An SEO content agency decides which questions a business intends to own, builds the pages that win them, and removes the ones dragging the rest of the site down. Writing is the visible part of that work and the least difficult part of it.

Four responsibilities sit under the label.

Demand and intent research. Establishing what your buyers actually type, what they mean by it, and which of those searches are worth money to you. Volume is the easiest thing to measure and the least reliable guide to value.

Architecture. Deciding how many pages a subject needs, which page owns what, and how they connect. Most sites do not have a shortage of content. They have twelve pages arguing over one subject and none of them winning it.

Production with real input. Getting something onto the page that a competitor could not have written, which almost always means extracting knowledge from people inside your company instead of paraphrasing whatever currently ranks.

Maintenance. Reviewing, updating, merging and retiring what already exists, on a schedule, because published pages lose ground quietly and nobody notices until a quarter of the revenue has moved.

Brand Vision has been building these programs since 2018, and across 500 or more projects one pattern keeps repeating. Companies buy volume when the thing they were missing was sequence. Leads across past engagements have risen an average of 250%, a figure that needs its limits stated in the same breath. It looks backward across programs of very different sizes and starting conditions, several of them running longer than a year, and it is not a number to build a budget against. The checkable one is our Clutch rating, 5.0 from 64 or more client interviews conducted by somebody other than us.

One thing we say early, and it loses us work. Content is the slowest lever in search. If a site cannot be crawled and read properly, that gets fixed first, which is technical SEO. If nobody credible references you, the ceiling on anything you publish stays low, and raising it is digital PR. Content sits between those two and pays best when both are moving, which is why we scope it as one track inside the wider search program instead of selling an article quota.

Eight, run as one program, because most of them are close to worthless in isolation. A calendar with no research behind it is a spreadsheet, and a research deck nobody publishes from is a cost.

1. Keyword and intent research. What your market searches, in what words, at what point in a buying decision, and which of those searches a business your size can realistically win inside a year.
2. Content gap and competitor analysis. A page-by-page read against the companies already ranking, looking for subjects they cover badly, subjects they avoid, and pages old enough to beat. This is also where we find the terms you already rank for by accident, which are usually the cheapest available wins.
3. Topic cluster and pillar architecture. How many pages a subject needs, what each one owns, and how they link, decided before anything is commissioned.
4. Editorial planning and calendars. A sequenced plan with owners, dates, formats and a brief per page, built around your review capacity instead of ours, since client review is the usual bottleneck.
5. Content production and expert-led writing. Pages written from recorded interviews with your practitioners and published under a named author with a real background.
6. On-page optimization. Titles, headings, entity coverage, structured data, internal links and formatting that lets a passage be lifted cleanly. Applied to new pages and retrofitted to existing ones, where the faster return usually sits.
7. Content refresh and consolidation. Updating, restructuring, merging and retiring what you already have, on a review cycle with dates attached to it.
8. Content performance measurement. Reporting against leads, pipeline and revenue page by page, with positions and citation appearance treated as leading indicators and never as the headline.

Where a program starts depends on the state of what exists. A site with three hundred old posts starts with an inventory. A six-month-old site starts with architecture. A site in a category nobody has covered properly starts with original research, because that is the one thing competitors cannot answer quickly.

If you want the full picture before committing, an SEO audit covers content alongside technical condition and authority. Where being cited inside generated answers is the priority, that discipline is AI search optimization and it is scoped beside this work. On-page changes and structured data usually need a deploy, so we work directly with whoever owns the templates instead of filing requests and waiting.

Search intent is what a person wants to happen immediately after they press enter, and matching it decides more outcomes than beating anybody on volume does. A page aimed at the wrong intent can reach position two and earn nothing, because the click bounces back to the results in seconds.

Four shapes, and a keyword tool will rarely identify which of them sits in front of you.

Informational. Somebody wants to understand something. What is this, how does it work, why does it happen. These pages build audience and trust, convert slowly, and are the most exposed to being answered inside an AI interface before anybody clicks.

Navigational. Somebody wants a specific destination. A brand, a login, a document. You win your own and you do not win anybody else's, and chasing a competitor's brand term organically is almost always wasted effort.

Commercial investigation. Somebody is comparing before they commit. Best, versus, alternatives, reviews, pricing, agency, services, near me. This is the most commercially valuable intent on most sites and the most under-built, because covering it properly requires taking positions about competitors that leadership finds uncomfortable.

