About
We were the case study before we were the firm.
Martenfield builds measurement instruments for AI answer engines and publishes what they find. We track a 32-token crawler registry across 16 operators, and we do the entity architecture and data publishing work that changes what proptech and CRE software companies get cited for.
What we have built
Built
The crawler registry
A table of 32 AI crawler product tokens across 16 operators. Every row carries the vendor page it was read from, the date it was read, and what the operator states about honouring robots.txt for that token. 12 of the 16 operators publish a crawler page, and all 12 were read end to end on 2 August 2026. The other 4 publish no page at all, so their tokens reach the table through third-party blocklists and are labelled undocumented rather than counted as though a vendor had confirmed them. It is re-read before every edition, because a stale token in a dataset about tokens produces a number that looks fine and is wrong.
Published
The measurement method
Written down before it was sold, and published in full. The measurement queries 5 engines at 3 runs per query, 50 queries in a diagnosis and 30 in an index category, recording per answer whether a company was named, cited, or absent. Those three stay separate. A single run is an anecdote rather than a measurement, and the method page says so in the section that explains the run count.
Current: v1.1
Method versioning
Every figure on this site carries the version that produced it. When the method changes the version changes, and figures published under an earlier version keep their original stamp rather than being restated under the new one. The one change so far split a single reported number into two, and the superseded version stays documented on the method page instead of being replaced.
Not run
The category census
The index of which commercial real estate software companies AI engines recommend does not exist yet. The data contract is written and the method above is the method it will run under, and no edition has been collected. There is no route for it and there will not be one before there is data behind it, because a page that exists is retrievable by anyone who guesses the URL whether or not it is linked.
The registry is not published yet, so a reader cannot check it. Take it as a statement of what we run rather than as evidence, and hold it to the same standard as the census above: it counts once it is readable by somebody other than us.
Where this came from
We operate REN.PH, a Philippine real estate data platform holding 234,337 zonal value rows and 25,264 broker records, structured to barangay level across 37,660 barangays. It was built as a data business, not as a marketing exercise, which meant entity architecture, provenance tracking, verification, and freshness systems had to work before anyone was sold anything. A national property dataset does not function without them.
In June 2026 a US commercial real estate company asked ChatGPT who does real estate AI search visibility. The answer pointed them at RealEstateSEO.ph, the service brand we run beside the platform, and they reached us through the parent consultancy. We know that because we asked the buyer and they told us.
This page said until 5 August 2026 that the buyer found REN.PH through ChatGPT. That was wrong, and it was wrong in the direction that flattered us. The engine recommended a service. The dataset was not the artifact it named, and an enquiry about the service is thinner evidence for the mechanism this firm sells than a citation of the data would have been. It is still one enquiry traced to an AI answer by the only method available for it.
Three weeks later a Zurich proptech company found REN.PH through Google Search and made contact on LinkedIn. That is organic search. It is not an AI referral, we never asked what they searched for, and it is written here because it happened rather than because it supports anything.
Separately, REN.PH publishes Philippine zonal values, the government land valuations, as pages down to barangay level. Lead capture on those pages is source-tracked. Between 14 May and 1 August 2026, a 79-day window, they produced 133 leads. Exactly one carried a chatgpt.com referrer, and it landed on a barangay-level page rather than on the homepage.
So we are testing it in the open
Every inbound enquiry across our properties now records how the buyer found us. We publish the running count of AI-sourced enquiries once it clears five, and we publish it whether or not it flatters the thesis.
An enquiry counts on one of two conditions. Either a referrer header named an AI tool, or we asked the buyer and they named one. A guess offered without being asked does not count, and neither does a Google referral, whatever the buyer believes sent them.
Until then the honest description of what we have is an anecdote with a plausible mechanism behind it, not evidence.
How we work
Client names are public, deal terms stay private
That a company is our client is public, and its name may be published. Publication rights are written into the contract at signature. Deal specifics stay private: what the client paid, its internal data, its strategy, and anything else the contract protects. Where a client appears in a table we publish, its row is marked as a client in the table itself, in every edition, permanently. Not in a footnote, and not on a separate disclosure page.
The niche is held narrow
Proptech and commercial real estate technology only. Broader framing tests better in the abstract and fails in practice: our evidence is a proptech platform cited into a proptech buyer's answer and it does not transfer, and query packs and source maps only compound inside one category.
Method is versioned and published
Every measured figure on this site carries a stamp saying when it was collected, where it came from, and which method produced it. A citation measurement names the engines queried and the runs per query, currently v1.1. A count that came from somewhere else, lead capture for instance, names that instead of borrowing a method it was not produced by, and where one block of figures has two sources the stamp says which figures each one accounts for. When the method changes, the version changes, and old numbers keep their old stamp rather than being quietly restated.
We measure ourselves too
martenfield.com sits in the same tracking pipeline as client work, against its own query set. We publish those results quarterly, including the early ones where we do not appear at all. It is the one case study whose permission we do not need.
Who runs it
Aaron Zara Principal
Aaron Zara is the Principal of Martenfield, and founded it in 2026. A PRC-licensed real estate broker since 2015, with 18 years across software engineering and digital marketing. Builder of REN.PH and operator of a portfolio of Philippine property and data businesses. Most of that work has been on systems where the data model was the product rather than a detail behind the marketing site.
Martenfield is based in Santa Rosa, Laguna, in the Philippines.
Martenfield exists because the same architecture that makes a property dataset trustworthy to a human is what makes it legible to a model, and almost nobody in proptech has built either.
Not a fit
- Individual agents and local brokerages
- No proprietary data to publish
- Budget under $2,500 a month after the diagnosis
- First question is the monthly rate
A written disqualify list is cheaper than a bad engagement, for both sides.
If your first question is the monthly rate, the honest answer is that we are probably not the right firm. Start with the diagnosis, or do not start.