Search intent is a business question, not a keyword attribute

Every keyword tool will hand you an intent label. Informational. Navigational. Commercial. Transactional. Four buckets, applied at scale, delivered in a column next to the volume figure.

The labels are not wrong. They are just too coarse to build anything on. Two queries carrying the same label can sit six months apart in a buying decision, need completely different pages, and be worth wildly different amounts to your business. A taxonomy that puts them in the same bucket is not telling you what to do next.

The more useful question is not what kind of search this is. It is: what has this person already decided, and what are they trying to decide now?

The same label, two very different searches

Take two queries that any tool will mark as commercial investigation.

“project management software” and “asana vs monday for a 12 person agency”.

The first is someone at the beginning. They may not have a budget. They may not have agreement internally that a tool is needed at all. They are building a mental map of the category — what these things are called, what they cost, who the players are. If you rank first for that query and send them to a pricing page, you have interrupted a person who came to look around and asked them to buy. They will leave.

The second is someone at the end. They have narrowed to two options. They have a team size, which means they have a budget range and probably an approval process. They are looking for the one piece of information that will let them stop deliberating. There is a decision waiting on the other side of that search, and whoever answers it best is very likely to be in the room when it gets made.

Same intent label. Same category. Radically different pages, different content, different conversion path, and different value per visit — often by an order of magnitude.

The second query has a fraction of the search volume of the first. It is worth considerably more.

Figure 1

Search volume Value per visitor

1. project management software Building a mental map of a category. May have no budget and no internal agreement that a tool is needed.
2. project management software for construction Has ruled out the generalists. Knows the sector constraint that matters.
3. asana vs monday for a 12 person agency Down to two. Has a team size, so a budget range and an approval process.

Distance to decision

The property that actually matters is what I think of as distance to decision: how many things does this person still have to work out before they can buy from anyone?

It is a rough scale, and rough is fine. What it gives you is a way to sort queries that has some connection to money.

Figure 2

Far from a decision At the decision

1. 01 Problem-aware Working out whether the problem is worth solving at all. Page's job: name the problem better than they can. Earn the right to be believed later.
2. 02 Solution-aware Working out what the options are, and how to judge between them. Page's job: give them the criteria. Be the source that taught them how to choose.
3. 03 Vendor-aware Down to a shortlist. Working out which one, and whether it is you. Page's job: answer the specific objection. Remove the last unknown.
4. 04 Ready Nothing left to work out. They are looking for the button. Page's job: do not get in the way.

Written out as a table, so you can put it next to your own query list:

Most content programmes are heavily weighted to the “far” column, because that is where the volume is and where the keyword tool points.

That figure is weaker than it looks. On Ahrefs’ own account, search volume is an annual average, with similar phrases grouped into one bucket and reported as a single number rather than counted per exact phrase — which is the subject of why keyword research keeps producing the wrong content, and much the same problem as the one behind most circulating local SEO statistics.

Most revenue sits in the “near” column, where the volume looks disappointing and the page feels almost too specific to be worth writing.

The pages that feel too specific to be worth writing are usually the ones worth writing first.

If your sales team answers a question on every single call, and no page on your site answers it, that question is a near-decision query with a known conversion rate. You do not need a tool to validate it. You need to write it down.

A buying decision, followed for four months

The framework above is abstract until it is applied to one person, in order, because sequence is the only way to see how far apart two searches with the same coarse label can actually sit.

Take a homeowner whose combi boiler is nine years old. In early March, nobody at that address is planning to buy anything from a plumber. By late May, someone does. Here is a realistic run of what they might actually type in between, hypothetical but ordinary — the kind of sequence a search console export would show if anyone thought to look at one person’s queries in order rather than one query at a time.

Two things about that sequence do not survive contact with a coarse intent label.

The first is the pair a tool would treat identically: “why does my boiler pressure keep dropping” in March and “boiler losing pressure again after repressurising” in April. Both read as how-to searches. Neither contains a commercial word. A keyword tool tags both informational and stops there.

