Situation and closer: how utility writing serves both humans and machines

For decades, weak content worked. Links, domain authority, and keyword-optimized articles elsewhere on the site carried the weight. The writing itself barely mattered as long as the technical signals were right. Sure, we did some meta text writing, some headings, some small pockets of keyword-stuffed texts here and there, where there some pixels left in the design.

But let's be honest, it wasn't exactly something to be proud of.

That is changing quite hard these days. Language machines do not follow links. They read text. And a page full of context-dependent, loosely-structured copy gives them nothing to extract.

And sure, they use indices like Common Crawl to decide what to crawl and what not (tip: read Metehan's research about it). But still, the direction is clear: language machines have removed that escape route, whether you accept it or not. So we need to up our game. You can no longer rank on authority alone while burying the promise in context-dependent prose. The writing has to carry the weight - or at least: waaaaaay more of it - now.

In fact, I'd argue that this should have been the main mantra in SEO for years already, especially since Google started using passage indexing.

See what I did there? With the quote? That's utility-writing in effect.

If you haven't already, now is a good time to read my two preceding articles about language machines, the way they interpret text, what all of them (roughly) have in common, and how to deal with that. A fair warning though: they're long, information-packed, with lots of examples. Take your time, it's worth it, i promise:

Did you read them? Kudos. Let's move on then!

In my utility-writing article (I'm linking here again, you really should have read it by now), I ended with two things I hadn't figured out yet.

Working on a content briefing for a client gave me a concrete answer to both questions I was still left with. Not as a theory, but as something I actually had to write and make decisions about sentence by sentence. What emerged from that process is what this article is about.

When someone arrives at a page, they don't start at the top and read to the bottom. They scan. They're looking for a signal that tells them this page holds what they came for. If that signal appears, they stop and read. If it doesn't appear fast enough, they leave.

This isn't a modern attention-span problem. It's how information seeking has always worked.

Peter Pirolli and Stuart Card formalized it in 1999 under the name information foraging theory¹. The analogy they drew was deliberate: humans seeking information behave like animals foraging for food. They move through an environment, picking up signals that indicate whether a given patch is worth pursuing. When the signal is strong enough, they stay and feed. When it weakens below a threshold, they move on.

The key concept is information scent: the signal a piece of content emits that tells the forager how likely it is to satisfy their current need². The scent is not the content itself. It's the cue that the content is there. A heading that names the situation. A first sentence that describes the problem. A phrase that the reader recognizes as their own.

Nielsen Norman Group's eye tracking research confirms what this looks like in practice³. Across studies with over 500 participants and more than 750 hours of observation, the finding is consistent: people don't read web pages, they scan them. The specific pattern varies with page structure and reader intent; but what stays constant across all patterns is that most of the page is never fixated. What gets read is what the scanner found strong enough to stop at.

The practical consequence for anyone writing web content: if your most important sentence sits in the middle of a page, behind three paragraphs of context, most readers never reach it. Not because the content is bad. Because the scent wasn't strong enough to pull them that far.

So the concept is this basically:

Language machines (LLMs, search engines, AI assistants, and agents built on transformer architectures: systems that retrieve and rank text by scoring chunks against learned semantic representations) do not read a page from top to bottom either. They select. The retrieval process works by scoring chunks of text against the query and extracting the highest-relevance passages. Petrovic's analysis of Google's Gemini grounding system found a total budget of approximately 2,000 words per query, distributed across sources by relevance rank⁴. Of a 6,000-word page, only a fraction gets selected. The rest is invisible.

The mechanism is different from human scanning though. A human scanner follows visual and semantic cues: headings, bold text, first sentences, whitespace. A language machine scores text chunks against a query using semantic similarity. It doesn't see visual hierarchy.

But the practical consequence for the writer is the same: a sentence that doesn't carry its own context is invisible to both readers. The human scanner doesn't pick up the scent. The language machine can't extract the chunk as a standalone, usable proposition.

This is the connection utility writing was building toward but hadn't fully named. The five lenses (position your best material, give the system a reason to pick you over alternatives, make every sentence survive on its own, state relationships explicitly, include a sentence a language machine could directly quote) are all, at their core, scent-building techniques. They make sentences strong enough to be found by either type of reader.

What this article adds is a sequencing layer: not just which sentences need to carry their own scent, but in what order the different types of sentence need to appear, depending on what the reader is hunting for.

Not all readers arrive at a page hunting for the same thing. And that changes which sentence is the one to start with.