Transactional. Somebody is ready to act. Buy, book, hire, quote, demo. Short queries, heavy competition, and a product or service page instead of an article.

Here is how to check instead of guess. Open the results for the term and read what is already there. If the first page holds nine definitional articles, the search engine has told you a service page will not rank there however good it is. If it holds nine vendor pages, an article will not either. That one check kills more bad briefs than any tool subscription.

Volume deserves a fair hearing, because a large number does prove the demand exists. What it hides is who sits behind it. Eight thousand monthly searches on a broad category term can be mostly students, job seekers and competitors, while three hundred searches carrying your service and your city are people with a budget and a deadline. In technology and SaaS that gap is extreme, since the head term collects developers who will never buy anything. Where the high-intent terms are genuinely out of reach this year, buying them through paid media while the organic work matures is usually the correct move, and it is better settled in month one than discovered in month five.

Content is priced as a monthly program because position compounds and a single burst of publishing decays, and the number is set by depth per page far more than by count of pages. Depth is the expensive part. Four pages built from practitioner interviews and original data out-earn a dozen assembled from a keyword and an open browser tab, per dollar, over any window past a quarter.

How the first six months usually run.

Weeks one to three. Research and decisions. Demand and intent work, competitor and gap analysis, and an inventory of what is already published with what each page earns now. What comes out is a decision list and not a topic list.

Weeks three to six. Architecture and first briefs. Clusters defined, pillars chosen, overlap mapped, interviews booked, briefs written. First publishing normally lands in week four or five.

Months two to four. Cadence. New pages ship at whatever rate your review capacity supports, refresh work runs in parallel on pages already close to ranking, and linking changes go out with every release.

Months four to six. Reading results. Early winners get expanded, pages that missed get diagnosed instead of quietly dropped, and the next two quarters are planned from what happened.

On timing, plainly. New pages on a domain with existing trust can move within weeks, and on a young domain they will not. Meaningful gains land between month four and month twelve, and any firm offering first-page results inside thirty days is either misleading the client or chasing terms nobody types.

What moves the number.

• How much already exists. Three hundred legacy posts need an inventory and a consolidation plan before anything new makes sense, which front-loads both cost and return.
• Competition in your terms. The bar is whoever holds the position now, which in some categories is a well-resourced in-house team.
• Whether original research is in scope. Surveys and data studies carry real production cost and produce the most durable asset in the program.
• How much you keep in-house. Programs where we direct and your team writes cost less and move slower. Both work.
• Review speed on your side. This sets the pace more than anything we control.

Ongoing programs carry a six-month minimum for a reason worth stating. A program abandoned in month four has absorbed the whole build cost and collected none of the compounding that pays for it. You own the pages, the accounts and the data throughout. Before committing, a paid consultation will say whether content is your best next spend, enterprise SEO is scoped differently at estate scale, and you can tell us what you already have for a straight answer.

Some of it, and the part it takes is mostly the part that was never earning anything. Pages built to define a term in four hundred words are being answered inside the interface. Pages carrying a number, a judgment or an experience that exists nowhere else are being cited, and cited pages get clicked by people much closer to buying.

What genuinely changes. Extraction happens below the level of the page. An assistant lifts a passage, so the unit being optimized is the section and not the document. In practice that means a heading written as the actual question somebody asks, an answer delivered in the two sentences underneath it, and the qualifications and evidence arriving after. Most corporate writing does the reverse, warming up for three paragraphs before committing to anything, which reads acceptably to a skimming human and gives a machine nothing liftable.

Attribution follows the origin of a claim and not the best summary of it. A page restating what four other pages already say gives a model no reason to name it. A page that is where a number came from leaves naming you as the only honest way to use it.

Gating breaks the entire mechanism. A gated asset cannot rank and cannot be cited, because neither a crawler nor an assistant fills in your form. Putting your strongest evidence behind an email field hides it from the two systems deciding whether anybody hears about you. The version that works is publishing the substance in full and gating the operational artifact, meaning the template, the calculator or the raw dataset, the thing somebody wants to use instead of read.

Access matters more than it used to. More than one of the large AI crawlers will download a site's JavaScript and never run it, so anything assembled in the browser is simply absent from what they collect. That is a rendering and crawl access problem before it is a writing one.