They were typed by the same person six weeks apart, and the second one arrives after a failed attempt at the first one’s advice — a signal no tool captures, and exactly the evidence that turns a symptom into a suspected fault. The word “again” moved that query considerably closer to a decision than its label admits.

The second is the brand-name query in May. It looks informational — a fact-finding search about a named product’s common faults — and it is very likely the single nearest-to-decision query in the whole sequence, because it almost certainly follows an engineer’s visit and a quote, and exists to verify that quote before money changes hands.

A business writing for “boiler pressure” head terms while treating brand-plus-fault queries as low-priority informational content is aiming at the wrong end of its own funnel.

Reading the results as evidence

You cannot deduce intent from the words in a query. You can observe it, and the observation is free.

Search the term and read the page Google has already built. What ranks is Google’s current, continuously-updated model of what people want when they search that phrase — Google’s own account of how it ranks results starts from the same premise, that before anything else its systems have to “establish what you’re looking for — the intent behind your query”. It is drawn from more behavioural data than you will ever have. Argue with it and you will lose.

Read four things.

The format. Are the top results guides, comparisons, product pages, tools, forum threads, videos? If nine of the ten results are third-party listicles, a vendor product page is not going to rank there. That is not a content-quality problem you can write your way out of; it is Google telling you what kind of thing belongs in that slot.

The publisher. Vendors, independent review sites, communities, news? If Reddit and a couple of independent blogs occupy the top of the SERP, the query has a trust profile that commercial pages struggle to meet. Sometimes the right answer is to earn a place in those results rather than to publish your own page.

The SERP features. A shopping carousel means transactional. A local pack means the query has a geographic component whatever the words say. An AI Overview or featured snippet means there is a short factual answer that your page needs to give cleanly and early if you want to be the source of it.

The gap. What does every ranking page fail to do? Nearly always there is one — an assumption they all share, a step they all skip, a reader they all ignore. That gap is the only reason for your page to exist. If you cannot find one, the honest conclusion is that this query is already well served and your effort belongs somewhere else.

Put together, the four readings give you a verdict before anyone writes a word:

That last row is the interesting one. A SERP that cannot make up its mind is a SERP where a genuinely well-aimed page can settle the question, and those are the easiest wins available to a site with any authority at all.

That last point is the one most content processes have no way of reaching. A brief that starts from a keyword can only ever produce another version of what already ranks.

What the SERP doesn’t tell you

The four readings above work when the results page disagrees with itself, or when the dominant format gives the intent away outright. They are less useful for the harder case: a SERP made entirely of one plausible-looking format, for a query where you genuinely cannot tell whether the person is at the beginning or near the end.

Google’s own guidance to the people it employs to judge its results is a reasonable place to start, because it says plainly that a single query rarely has one meaning. Its public Search Quality Rater Guidelines overview uses “mercury” as its own example — a searcher might mean the planet or the element — and instructs raters to judge a result against “the dominant, common, or reasonable minor” interpretation of a query, not a single assumed one.

If the organisation building the ranking system treats a query as an interpretation problem rather than a lookup, a content team should too.

Three things worth checking when the format alone will not settle it:

The exact words, not the paraphrase. “Again”, “still”, “won’t” and “keeps” are a distance signal a keyword tool discards the moment it groups a query into a topic. A query with “again” in it has usually already tried something and failed, which sits closer to a decision than the same head phrase without it — whatever intent label both carry.

Modifiers that imply a completed step. A brand name, a specific model, a quoted price or a named competitor inside a query usually means something has already happened — a quote was given, a shortlist was made, a product was chosen — that the plain head term does not carry. These queries are routinely mislabelled informational or navigational by a tool, because they contain a proper noun rather than a commercial verb.

What sits next to it in the data. Related searches, “people also ask” entries and, where you have it, your own site-search or next-click analytics are the cheapest evidence available. A query that is frequently followed, in the same session, by a comparison or pricing search is nearer a decision than its own wording suggests; a query that dead-ends is probably genuinely early.

