each business in the sample is evaluated by Illuminate, then independently reviewed by Firefly against the live site and its available external signals. Where Illuminate's findings and Firefly's manual review agree, the pattern is treated as provisionally validated. Where they diverge, the case is set aside for further study rather than forced into a conclusion. That discipline is slower. It is also the only way the resulting findings are worth publishing.
What We Are Not Measuring
The clearest way to define a new field of research is to say what it isn't. The Illuminate Research Initiative is not a search engine optimization study, and the metrics it tracks reflect that distinction deliberately.
- Keyword rankings
- Backlink counts
- Domain or SEO scores
- Page speed in isolation
- Understanding — what AI can accurately state about a business
- Confidence — how certain AI is in that understanding
- Recommendation potential — whether that confidence is sufficient to be named
- Entity clarity and semantic consistency across the site
This is not a rejection of SEO. Traditional SEO and AI visibility share infrastructure — both depend on a website that is well structured and accurately described. But ranking well has never guaranteed that an AI system understands a business correctly, and the initiative's findings so far suggest the two are more loosely correlated than most businesses assume.
Throughout this research, Firefly uses the term AI Confidence to describe the degree of certainty an AI system has when describing, classifying, or recommending a business. For purposes of this research initiative, confidence is treated as distinct from visibility. A business can be visible to an AI system — findable, indexed, mentioned — and still be described with low confidence, in generic terms, without being recommended.
This is a working term, not an established industry standard. Firefly is introducing it here because the existing vocabulary around AI visibility does not yet distinguish between a business AI knows about and a business AI is confident enough to recommend — and that distinction has been consistent enough across the validation sample to be worth naming.
The Six Dimensions Of AI Confidence
Across the validation sample, recurring patterns in what raises or lowers AI Confidence began to organize themselves into six consistent dimensions. This framework is early — it will be refined as the sample grows — but it has held up consistently enough across industries to publish as the initiative's working model.
Whether a first-time visitor — human or AI — can determine what the business does and who it serves without prior context.
Whether the business is represented as a single, consistent entity across its website, directories, and structured data — or fragmented into conflicting versions.
Whether site architecture, headings, and content hierarchy are organized around the questions customers actually ask, in language AI systems can parse.
Whether credentials, reviews, affiliations, and history are present, current, and machine-readable.
Whether independent, third-party sources corroborate what the business claims about itself — beyond its own website.
Whether the combined signal across the prior five dimensions is sufficient for an AI system to name the business unprompted in a relevant query.
The dimensions are sequential in practical effect, if not strictly causal: a business with strong authority signals but poor website clarity is rarely recommended, because the AI system cannot resolve what it is recommending the business for. This is the framework the initiative will reference, test, and refine in every report that follows.
Emerging Findings
The findings below are not statistics. They are observations — patterns specific enough to be useful, provisional enough to be revised as the sample grows. Each is numbered as part of the initiative's ongoing record.
#001
#002
#003
#004
Research Questions We'll Be Publishing
Each report in the Illuminate Research Initiative will take on a specific, testable question rather than a general theme. The current research queue includes:
Roles Within The Initiative
Created and leads the Illuminate Research Initiative. Designs the research methodology, validates findings against live businesses, implements the structural and semantic changes they suggest, and publishes the resulting reports.
Develops Illuminate, the AI visibility analysis platform that generates initial findings on how AI systems interpret a given business across discovery surfaces — one of the initiative's primary research inputs.
Findings from Firefly's implementation work feed back into Faeth's continued development of Illuminate, and the cycle repeats with the next business, the next industry, the next report.
Ask an AI system what it knows about your business, and compare the answer to how confidently — not just how accurately — it responds. Hesitation, hedging, and generic language are signs of low AI Confidence even when the underlying facts are correct. That gap between accurate and confident is the starting question of every report this initiative will publish.
What This Means For The Web
The internet was built for people.
The next generation of the web must communicate equally well with people and AI.
That transition will not happen through guesswork.
It will happen through research, implementation, measurement, and continuous improvement.
The Illuminate Research Initiative exists to document that evolution.
Firefly Web Labs will continue that work through future research reports, implementation studies, and cross-industry analysis.
- Search Engine Land. Coverage of AI-Driven Search and Entity Evaluation. Search Engine Land, 2024–2026. Ongoing industry reporting on how AI discovery platforms evaluate and rank business entities.
- Google. Search Quality Evaluator Guidelines. Google, 2024. The E-E-A-T framework underpins how AI systems weight trust signals and source credibility when interpreting a business.
- Semrush. State of Search 2024. Semrush, 2024. Industry data on the growing prevalence of AI-generated answers and the signals associated with inclusion in those answers.
About The Illuminate Research Initiative
The Illuminate Research Initiative is an ongoing editorial research program published by Firefly Web Labs to study how AI systems interpret, understand, and recommend business websites. Reports combine real-world website analysis, implementation findings, cross-industry observations, and practical guidance to help businesses improve AI visibility and digital clarity. Learn more in the AI Visibility Hub or explore Firefly's Research and Insights archives.
Future reports in this initiative will be published at fireflyweblabs.com/research.