What does not change. Clarity, evidence and third-party corroboration still decide it, which is why independent coverage has become a content input and no longer only an authority one.

On measurement we state the limits out loud. Nobody holds position four inside a generated answer. What is measurable is appearance frequency across a fixed prompt set and the identity of whatever sources sit beside you. We map how the assistants handle your category first, and we will not forecast a share we have never observed.

Four situations make a content program the wrong purchase, and you will hear all four from us in the first conversation instead of in month five. Content is the slowest and most expensive lever in search, which makes it the worst one to pull while something upstream is broken.

1. The site is technically broken. A migration nobody finished, crawl traps generating thousands of near-identical URLs, pages that only appear once a script runs, or response times that make a crawler give up. Commissioning writing into that is paying for pages nobody will read. A full audit of the site establishes which of those are real, and the repair comes first.
2. The positioning is unresolved. If two people on your leadership team answer who this is for and why you differently, no writer resolves it on their behalf. What comes back is competent and says nothing, and that is the road to forty pages that make no argument. That is brand research and strategy work, and it costs less than a year of hedged writing.
3. Traffic is already adequate and the site does not convert. Sending more readers into a page that leaks multiplies the loss. Fixing the path is the better spend, and where the template is what limits you, the honest answer becomes a redesign.
4. The arithmetic does not close. Some categories genuinely have a few hundred relevant searches a month nationally, and some products are bought entirely through channels with no relationship to search. Multiply realistic volume by a realistic click share by your conversion rate and your close rate. Where that number lands under the cost of the program, we show you the calculation and recommend against it.

A fifth case is worth naming, which is urgency. If you need pipeline inside this quarter, content will not produce it, and pretending otherwise is how agencies lose accounts in month six.

What a smaller version looks like, since the answer is rarely nothing at all. A fixed-scope research and architecture piece with no production attached, so you own a plan and execute it however suits you. Or a refresh sprint on the twenty pages already sitting on page two, which is a few weeks of work with no retainer behind it. Both leave you better off than a quota you did not need, and both are easier for us to sell honestly.

Because a page already sitting at position eleven has proven a search engine can find it, read it and trust it, and moving it to position four is a smaller job than earning position eleven from nothing. On most sites we take over, the fastest returns available are already published and nobody has looked at them in two years.

The reason they need looking at is decay, and decay is quiet. A page loses ground for four reasons and none of them shows up as an error in any tool. The subject moved and the page did not. A competitor published something better. Newer pages diluted the internal links that used to point at it. Or the query changed shape, and what people now mean by those words is not what the page answers.

The intervention scales, and choosing the lightest one that works is the whole discipline.

Lightest. Update in place. Correct the facts, replace dated examples, add the two or three subheadings every current result covers and yours does not, and tighten the opening so the answer arrives first. Days of work per page.

Middle. Restructure on the same address. Same URL, reorganized around what the query now wants, preamble deleted, strongest material moved up. Keeps every link and every year of history.

Heavier. Consolidate. Three or four thin overlapping pages merged into one strong one, the URL holding the most equity kept, the others redirected into it permanently so their links and history transfer instead of evaporating.

Heaviest. Retire. Some pages should be removed with nothing salvaged. A page earning nothing still spends crawl attention and blurs the topical signal of the pages around it, so pruning is a gain and not a sacrifice.

Then the cadence, which is the part almost nobody puts in a plan. Every page gets a review date the day it publishes. Commercially important pages get looked at twice a year. Anything referencing prices, regulations or platform behaviour gets looked at quarterly. The long tail gets an annual pass. That schedule belongs in search work as a whole from month one instead of being discovered in year three.

The honest ceiling on all of it. A page aimed at an intent it can never serve does not improve with editing, and no amount of refreshing rescues a page that should not exist. Decay also turns catastrophic during replatforming when URL mapping is done carelessly, which is the most common way a company deletes a decade of content equity by accident.

A cluster is one broad page on a subject with a ring of narrower pages around it, each handling a single component, wired together so a reader and a machine can both see how they relate. The alternative, which is what most sites have, is a pile of unrelated articles that support nothing and compete with each other.

The pillar. One page covering the whole subject at a level a newcomer can follow, written to be the best available overview and nothing narrower. It carries the broad head term, it is usually the page that attracts external links, and it exists to be the place everything else points at.