None of this replaces judgement about a person mid-decision. It replaces guessing at one.

Where intent and margin meet

Here is the part that makes this a business question rather than an SEO one.

Distance to decision tells you how close someone is to buying. It does not tell you whether you want them to buy. Those are different questions, and the second one is answered by the business, not by search.

Cross the two.

Figure 3

High margin

Build second Worth building once the commercial pages exist and rank. Where the compounding happens — and a poor place to start.

Build first In order, before anything else. This is where organic search pays for itself and where you get a number to point at.

The default internet The largest category of published content there is, and the reason so many blogs have a rising traffic chart and flat revenue.

Worth having Worth not over-investing in. A competent page, not a flagship.

Low margin

Far from a decision Near a decision

Most keyword strategies never make this cross-reference, because the person doing the research does not know the margins and the person who knows the margins is never asked. That conversation takes an hour. It changes the shape of the entire programme.

What this changes in practice

Four things, concretely.

One page, one decision. Every page should be built for someone at a specific point in the process, and it should be obvious from the first paragraph who that is. A page that hedges — half explanation, half sales pitch — usually satisfies neither reader and ranks for neither query. Google’s own guidance on creating helpful, people-first content asks the same thing from the other direction: whether someone will “leave feeling they’ve learned enough about a topic to help achieve their goal”.

Sequence commercially, not alphabetically. Publish near-decision pages first, even though they have less volume and feel less impressive on a content calendar. They rank faster, because there is usually less competition, and they produce revenue you can point at when someone asks whether this is working.

Judge pages on the right measure. A far-from-decision page that produces no direct enquiries is not failing; that is not its job. Measure it on assisted conversions, on returning visitors, on whether people who read it come back. A near-decision page that produces traffic and no enquiries is failing, and no amount of traffic redeems it.

Re-read the SERP periodically. Intent is not a fixed property. Categories mature, language shifts, and Google reinterprets queries. A page that stops ranking without an obvious technical cause has often been overtaken by a change in what the query means rather than by a better page.

Intent decides the page type, not just the angle

Most of the advice above stops at content angle — write for a beginner, write for someone comparing, write for someone ready to buy. That is necessary and not sufficient. Distance to decision should decide the type of page before anyone thinks about the angle of the prose on it, and the two decisions get conflated more often than they should.

The tool row is worth pulling out on its own, because it is the format most content programmes never build, and the one that fits the middle of a decision best. Someone weighing “boiler repair vs new boiler” is not well served by an article listing the general pros and cons of each path — a properly built brief can sharpen that article considerably, but it cannot fix a format mismatch the page type itself creates.

They are better served by something that takes their specific inputs — the age of the unit, roughly what a quote came in at — and returns a specific answer. A local SEO ROI calculator is this site’s own version of that format, built because the question it answers is a middle-of-decision one that no article resolves as cleanly.

That is a different production to a written article, and it is worth building deliberately rather than defaulting to prose, because a tool answers a solution-aware question the way the searcher is actually trying to answer it: by doing the comparison, not by reading about how to do one.

The near-decision row is where a mismatch is most expensive, because the traffic still arrives — the query still gets searched, the page still gets the click — and it fails anyway, quietly, because the format was wrong rather than the content.

A blog post ranking for a near-decision query and converting at close to nothing is not an SEO failure. It is a page-type failure wearing an SEO failure’s symptoms, and it usually needs the conversion fixes a local website guide covers rather than another round of optimisation.

One business question, five keywords, one page

There is a second failure sitting underneath one this site has already written about at length: why keyword research keeps producing the wrong content is mostly the story of building one page per keyword when several keywords actually describe a single decision. Grouping by decision instead of by vocabulary fixes most of that problem.

It does not fix all of it, because a single decision can still contain several distances, and treating “same decision” as though it means “same distance” is the subtler version of the same mistake.

