- Client
- NoorRank Labs (our own research division)
- Industry
- Open AEO research
- Engagement started
- Results measured
Results
| Measure | Before | After | How it was measured |
|---|---|---|---|
| ChatGPT entity recognition | Not recognized — returned information about the fruit | Recognized day 9, cited noorranklabs.com, stable 11+ days | Fixed direct and entity-focused queries run in clean sessions and logged as dated milestones |
| Perplexity entity recognition | Not recognized — returned information about the fruit | Recognized day 17, lost day 19, recovered day 20 | Same fixed query set; the day 19 loss and day 20 recovery are both recorded in the public log |
| Claude entity recognition | Not recognized — returned information about the fruit | Still not recognized when the experiment closed on day 21 | Same fixed query set, tested through to end of experiment |
| Google Search impressions | 0 — domain registered day 1 | 257 impressions, 10 clicks by day 21 | Google Search Console, measured over the 21-day window |
| Time to first page indexed | No domain existed | Homepage indexed on day 1 | Search Console indexing status, after sitemap submission on the day the domain was registered |
| External corroborating sources | 0 | 5 by day 10 — Medium, Dev.to, Reddit, LinkedIn, GitHub | Counted as published, externally-hosted references to the entity |
The challenge
Every AEO agency claims it can make a brand visible to AI. Almost none will say how long it takes, on which platforms, or what happens when it does not work — because doing so means publishing the failures alongside the wins.
There was also a harder problem hiding in the name. "Miracle berry" is an established entity: a West African fruit with a Wikipedia page and decades of search history. A brand-new project sharing that name starts out competing against a well-understood thing, which is the same disambiguation problem faced by any company whose name is also an ordinary word.
So we ran it on ourselves first. We registered a domain with no history, no backlinks and no brand, and committed in advance to publishing whatever happened over 21 days.
What we did
Day 1: domain registered, a public GitHub repository created, Search Console verified and the sitemap submitted the same day. The homepage was indexed within 24 hours.
Entity architecture first. Persistent Schema.org identifiers and consistent descriptions across every page, written specifically to distinguish the protocol from the fruit — the disambiguation problem handled in structured data rather than in copy.
Corroboration second. Between day 3 and day 10 the work was documented externally: a Medium article on the methodology, a Dev.to technical piece on the modular schema architecture, and posts to Reddit and LinkedIn. Five independent sources by day 10.
Then measurement, daily. A fixed set of direct and entity-focused queries run against ChatGPT, Perplexity and Claude, with every result logged as a dated milestone — including the days when recognition disappeared.
The outcome
ChatGPT recognized the entity on day 9 and cited noorranklabs.com directly. That recognition held for the remaining 11+ days of observation without degrading.
Perplexity was the interesting result. It recognized the entity on day 17, lost it entirely by day 19 — reverting to fruit information — then recovered it on day 20 with more detail than before. A first citation is not the same as durable recognition, and we would not have learned that from a shorter experiment.
Claude did not recognize the entity at any point in the 21 days. We published that too.
Alongside the AI results, conventional search responded faster than expected: indexed on day 1, and 257 impressions with 10 clicks by day 21 from a domain that did not exist three weeks earlier.
The honest summary is two of three platforms, one of them unstable, from a standing start in under three weeks. This is one case study on one domain — it is not a guarantee of timing for any particular brand, and platform behaviour shifts with model updates and query wording. The full day-by-day log, the structured data implementation and the raw data are public on noorranklabs.com and GitHub.