The cluster pages. Each one owns exactly one sub-question and answers it in more depth than the pillar could afford. One page, one intent, one primary query. The discipline sits in the boundary drawing, because the moment two cluster pages could answer the same query you have manufactured a problem for yourself.

The links. Every cluster page links up to the pillar with descriptive anchor text naming the subject. The pillar links down to each cluster page. Cluster pages link sideways only where a reader genuinely needs the other page. Links live inside body copy where they are earned, and not in a related-posts strip that a crawler discounts and a reader ignores.

Three things that structure buys you. Authority collects on the pillar instead of spreading across twenty orphans. Crawlers get an explicit map of what you claim to cover and how the parts relate to each other. And the architecture forces scope decisions up front, which is the only reliable prevention for your own pages fighting.

How large a cluster should be is decided by how the market splits the subject and never by a target count. In education marketing a program page behaves as a pillar, with admissions, funding, outcomes and delivery format pages beneath it. In legal services each practice area is a pillar and the cluster is process, cost, jurisdiction and eligibility. Identical structure, completely different shape.

One limit worth naming. Hand-placed internal linking works up to a few hundred pages and then stops working. Past that it has to be generated by rules built into templates and components, which is why linking across large site estates is an engineering problem before it is an editorial one.

Cannibalization is several of your own pages competing for one query, so a search engine cannot tell which page you meant and ranks none of them as well as one would have ranked alone. It is the most common self-inflicted content problem we find, and it is nearly always a by-product of publishing without an architecture.

How it shows up in the data.

• Positions flipping between two of your URLs week to week for the same term
• A page that peaked and then declined in the month a similar page went live
• Impressions and clicks for one query spread across four URLs where one used to hold them
• Internal links about the subject split across several destinations, so no page accumulates the signal
• A sales team that cannot say which page they would send a prospect to

Diagnosis takes an export of queries by landing page and a grouping step. Anywhere more than one URL earns clicks for the same query you have a candidate. Candidate is the operative word, because a second symptom should be present before anybody starts merging things.

The fix, in order.

1. Pick the winner on merits. Existing position, external links, conversion history and fit to the intent. Sometimes the older, weaker-looking page wins because it holds the links.
2. Move the unique substance across. Everything the losing pages say that the winner does not, folded in properly instead of appended at the bottom.
3. Redirect the losers permanently, so their equity and history transfer to the winner instead of dying with the URL.
4. Repoint internal links at the winner, including navigation and any campaign destinations pointing at the old address.
5. Fix the cause, which is the brief process that allowed a second page to target an intent already covered. Skip this step and you repeat the whole exercise in eighteen months.

Where it bites hardest is scale. On a large product catalogue filter and category overlap generates it structurally, and for multi-location businesses the city pages compete with one another, a failure mode well known in local SEO.

The counter-argument deserves room. Not every overlap is cannibalization. A definitional page and a comparison page on one subject serve genuinely different intents and are supposed to coexist, and merging them because a tool flagged an overlap is a mistake we see made in the other direction just as often.

Senior writers who are not experts in your field, working from recorded interviews with the people in your company who are. Anybody claiming their writers already understand your industry the way you do is overselling, or about to hand your account to a generalist with a browser open.

How a brief gets built when the writer is not the expert.

1. Interview the practitioner. Thirty to forty-five minutes with the person who actually does the work, and not with the marketing contact. What clients consistently get wrong. What you refuse to do and why. The number you find yourself quoting every week. The objection that arrives on every call. Recorded, transcribed, and mined for the three things nobody outside your company could have said.
2. Write a brief that carries an argument. The question the page answers, the intent behind it, the position the page takes, the proof for that position, the internal links in and out, and what a reader should be able to do afterwards. A brief consisting of a keyword and a word count produces exactly what you would expect.
3. Draft against the brief, preserving the practitioner's own phrasing wherever it beats ours, which is frequently.
4. Send it to the practitioner for review, and not to marketing. Marketing review catches tone. Practitioner review catches the wrong claim, and in regulated categories that distinction is the entire risk. In health and wellness work, clinical review is a gate and not a courtesy.
5. Publish under a named author with a real biography, real credentials and a page saying what that person has actually done. A reader weighs a claim by who made it, and machine systems increasingly resolve trust through people and organizations they can identify by name.