Look back at the boiler sequence above. “Why does my boiler pressure keep dropping”, “boiler losing pressure again after repressurising” and “boiler repair vs new boiler cost 9 year old combi” arguably belong to one decision — should this boiler be repaired, and by whom — the way decision-based clustering would group them. A tool grouping on shared vocabulary would put all three on one cluster map without hesitating.

They are still three different distances from a decision, produced by the same person over roughly seven weeks, and a single page trying to serve all three at once will do what a hedged page always does: satisfy the reader at none of the three points fully, because the opening paragraph has to guess which of the three people showed up.

The practical fix is the same discipline this site applies to briefs generally, applied one level below clustering: once queries are grouped by decision, check whether the cluster itself spans more than one distance, and split again where it does.

A decision cluster that runs from “is this worth fixing at all” to “verify the quote I was just given” is not one page with several headings. It is at minimum two — an explainer for the earlier distance, and an objection- or verification-focused page for the later one — even where a clustering tool, and even a careful human doing the grouping by hand, would reasonably see one topic.

The same question, three surfaces

Everything above has assumed the query lands somewhere close to a conventional results page. Increasingly it does not, and where it lands changes what the page needs to do — sometimes more than the distance-to-decision judgement itself does.

BrightLocal’s 2026 survey of US consumers, collated in this site’s own local search behaviour data, found the same underlying local search now starts across a genuinely mixed set of channels: 52% on Google Search directly, 9% on Google Maps, 9% on social media and 8% on an AI tool, with the remainder split across review sites, Apple Maps and voice search — none large enough individually to report at that sample size.

Three in four consumers used more than one channel during their most recent search. The surface a question first arrives on is not a fixed property of the question; it is closer to a habit than a decision.

Two findings from that same research matter more than the raw split, because they show the surfaces do not behave the same way once a question lands on them.

An AI-assistant start is rarely the whole journey. Only 2% of people who began their search on an AI tool used it as their only channel — 43% went on to check Google, and 39% checked social media, before deciding. A page built only to be quoted well inside an AI answer, with no thought for what that same person searches next, is optimising for a third of a journey that continues somewhere else two times out of three.

Google’s own AI Overviews intercept the far end of a decision far more than the near end. Whitespark’s case study across 540 queries in three US cities and six trades — plumbers among them, and recorded in full here — found AI Overviews triggered for 92% of purely informational queries in a trade, against 15% for narrowly local, transactional ones like “personal injury lawyers in Phoenix”.

That gap maps almost exactly onto the distance-to-decision scale this piece has argued for throughout: the far end of a decision is where an AI answer is now most likely to intercept the click before it reaches a business’s own page; the near end, where the booking actually happens, still returns something close to a conventional results page most of the time.

Read against the boiler sequence earlier, the implication is specific rather than general. “Why does my boiler pressure keep dropping” — the far-end query, typed in March — sits in the intent category this dataset says gets answered inside an AI box roughly nine times in ten. “Emergency plumber [town] no heating” — the query typed in May, at the decision — sits in the category that still mostly reaches a business’s own listing or page.

The content most exposed to AI interception is not the near-decision page most local businesses worry about protecting. It is the explainer sitting above it, the one most local content programmes build last, if they build it at all. What to do about that layer specifically is the subject of the local SEO for AI search guide.

None of this is a reason to stop writing the far-end explainer — a business absent from that layer forfeits the one place it might still get named inside the answer. It is a reason to write it knowing what it is actually competing against, which is not another article. It is an AI Overview reading the query at a 92% trigger rate and deciding, on the searcher’s behalf, whether a click is even necessary.

The question to ask about every page

Before anyone writes a brief, someone should be able to answer this in one sentence:

What has this reader already decided, and what will they be able to decide after reading this?

If the answer is “nothing in particular, but the keyword has 8,000 searches a month”, you have a keyword, not a strategy. The page will probably get written, it may even rank, and it will not do anything for the business.

Intent is not an attribute you look up. It is a judgement you make about a person mid-decision, and it belongs in the same conversation as pricing, margin and sales — which is why it sits inside the FOUND, TRUSTED, CHOSEN framework rather than beside it.