Eugene Schwartz described this in 1966⁵. His five stages of customer awareness (from unaware through problem-aware, solution-aware, and product-aware, to most aware) are a map of how much of a defined question the reader has already formulated before they arrive.

His core observation: "In its natural development, every market's awareness passes through several stages. The more aware your market is, the easier the selling job, the less you need to say."⁶

The logic: a reader who hasn't formulated a problem yet can't evaluate a solution. Before they can assess information, they first need to recognize that the information is relevant to them. The sentence that creates that recognition doesn't do less work than the sentence that delivers the specification. It does different work, and it has to come first.

Schwartz described this at the level of campaigns and headlines. What this article adds is the sentence level: which type of sentence needs to open a given block of copy, depending on what question the reader has already formulated⁷.

There is data that supports the underlying principle at scale. Unbounce's 2024 analysis of 57 million conversions across 41,000 landing pages found that email traffic converts at a median of 19.3%, while display advertising traffic converts significantly below that⁸. The pages are often identical. What differs is the reader's awareness stage on arrival. Email readers already know the brand and have a partially formulated question. Display readers are still hunting for the scent. Same page, same copy, conversion rates that differ by a factor of four or five⁹.

Pirolli and Card showed that what counts as a strong enough signal depends entirely on what the forager is looking for. Schwartz showed that what a reader is looking for depends on how much of a question they have already formulated. Kahneman showed that without a defined question, System 2 doesn't activate and specification-dense content passes unprocessed.

The practical consequence of combining these: the sentence that needs to open a block of copy is not fixed. It depends on the reader's state on arrival. A reader without a question needs a situation sentence first. A reader with a fully formed question needs the answer first.

This is my synthesis of the three, and it seems to be useful¹⁰.

I use four distinct reader states, each defined by how much of a question the reader has already formulated on arrival:

Each state calls for a different sentence to open. That is what the rest of this article maps out¹¹.

Working through the briefing, I found myself making the same decision repeatedly at the sentence level. Every block of copy had two layers. One layer did recognition work. One layer did anchor work. And the order between them depended on what question the reader had already formulated.

I've been calling them situation sentences and closers:

Both can carry emotional weight. Both can carry factual content. What distinguishes them is function: the situation sentence emits a signal, the closer is what's found when the signal is followed.

The two are not equal in weight. The closer is the foundation: it is the unit that survives extraction, serves the comparing reader, and gives the language machine what it needs. But it only reaches the browsing or recognizing reader if a situation sentence has first made it worth their while to stop¹². Write the closer first. Then decide how much situation the reader needs to get there.

The distinction maps onto two different writing traditions. Situation sentences are the domain of advertising writers (the kind of sentence Schwartz, Ogilvy, and Bernbach spent careers perfecting). They are psychological before they are informational: they work because they name something the reader already feels. A promise, in other words.

Closers are where that promise is kept. They are the domain of utility writing: no atmosphere required, just a claim that stands completely on its own. And as language machines take over more of the conversation (selecting, extracting, and summarising on behalf of the reader) the closer becomes the unit that survives. The situation sentence draws the reader in. The closer is what gets found.

A situation sentence names a moment the reader is already in. Not a product. Not a benefit. A moment.

The fastest test: can the reader say "that's me" when they read it? If yes, it's a situation sentence. If it requires knowing something about the product first, it isn't.

One practical observation: the most effective situation sentences tend to be loss-framed rather than gain-framed. A gain-framed sentence promises something better: "imagine running without effort." A loss-framed sentence names something already wrong or unresolved: "most shoes are built for tracks." The difference matters because loss-framing activates faster. The reader isn't being offered a future state; they're being recognized in their current one. Prospect theory predicts this: losses register more immediately than equivalent gains¹³.

That said, here are some examples across different industries and page types. All loss-framed:

A closer contains everything needed to understand the claim without reading anything else on the page. It can be extracted by a machine and understood by a reader arriving cold.

The fastest test: if you paste this sentence into a blank document with no context, does it still make sense? If yes, it's a closer. If it relies on what came before, it isn't.

Examples:

What these have in common: named entities, specific numbers, conditions stated inline, no pronouns without antecedents, no vague qualifiers. A human reader arriving cold can evaluate it. A language machine can extract it as a standalone answer.

No "we", "our", "it", "this", "here". Every reference is resolved. Every condition is explicit.

The question isn't which type of sentence is better. It's which type the reader needs first, given what they already know when they land on this page.

