Content Written to Be Quoted, Not Just Ranked
Most marketing content is invisible to language models — not because it is bad, but because there is nothing in it a model could safely repeat. We write the kind that gets lifted into the answer.
Most marketing content is invisible to language models — not because it is bad, but because there is nothing in it a model could safely repeat. We write the kind that gets lifted into the answer.
Think about what a model is doing when it composes an answer. It is assembling statements it has reason to believe are true, from sources it has reason to trust, in response to a question. Every sentence it produces is a small risk.
Now look at a typical services page. “We deliver best-in-class solutions tailored to your needs.” There is no fact there. Nothing to attribute, nothing to verify, nothing a model could quote without saying something meaningless. So it doesn't.
This is why content volume so rarely moves AI visibility. Publishing forty more pages of the same unquotable prose gives a model forty more things it cannot use. One page containing a specific, sourced, checkable claim about your field is worth more than all of them.
The practical shift is uncomfortable for most marketing teams: fewer, sharper claims, with the evidence attached, written to be extracted rather than to be admired.
These are the rules our content operates under, and they are visible in the output.
01
A quotable passage carries its own support: the assertion and the source sitting together, so a model can lift both without hunting. Split them across a page and the claim travels alone — which usually means it does not travel at all.
Rule: never orphan a claim
02
Lead with the direct answer to the question the page exists for, then supply the reasoning. This is how models extract passages, and it happens to be how impatient humans read. The two audiences want the same structure more often than the industry admits.
Rule: the answer in the first sentence
03
“Faster response times” is unusable. “Median first response of 47 minutes across 1,200 tickets in Q2” is quotable, checkable, and hard for a competitor to imitate honestly. Specific claims are also far more defensible, which matters when a model is deciding whether to trust you.
Rule: numbers, dates, and named sources
04
Brand voice tends to imply what a business does rather than state it. Models do not infer reliably. Naming your services, markets, and credentials in plain language costs a little elegance and buys a great deal of comprehension.
Rule: state it, don't imply it
05
If a page contains nothing that could not be found in the ten pages already answering that question, there is no reason for any system to prefer it. Original data, first-hand experience, and a genuine position are the only durable differentiators left.
Rule: bring something new or don't publish
06
Clear hierarchy, self-contained sections, and no critical fact stranded in a table caption or a graphic. A passage should still make sense when it is pulled out of the page and dropped into an answer alongside three competitors.
Rule: every section stands alone
The questions that come up when teams dig into this.
Either. Some clients want us producing the work, some have capable writers who need the framework and an editorial standard to hold to, and some want us to audit and rewrite what already exists.
The rewrite path is often the highest-return option and the one clients least expect. Most companies already have the expertise on the page — it is buried under vague phrasing and missing sources.
No. We use AI in the research and drafting process the way any competent team does now, but the substance comes from your expertise and real sourcing, and a human writes and edits the result.
There is a practical reason beyond principle: unedited AI output is exactly the undifferentiated, unsourced prose that fails the quotability test. Using a model to mass-produce content that no model will cite is a self-defeating strategy.
Less than you have been told, done considerably better. A dozen genuinely authoritative pages that own your core questions will outperform a hundred thin ones, in both traditional search and AI citation.
We would rather publish two pieces a month that get quoted than eight that get indexed and ignored.
Usually worth auditing before writing anything new. A page that already ranks and gets traffic but never gets cited is normally a sourcing and structure problem, not a subject problem — and fixing it is far cheaper than starting over.
We typically find a handful of pages carrying real expertise that simply need claims sourced, answers moved to the top, and facts stated explicitly.
Citation rate for the questions each piece targets, tracked across models over time — plus the traditional metrics, which we keep reporting because organic search still pays.
Content is the slowest of the levers. Entity and technical fixes can show up within weeks; content-driven authority is a three-to-six month horizon, and we set that expectation before anyone signs.
The audit shows which sources the models cite in your category — and whether any of them are yours.