Research
We don't just talk about AI. We measure it.
Original measurement is the only thing that makes a company in this category worth citing. This page is the programme — what we are measuring, how, and exactly how far along it is.
Nothing published yet — and we are not going to pretend otherwise
Greenbug has collected zero live AI observations to date. The measurement infrastructure is built and the first baselines are in progress, but no study below has results, so none is shown with numbers.
A research page full of invented statistics would be the fastest possible way to disqualify everything else on this site.
Research in progress
Each study states the question it is designed to answer and the method that will answer it — published before the results exist, so the method cannot be retrofitted to whatever the data turns out to say.
- Research in progressQuarterly
DMV AI Recommendation Index
Which businesses do AI assistants actually recommend across DMV service categories, and how consistently?
Method: A fixed benchmark prompt set per category, run across the benchmark platforms with repetition, scored with the published methodology. Businesses are included on measurement alone — inclusion is never purchased and being a Greenbug client cannot affect placement.
Method, planned output and limits → - Research in progressQuarterly
AI Citation Studies
Which websites do AI assistants actually rely on when recommending local businesses in this region?
Method: Extraction and frequency analysis of every source cited across the observation corpus, broken down by platform, category and geography.
Method, planned output and limits → - Research in progressOngoing
Local Business Recommendation Studies
How does question phrasing — city versus county, 'near me' versus a named place — change which businesses an AI recommends?
Method: Controlled prompt variants differing only in geographic phrasing, run in the same period, comparing the resulting business sets.
Method, planned output and limits →
Planned
These require a larger corpus than currently exists. They are listed so the programme is legible, not to imply they are imminent.
- Planned
State of DMV AI Recommendations
How is AI recommendation behaviour across the DMV changing over a full year?
Annual
- Planned
Platform Comparison Research
How often do ChatGPT, Gemini, Claude, Perplexity and Google AI agree on which local businesses to recommend?
Quarterly
- Planned
Industry Recommendation Benchmarks
What does a typical AI Recommendation Score look like in a given category, so a business can tell whether its own score is good?
Quarterly
The publishing commitment
Stated publicly because a cadence you publish is a cadence you can be held to. It begins with the first completed measurement cycle.
| Frequency | Output |
|---|---|
| Monthly | One original mini-study using our own measurement data |
| Monthly | One case-study update showing real movement, or no movement |
| Monthly | One category or city AI recommendation leaderboard |
| Monthly | Two evidence-based educational articles |
| Quarterly | The DMV AI Recommendation Index |
| Annual | State of DMV AI Recommendations |
Rules that make this worth reading
A ranking that can be bought is an advertisement. These constraints apply to every index, leaderboard and study Greenbug publishes.
- Rankings cannot be purchased. There is no paid placement in any Greenbug index.
- Being a Greenbug client cannot raise a score or a ranking. Client scores are computed by the same formula from the same kind of evidence as everyone else's.
- Where Greenbug has a commercial relationship with a business that appears in published research, that relationship is disclosed on the page.
- If sponsored placement is ever introduced, it will be visually and structurally separated from editorial rankings and labelled as sponsored.
- Greenbug.ai measures itself with the same methodology and publishes its own result, including when it is unflattering.
Want to be measured?
Client measurement and published research use the same methodology — but a client relationship can never affect an editorial ranking.