The main question: does this reader arrive with a specific problem to solve?

Getting the order wrong is not a subtle problem. The reader either leaves or stalls, and neither shows up in your copy review.

So, here is a table that summarizes it:

Each moment calls for a different pattern on the page. The examples below show what that looks like in practice.

The four-moment model maps onto pages. The same product category, in any given industry, can have pages that serve completely different moments.

For example, a law firm's general practice page catches browsers. Its dismissal page catches recognizers. Its fee structure page catches comparers. Its intake form page catches confirmers. Each needs a different sentence to open.

Below are some more examples.

The reader landed on a specific product but hasn't yet answered "is this for me?" The situation sentence makes the experience of the product visible. Once that recognition has landed, the closer earns its place.

Situation sentence:

Closer:

Without the situation sentence, the closer is accurate but cold. Without the closer, the situation sentence resonates but anchors nothing. The pair works because one earns the right for the other to land.

The reader knows what the product does and is deciding whether to commit. They came for criteria. The closer opens:

Optional situation sentence close:

The closer does the work. The four-word situation sentence removes the last friction without reopening the case. Whether to include it depends on how much anxiety sits underneath the decision. Higher stakes, more useful.

The reader has decided. More copy actively hurts. One closer:

Everything in one sentence. Adding anything after it is noise¹⁴.

The examples above each show a single situation sentence paired with a single closer. But the model also works at page level, across multiple closers. A listing page (a category page, a comparison overview, a legal rights summary) can open with a situation sentence that names the reader's context, follow with a closer that describes the category as a whole, and then use utility headings to group the product closers beneath.

The situation sentence earns attention. The category closer gives the language machine a single extractable claim about what this page covers. The utility headings act as taxonomy markers. And each product closer stands on its own: findable, scorable, and usable without the context above it.

Three examples across different industries:

Situation sentence:

Category closer:

With below it lists with category listings, starting with "daily trainers":

And then "Stability shoes":

Situation sentence

Category closer

Category, per-seat pricing:

Another category, workspace pricing:

Situation sentence:

Category closer:

And then the first category, Financial entitlements:

And the second category, procedural entitlements:

The five utility-writing lenses haven't changed:

What this article adds is a sequencing layer on top of those five. The lenses describe what makes a sentence useful to a machine. The moment-based pattern describes when each type of sentence does its best work for a human, and how to align the two.

In browsing and recognizing: the situation sentence emits the scent that pulls the reader toward the closer. In comparing: the closer opens because the reader has already found the scent and wants the content. In confirming: the closer is the whole thing.

The closer is not optional. It is mandatory. A self-contained, fully-resolved proposition with named entities, inline conditions, and no unresolved references is the unit a language machine can extract without surrounding context. That is not a stylistic preference. It is a structural requirement for content that survives AI retrieval.

Everything else in this article is about getting the human reader to the closer. The situation sentence activates the schema. The sequencing logic calibrates the order. The four moments tell you how much seduction the reader needs before the utility can land. Some readers need none. A confirming reader and a language machine have identical requirements: one closer, nothing else.

Machines are trained on human behaviour. They have learned to find what humans find useful. If you optimize for the machine first, you strip the humanity from the content and lose the human reader. If you optimize for the human first, the machine follows. It always has.

Write the utility first. Build the seduction around it.

Utility-writing is a lever to pull. Something that gets you results. I know we've looked at pagerank sculpting, and speed, and log file analysis, and serverside rendering, and all that stuff. But pretty please with cream and sugar on top: realize that this is way more useful than putting some keywords in a title.

If you ask me, it's a new level that we need to unlock in this optimizing game. Don't ignore it!

The sentence-order hypothesis is consistent with information foraging theory, Schwartz's awareness stages, and dual-process theory, and with what I've observed in content that performs well. But the research I'm drawing on comes from controlled experiments and navigation studies, not from content optimization at the sentence level. Test this in your own context. Break it deliberately. See what the pattern actually is. I'm know I am.

What counts as the dominant question varies by page type and product category in ways that are probably more complex than the examples here suggest. The framework applies; the calibration is always specific to the page. And the situation sentence / closer labels describe function, not content: the same sentence can do situation work in one context and closer work in another.

A gap I haven't addressed: measurement. The article describes what to write and in what order, but gives no signal for how to know if the sequencing worked. I think, assume that conversion rate is too noisy. Dwell time is too blunt. But to be honest: I haven't scientifically validated all this yet. My goal is application, not science. And I'm too busy getting results anyway :)