Retainer

Who this is for

Companies holding proprietary data that their buyers ask about and their market cannot currently read. This is where we recommend starting after a diagnosis, and it is the default retainer.

How it runs

1. 01

   Baseline

   The diagnosis, on the same query set, before any work starts. Without it the 90-day number means nothing.

2. 02

   Technical and entity layer

   Crawler access, rendering, schema, and entity architecture, verified against live crawler requests rather than against a testing tool.

3. 03

   Publishing pipeline

   We build the path from your database to public, readable pages. This is the part that compounds, because it keeps producing assets after we leave.

4. 04

   Weekly tracking and a monthly report

   The contracted query set runs weekly. The report states the movement and what caused it, and where we cannot attribute a change to our work, we say so.

5. 05

   Re-measurement

   Same queries, same engines, same run count, 90 days after the baseline. We publish the delta to you, including when it is smaller than we hoped.

Measurement is the same at both retainer prices

Same depth on both tiers

Both retainer tiers track the same 50 queries across the same 5 engines at 3 runs each, against the same competitor set, on the same reporting cadence. We do not charge more to run more prompts.

Why depth is not what the price buys

Collection is the cheapest part of this work, and the guide to what GEO costs works through why. Pricing on it would put the premium on the part that costs least and invite comparison with dashboards that sell tracking by the prompt. Nobody gets a shallower read for paying less.

What the higher tier adds

Work. The $4,000 tier adds earned-mention work on top of everything in the $2,500 tier. The view of that work, the queries, the engines, the competitors and the report, is identical.

How will we know if any of this produced leads?

What we install

First-touch capture of the referrer and the landing URL, query string included, cookied at session start and carried through to the form and into your CRM as hidden fields. Source classification into a clean label, with the raw string kept beside it. An optional field on lead forms asking how the buyer found you. Edge logging of AI user-fetch bots.

Why first-touch matters

Most sites store last-touch, so a visitor who reads a second page before converting loses the source they arrived on. The referral gets attributed to internal navigation, or to whichever search engine they passed through on the way. Correcting this is usually the single largest change to a client’s existing numbers.

The part that is not commodity work

Referrer tracking is a two-day job any agency could copy. Edge logging of ChatGPT-User, Perplexity-User and Claude-User hits, joined to the page each one fetched, is not. Those hits mean a person asked a question and the model went and read that specific page to answer it, which happens whether or not anyone clicks. Most of the time nobody does, and no referrer records the ones who do not, so a fetch log is the only place that reader appears at all. We do not publish a click-through rate for AI citations, ours or anyone else’s, because we have not measured one.

We report a floor, never a count

A share of leads arrive with no referrer at all, and the share is larger for AI sessions, because mobile app webviews strip the header. We do not put a percentage on either, ours or a borrowed one, and the floor rule does not need one: whatever is missing is missing upward. Google AI Overviews and AI Mode clicks cannot be told apart from ordinary organic search on top of that. Every dashboard we deliver states all of this in the interface rather than in a footnote, because a clean-looking figure of 3 AI leads without it is the same overclaiming we criticise everywhere else on this site. Once your own data can measure your own missing-referrer rate, we report that instead of the disclaimer.

Privacy

A referrer and a landing URL attached to a named lead record are personal data. It is covered in the engagement letter, and your privacy policy needs a line about it.

The proptech capability list, item by item

Authority built around neighbourhood and market data

This is the centre of the build rather than one item in it. Your neighbourhood and submarket data is the asset a model has no other source for, and it is usually the part sitting inside the product where nothing can read it. We build the path from that database to public pages, one page per entity, each owning a specific figure.

Original market reports

Yes, and this is what the source-grade assets are. A report a model will cite as evidence has a defined method, a stated sample, dated data and figures a reader could check. Most market reports published in this category have none of those, which is why they get skimmed as marketing rather than cited as a source.

Structured property content

Yes. The publishing pipeline is the deliverable and it keeps producing after we leave, which is the difference between this and a content project. Structure here means one resolvable entity per page with a stable identifier, not a template applied to a spreadsheet.

Schema markup

Yes, and it is the cheapest item on this list, which is why it is oversold. Schema tells a parser what a page is about. It does not make a page worth citing, and a correct schema graph over thin content changes nothing. We do it as part of the entity architecture rather than as a line item.

Entity optimisation for the company, its executives and its products

Yes. The company, the products and the named executives each resolve as entities with stable identifiers, cross-linked, and reconciled against the public records the engines already hold. Executives matter more here than in ordinary search, because a model asked who leads a category answers with people as often as with companies.

Digital PR with real estate publications

On the $4,000 tier. This tier produces the data assets a pitch needs, and the outreach, source acquisition and comparison-layer placement are the earned-mention work the higher tier adds. That page states exactly what we do and do not do there.

What continues once the build is done

A page can lose ground without anything happening to it

A figure that was current when you published it becomes the second most recent version of itself the moment somebody publishes a newer one, and at that point it stops being the lowest-risk way to ground a claim. Nothing about your page got worse and nobody outranked you. We have not measured how fast that happens, to us or to anyone, and we will not put a decay rate on it, so treat this as the reason the tracking continues rather than as a countdown.

Which is an argument for cadence, not for volume

The answer is updating what already earns citations on a schedule, not publishing more. Our re-measurement cadence is 90 days for the same reason: it is short enough to catch a slide while it is still a slide. If your data updates quarterly and your pages do not, the pages are wrong for most of every quarter and the engines find out before you do.

What that costs you if you do it yourself

Less than this retainer, and that is a real option. The pipeline the build installs runs on your infrastructure, the pages are on your domain, and nothing here is designed to make leaving expensive. What you would be taking on is the weekly tracking and the judgement about where the source map moved.

Questions

Where is the build fee?

Inside the retainer. The build is done in the first 3 months on both tiers, and it is never priced separately. The first 90 days on this tier total $8,700 with the diagnosis included, and that is the whole first commitment.

Why a 3-month minimum?

Short enough to run alongside an incumbent agency rather than asking you to switch first. Long enough to finish the build and take one re-measurement at 90 days, which is the first point where a number means anything.

Do we need the diagnosis first?

Yes. The retainer works from a baseline and a source map, and both come out of the diagnosis. Without them there is nothing to measure the work against, which means paying monthly for work you have to take on faith.

Who owns what you build?

You do. The pipeline runs on your infrastructure and the pages sit on your domain. There is no dependency designed into the engagement.

Will we be able to attribute revenue to AI answers?

No, and neither can anyone else. We instrument the part that is measurable, we label the rest as unmeasured, and on this tier we contract on citation share, because it is the outcome this work moves and we can measure it identically before and after.

What happens if the numbers do not move?

We show you that they did not, and why we think so. Model behaviour changes and some categories are harder than others. The tracking is honest in both directions or it is not worth running.