The classic advice for agent blogs is a numbers game: here are 115 ideas, post weekly, stay consistent, and traffic will come. Meanwhile the systems that increasingly broker introductions, AI assistants answering movers' questions, quote almost none of what that advice produces. The gap between those two facts is this post. You do not need 115 ideas. You need the five page types machines actually cite, and the discipline to skip the rest.
Why most blog ideas earn nothing now
Read any big list of real estate blog ideas closely and one property repeats: the entries are generic. Spring staging tips. Questions to ask your realtor. A moving checklist. Renting versus buying. These were always weak differentiators, but AI search makes the weakness terminal, because an assistant asked a generic question answers it from general knowledge. It does not need your page, so it has no reason to retrieve it, and no citation ever happens.
The filter for every content idea is one question: could a machine answer this without me? If yes, the idea is exercise, not marketing. What survives the filter is knowledge with an owner, which is also precisely what Google's own helpful content guidance selects for: content demonstrating first-hand expertise, made for people, with an explicit warning against automation used primarily to manufacture rankings. Two different systems, Google's rankers and chat assistants' retrieval, converge on the same scarce ingredient: what only you know.
Page type one: the neighbourhood page built from presence
The strongest citable asset an agent can own, and the rarest. Not the portal version, population, average price, close to transit, which is generated identically for every place on earth. The version only presence produces: which blocks flood in November, which buildings carry levy histories, where visitor parking actually is, what sells in a week versus what sits two months, and why.
We called this the generation test in the agent websites teardown: if ChatGPT could have written your neighbourhood page without visiting, the page earns nothing. Aim for the opposite: the page an assistant has to quote because nothing else on the internet contains its facts.
Page type two: the answer page with a local variable
"Can my landlord sell the condo I rent?" "What is a depreciation report and when should it scare me?" "How do assignment sales get taxed here?" Questions like these have a shape: general frame, local variable. The generic national answer matches every jurisdiction weakly. The page that names the actual rule in your province or city, in plain language, with the date it was last checked, matches one jurisdiction strongly, and strong match is what retrieval selects.
These pages also do quiet double duty: they are what your AI-referred leads actually land on, arriving with exactly the question the page answers.
Page type three: the market note that interprets
Monthly stats posts fail the filter: board numbers are public, machines already have them, recitation adds nothing. What passes is interpretation, clearly attributed as your observation: what you are seeing at showings that the averages hide, where offer behaviour changed first, which segment the headline number misrepresents. Date-stamp these visibly, because a market note quoted a year late becomes misinformation with your name on it. The stats are the commodity. Your read is the content.
Page type four: the process guide anchored to your jurisdiction
Buying and selling processes are governed locally: disclosure rules, deposit conventions, tenancy law, closing mechanics. A process guide that walks one transaction type through your jurisdiction's actual rules, plainly, is durable, citable, and nearly always missing, because writing it requires knowing the process rather than paraphrasing someone else's checklist. One caution that follows from the buyer data: most of your readers will still hire an agent, per NAR 88% of buyers purchased through one, so the guide's job is not to replace yourself but to be the demonstration of competence that makes you the agent they hire.
Page type five: the record that verifies you
The last type is not a blog post at all, and it gates the other four. When an assistant weighs quoting your neighbourhood page, it checks who wrote it: your bio's facts block, your business profile, your reviews, and whether they agree. An unverifiable author weakens every page they publish. This is why identity work comes first in the sequence, before the first neighbourhood page, not after the fiftieth generic post.
How to get the knowledge out of your head
The most common objection to this strategy is not disagreement. It is that writing feels slow, and the knowledge that would make a page citable is tacit: you know which buildings to avoid, but you have never written a sentence explaining why.
Three habits solve this cheaply. Keep a running note of the questions clients actually ask you, verbatim, because those are your content calendar and they arrive free every week. After a showing or an inspection that taught you something, dictate two minutes into your phone while it is fresh and let a transcript be the raw draft. And when you catch yourself explaining something twice to two different clients, treat the second time as the signal: that explanation should be a page.
The failure mode to avoid is asking an AI tool to generate the page from the title. It will produce a competent, generic article, which is precisely the thing that earns nothing, and Google's guidance separately warns about automation used primarily to manufacture rankings. Use tools to tidy your sentences, never to supply the substance. The substance is the only part with value, because it is the only part nobody else has.
The cadence that sustains it
The old advice said weekly, because volume proxied for freshness. The citable strategy inverts it: one necessary page a month, built properly, beats fifty-two skippable ones a year, and it is a pace a working agent can hold. A realistic first quarter: month one, fix the identity layer and pick your two core neighbourhoods. Month two, write the first presence-built neighbourhood page. Month three, the first answer page with a local variable, chosen from questions clients actually asked you that quarter. Every page goes through the same one-question filter before a word gets written: could a machine answer this without me?
Two small disciplines keep that library from turning into a liability, and both take seconds.
Date what decays. Market observations, rules, and fee structures all age. Put a visible last-checked date on those pages and honour it. The risk is specific: an assistant that quotes your undated 2024 market read as current advice has made your page the source of a wrong answer, with your name attached.
Write down where each fact came from. When you state a rule, note the regulator, board, or statute you took it from, in the sentence. This is the same standard we hold ourselves to, and it does double duty. Readers can check you, and assistants treat sourced statements as safer to quote than bare assertions.
Neither habit makes writing slower once it is a reflex. Both make the difference between a page that earns trust for years and a page you eventually have to quietly delete.
The takeaway
Stop collecting ideas and start applying the filter. The five page types, neighbourhood presence, local-variable answers, interpreted market notes, jurisdiction process guides, and the identity record that verifies them, are the whole strategy, and everything that fails the filter is time returned to your actual practice. The structural half, templates, structured data, and linking that make these pages machine-readable, is what the free rebuild preview shows on your own site; the knowledge half was always yours to write down.
