Is this different from GEO for a real estate brokerage?

Yes. A brokerage competes on local queries against portals and listing sites, and its content is about places. A software vendor competes on category and technical queries, and its strongest asset is a dataset sitting inside the product where nothing can read it. The mechanism is the same and the work is not.

We cannot publish our data, it is the product. What is left?

More than you would expect, and less than the full move. What gets published is an aggregate: a median, a distribution, a count, dated and scoped, at a level that answers a buyer's question without replacing a subscription. If a slice at that level would replace the product, it is the wrong slice. The rest of the work is entity clarity and source presence, neither of which requires publishing a row of anything.

Do you have proptech results to show?

No. We have no proptech clients yet, so no 90-day re-measurement exists. Our results page is empty and stays empty until a real re-measurement exists. What we can show is a property we operate, measured with the same instrument, on the REN.PH evidence page, which states at the top what it does not demonstrate.

GEO for proptech · Martenfield

Most writing about generative engine optimisation is written for brands. This is written for a company whose product is a dataset and whose buyer is an acquisitions analyst.

The mechanism is the one described in the other guides. What changes in this vertical is which part of it you are losing on.

What is different about selling software into commercial real estate

GEO for a brokerage or an agent is a local visibility problem. The queries name a place, the competitors are other firms in that place, and the sources the engines retrieve are portals and listing sites.

A proptech vendor is somewhere else entirely, for two reasons.

The asset is a dataset sitting inside a product. CompStak’s lease and sale comps sit behind an exchange account, where members earn credits by submitting comps of their own. Northspyre tells development teams they can standardise deal, cost and vendor data across a portfolio and build a proprietary database out of it, which puts that database inside the platform. Archer, Rockport VAL and ARGUS hold the underwriting and valuation inputs their models run on. Every one of those is close to the best available answer to questions buyers are putting to models today, and a retrieval step cannot see any of it.

The buyer is a specialist. They do not ask what proptech is. They ask what a lease abstraction tool costs, whether something integrates with Yardi, what industrial absorption did in a submarket last quarter. Narrow questions retrieve narrow pages, which is the good news in this vertical: the surface is small enough to own, and category commentary does not compete for it.

The four outcomes, and they move separately

Worth pulling apart, because one score cannot tell you which of them you are missing.

Does the model know what you are. Asked to describe your company, does it get the category right, or does it merge you with a similarly named firm. Consolidation is what breaks this: redIQ now presents as Radix Underwriting after an acquisition, and a model has to decide whether that is one company with a history or two companies with half of one each.

Does it cite your pages. Citation share.

Does it name you as a vendor. Recommendation share.

Does it keep you as the question gets specific. The broad category question and the narrow technical one frequently return different vendors, and the narrow one sits closer to a purchase.

Both metrics are defined, with the difference between them stated, on our method page. That page also says what a single run is worth, which is not much.

What moves each one

Entity clarity is schema and profiles. An Organization node with sameAs pointing at every profile you control, consistent naming across all of them, and the old name linked to the new one if you have been renamed or acquired. Cheap, usually missing, and covered in does ChatGPT read schema.

Citation share is publishing the data. An aggregated, anonymised, dated slice of what your product already holds, with a method note and a real table, at a URL a crawler can fetch. This is the expensive half of a proptech engagement, and the reason is engineering rather than a vertical premium, which is set out in what GEO costs.

Recommendation share is mostly not on your site at all. It comes from the sources that discuss options in your category: review sites, trade press, industry associations, the forums where operators actually ask each other. Which of those the engines prefer for your category is a measurement rather than a guess, and the method is in how to get cited by ChatGPT.

Staying in as the question narrows comes from owning the specific answer rather than publishing a page about the topic. Same work as citation share, pointed at the long tail of technical questions instead of the headline ones.

What does not work

The volume content retainer. Twelve posts a month on the future of proptech competes with every other vendor publishing the same twelve posts, and not one of them is the best available source for anything. A single page holding a figure nobody else publishes outperforms the lot.

The advertorial version of digital PR. A paid placement that reads as a paid placement is not a source an engine prefers, and the sites selling them in volume are rarely in anybody’s source map. Getting into the sources that are already retrieved is slower and is a different activity, done by different people.

Buying a tracking tool and calling it a programme. The trackers report brand numbers. None of them reports whether your data pages are being used as sources, which is the gap set out in AI visibility tools compared.

What we have not measured

The proptech census has not run. We have published no measured baseline for CRE underwriting and valuation software, so nothing on this page is a finding about how proptech companies currently perform in AI answers. Everything above is what the mechanism implies, and it is written as that rather than as a result.

The engines change retrieval behaviour without notice and without a changelog. A number that moves may be a model update rather than anything anyone did.

We have no proptech clients yet, so no 90-day re-measurement exists. There is no roster. What we can show instead is a property we operate: the REN.PH evidence page publishes both console exports and says how much of each console’s own total they account for, which in Bing’s case is about half. It states at the top what it does not demonstrate.

If you want the size of your own problem before committing budget to any of this, the Diagnosis is the fixed-scope version, and its price is on the page.

Questions

Is this different from GEO for a real estate brokerage?: Yes. A brokerage competes on local queries against portals and listing sites, and its content is about places. A software vendor competes on category and technical queries, and its strongest asset is a dataset sitting inside the product where nothing can read it. The mechanism is the same and the work is not.
We cannot publish our data, it is the product. What is left?: More than you would expect, and less than the full move. What gets published is an aggregate: a median, a distribution, a count, dated and scoped, at a level that answers a buyer's question without replacing a subscription. If a slice at that level would replace the product, it is the wrong slice. The rest of the work is entity clarity and source presence, neither of which requires publishing a row of anything.
Do you have proptech results to show?: No. We have no proptech clients yet, so no 90-day re-measurement exists. Our results page is empty and stays empty until a real re-measurement exists. What we can show is a property we operate, measured with the same instrument, on the REN.PH evidence page, which states at the top what it does not demonstrate.