The asset that outlasts everything else is original research. A survey of your own customers, an analysis of data only you hold, a benchmark drawn from your own operations. Field convention holds that around four hundred responses puts you near five points of margin of error, which is enough to publish a figure and defend it when somebody asks. That kind of page earns references because no second source for the number exists, and it keeps earning long after an opinion piece has stopped.

You can look at the senior people who do the work before committing to anything. Subject-matter background is a real hiring criterion here, which is visible in the kind of people we hire.

Long enough to answer the question completely and no longer, which makes word count a symptom of thoroughness and never a target to hit. A brief specifying eighteen hundred words produces eighteen hundred words, several hundred of which exist only to reach eighteen hundred words.

The myth persists because long pages do correlate with ranking, and that correlation is real. Covering a subject thoroughly takes room, so pages that cover subjects thoroughly tend to be long. The error is reading the correlation backwards and treating length as the cause. Padding a four-hundred-word answer out to two thousand adds no coverage. It adds distance between a reader and the thing they came for, which shows up as people leaving early and, now, as a page a model cannot lift a clean passage from.

Five tests more useful than a length target.

• Does the page answer its own question inside the first hundred words?
• Does it cover every sub-question the current results cover, plus at least one they all miss?
• Could any paragraph be deleted without losing information?
• Would somebody who already knows the subject learn something?
• Can each section stand on its own as an answer when read out of context?

On-page work is where that gets operationalized, and it involves more than a keyword in a title. The title and H1 written as the query a person types. Headings written as questions where the material underneath is an answer. Entity coverage, meaning the related people, products, standards and places a thorough treatment would mention, since their absence reads as shallowness to a machine assessing depth. Structured data, so a system is told what the page is instead of inferring it. Internal links with descriptive anchors. Formatting that supports extraction, meaning short paragraphs, real lists, and definitions placed where somebody would look for them. Most of that ships through templates, which is why it moves faster when we work with the development side directly.

Two honest caveats. Some intents want a short page, and forcing depth onto them loses. A transactional query wants a product page and a price, not an essay. And reading behaviour differs enormously by audience. B2B buying committees will read four thousand words when the decision is expensive enough, while a consumer on a phone will not read four hundred. Where that answer genuinely matters to the format, UX research on your actual readers settles it faster than any published benchmark.

Content is judged on leads, pipeline and revenue page by page, with positions and citations treated as leading indicators that explain the commercial number instead of standing in for it. A report opening with impressions or with total sessions is avoiding a harder conversation.

What appears in the reporting, in the order it matters.

• Leads and revenue attributed to specific pages, tied to conversion tracking we verified before publishing began, since numbers drawn from measurement nobody validated are more dangerous than no numbers.
• Assisted influence, meaning pages appearing early in converting journeys without ever being the last click. Most educational content earns its place here and gets killed by teams measuring last click only.
• Non-branded organic entrances by page and by cluster, held apart from brand-name traffic, because arrivals on your own name track everything else you do while arrivals on problem language track this work specifically.
• Position movement across the tracked keyword set, with the terms we deliberately left alone listed beside the reason for leaving them.
• Citation and mention frequency across a fixed prompt set in the assistants your buyers actually use, reported as share and trend because there is no rank available to report.
• Decay watch, meaning pages losing position or clicks quarter over quarter, which is the queue refresh work gets pulled from.
• Production throughput against plan, including where a delay sat, which is often on the review side and should be visible instead of absorbed.

Two structural points about how this gets read. A page needs a fair window before anybody judges it, which on most domains is three to six months, so killing a page at week eight tells you only that eight weeks was too short. And judgment happens at cluster level as well as page level, because a supporting page whose job is funnelling authority into a pillar can look like a failure on its own numbers while doing precisely what it was built to do.

Every page carries a stated purpose from its brief, so measurement compares the result against the intended job instead of against a generic benchmark. Where a page missed, the report names which assumption turned out wrong. When a quarter disappoints, the explanation reaches you from us first, with the change we are making attached to it.

Content reporting sits in the same monthly view as everything else, since search rarely converts in isolation and cross-channel attribution is where content gets undervalued. That consolidation is what the broader marketing plan exists to provide. You can see how it has read on past engagements, and the team producing it sits in our Toronto office beside the people who wrote the pages.