The Greenbug experiment
Building Greenbug.ai in public
We are applying the methodology to ourselves, on the same schedule, with the same rules — and publishing what happens.
Greenbug.ai is a new brand in a category most business owners have not heard of yet. By our own argument, that means AI systems currently have very little reason to recommend us when someone asks who can help with AI visibility in the DMV.
That is an uncomfortable thing for a company in this business to admit, and it is also the most useful experiment available to us. We are in exactly the position our clients are in, we are going to do the work we would do for them, and we are going to publish the measurement either way.
We don't just tell you what we changed. We measure whether AI visibility changed.
Measurement underway
No baseline has been collected yet, so there is no score on this page. There will not be one until 240 recorded observations exist to compute it from.
If we published an encouraging number now, nothing else on this site would deserve to be believed.
What we are committing to publish
Including the parts that make us look bad. A case study that only ever goes up is a marketing asset, not evidence.
| What we will track | Why it matters | Status |
|---|---|---|
| Baseline AI Recommendation Score | Our own score, computed by the same formula from the same kind of evidence as any client's. | Measurement underway |
| AI recommendation presence | How often we are named when someone asks an AI assistant who does AI visibility work in the DMV. | Measurement underway |
| Citations earned | Which sources actually cite our research, and which never do despite the effort. | Measurement underway |
| Authority growth | Third-party corroboration accumulating — or not — across the sources that matter in this category. | Measurement underway |
| Source gaps identified and closed | The sources repeatedly cited in answers about this category, which of them feature competitors and not us, and which we actually manage to close. | Measurement underway |
| Changes implemented | Every change made to our own entity data, site and external footprint, dated — so a movement in the score can be attributed to something rather than guessed at. | Measurement underway |
| Content published | What we published, when, and whether any of it moved a measurable number. | Measurement underway |
| External mentions | Where we get referenced without asking. | Measurement underway |
| AI visibility change | Movement against the baseline, cycle by cycle, attributable because the benchmark never changes. | Measurement underway |
The rules we are holding ourselves to
- The same benchmark. Our prompt library is built the same way a client's is, frozen the same way, and run on the same schedule.
- The same formula. No separate scoring for ourselves, no adjustment, no rounding in our favour.
- Publish flat results. If a quarter of work moves nothing, the page says it moved nothing.
- Publish the failures by name. Content that earned no citations, sources that never materialised, tactics that did not work.
- No retroactive editing. When the methodology version changes, both versions are shown rather than history being rewritten.
- Being our own client changes nothing about any public ranking. We are excluded from editorial placement in our own indexes on principle.
Why this is worth following
Every agency in this emerging category is making the same claim: that it can improve how AI systems talk about your business. Almost none of them will show you a measurement of themselves.
This page is the test. If the methodology works, it should be visible here first — and if it does not, that will be visible here too.
Alongside this study
DEL Trees is the other end of the same question: a brand-new tree services division with no website and no history, which lets us measure AI visibility essentially from zero and document what happens as authority, citations, content and external evidence are built.
One established brand in a new category, one new brand in an established category. Between them they test whether the methodology describes something real.
All case studies →Related
The methodology we are using on ourselves →
Weights, denominators, limitations and conflicts policy, in full.
Research programme →
The original measurement this experiment is meant to produce.
The AI Recommendation Score →
The number this page will eventually carry.
Case studies →
The businesses being measured alongside us.
Run the same measurement on your business
Same benchmark, same formula, same rule about publishing what it actually finds.