The practical next step is getting that judgement out of your head and into something a writer can work from: the content brief is where it either survives or gets lost.

Questions

What are the four types of search intent?

Informational (learn something), navigational (reach a specific site), commercial (compare options before buying), and transactional (act now). Some practitioners add local as a fifth, though it behaves more like a modifier that can sit on top of any of the other four. The labels are a useful first sort and too coarse to build a page on, because two queries with the same label can sit months apart in a buying decision.

How do I work out the search intent of a keyword?

Search it and read the results. The pages Google already ranks are its current best model of what people want from that query. Note the format (guide, comparison, product page, tool), who publishes them (vendors, review sites, forums), and what the SERP features are. If every result is a listicle from a third-party publisher, a product page will not rank there no matter how well it is optimised.

Why does intent matter more than search volume?

Volume tells you how many people search a term. Intent tells you whether any of them are close to buying from you. A query with a tenth of the volume but sitting one step from a purchase decision will usually produce more revenue than a high-volume term searched by people who will never become customers.

Can one page serve more than one search intent?

Occasionally, and it is usually a mistake to try. A page that hedges between explaining a concept and selling a product tends to do neither well enough to rank. The exception is where the SERP itself is mixed — where Google is visibly uncertain and ranks both formats — in which case a single page with a clear primary intent and a secondary section can work.

How often does search intent change for a keyword?

Often enough that it is worth re-checking the SERP for your most important terms once or twice a year, and any time a page that used to rank starts sliding without an obvious technical cause. Intent shifts when a category matures, when a news event changes what people mean by a phrase, or when Google reinterprets a query and swaps the dominant result format.

How far apart can two 'informational' queries actually be in terms of buying readiness?

Very far, and a shared intent label hides it completely. Two queries can carry the wording of a how-to search and sit months apart in the same person's decision — an early symptom search typed before anyone suspects a real problem, and a near-identical search typed weeks later after a first fix has failed, which functions as evidence the issue is worse than assumed. What changed between the two is the distance to decision, not the intent label, which stayed the same the whole time.

Does a query's format on the results page tell you everything about its intent?

No, and it works best on the easy cases. Where the results page itself is ambiguous, Google's own Search Quality Rater Guidelines are explicit that a single query rarely carries one meaning, and instruct raters to judge a result against 'the dominant, common, or reasonable minor' interpretation rather than a single assumed one. The more reliable signals in an ambiguous case are the exact wording used — particularly words implying a repeated or failed attempt — and what a searcher looks at next.

Should every stage of a buying decision get a blog post?

No. Distance to decision should decide the type of page before it decides the angle of the writing on it. A far-from-decision question usually deserves a guide; a middle-of-funnel comparison is often better served by a tool that does the comparison directly than by an article describing how to do one; and a near-decision, local-transactional query needs a service page and a complete business profile, not an article at all. Publishing an article by default at every stage produces a content library that looks complete and converts unevenly.

If several keywords belong to the same buying decision, should they always share one page?

Only if they also share a distance from that decision. Grouping keywords by the decision a searcher is trying to make, rather than by shared vocabulary, is a real improvement on standard keyword clustering — but a single decision can still span several distances, from someone first questioning whether the problem is real to someone verifying a quote they have already been given. A cluster that spans that whole range usually needs splitting into at least two pages, even where a clustering tool, or a careful human doing the grouping by hand, would reasonably treat it as one topic.

Does Google's AI Overviews feature affect every local search query the same way?

No, and the difference is large enough to change where a content programme should focus its effort. A Whitespark study of 540 queries across three US cities and six trades found Google's AI Overviews triggered for roughly 92% of purely informational queries in a trade, against about 15% for narrowly local, transactional ones such as 'personal injury lawyers in Phoenix'. The click risk from AI Overviews currently concentrates in the far-from-decision, informational layer of local content, not in the near-decision service and location pages most local SEO effort protects first.