Keyword research has a deservedly poor reputation. For fifteen years the standard deliverable was a spreadsheet with eight hundred rows sorted by volume, which the client looked at once and filed forever. It was useless because it never answered the one operational question: which page do I write, about what, and in what order?
This article describes how we do it now. The main difference is not the tools but the deliverable: instead of a word list, it produces a page map with one intent assigned to each page. And in the process there is one source most agencies still underuse: the client's own data, which is free and real.
This has happened to us too often: impeccable, technically perfect research on a business that had been misunderstood. Before looking at a single volume figure, these five answers must be written down.
Which service or product carries the best margin? Not the best seller: the best margin. They are different things and SEO should chase the second.
Who do you not want to attract? As important as the above. A firm that works with companies does not want consumer enquiries, even at ten times the volume.
What is your real service area? The city only, the metropolitan area, the whole region, national. It changes the map completely.
What do customers ask on the phone before buying? This is the gold mine and almost nobody works it. Real questions, in real language, from people with real intent.
Which objections come up every single time? Price, timescales, guarantees, prior experience. Every recurring objection is a page somebody is searching for.
Ask those five questions to whoever answers the phone or the WhatsApp, not to the manager. The person handling enquiries all day knows the customer's real vocabulary better than any keyword tool.
This is the most skipped step and the one with the best cost-to-value ratio. If the site has been running more than six months, you are already appearing for queries you never chose and nobody has looked at.
There are three distinct treasures in there:
| Extraction | What to look for | What to do with it |
|---|---|---|
| Impressions without clicks | Queries with volume where you sit in low positions | Confirmed demand you are not capturing: candidates for their own page |
| Queries with no dedicated page | Terms ranking with a page that is not about them | The clearest signal of missing content |
| Unexpected vocabulary | Phrasings you would never have written | Your customer's real language: use it in titles and headings |
All of this comes out of Search Console analytics in a one-hour session, and it has a decisive advantage over any third-party tool: it is not an estimate. A tool tells you a query has "210 searches a month" according to its model. Your impressions tell you that query showed you 340 times last month, in your market, in your geography. The second figure is worth more.
The natural complement is looking at who occupies those SERPs. In rank tracking you can see what type of page ranks for each candidate query, which is what genuinely reveals intent — and, incidentally, whether you compete against businesses like yours or against unreachable national portals.
Here is the conceptual shift that turns a list into a plan. Two queries can be about the same topic and require completely different pages, because the person typing them is at a different moment.
| Intent | Example | Page that resolves it | Commercial value |
|---|---|---|---|
| Transactional | "full flat renovation valencia price" | Service page with prices or ranges | Very high |
| Commercial | "best renovation company valencia" | Comparison, cases, testimonials | High |
| Informational | "how long does a bathroom renovation take" | Guide linking to the service | Medium, early acquisition |
| Navigational | "[your brand] reviews" | Nothing new: look after your reputation | High but captures nobody new |
The operational rule that follows is short and strict: one intent, one page. If two queries share an intent, they go on the same page even when the vocabulary differs. If a query has mixed intent, the SERP tells you: look at what ranks. If the top 10 mixes guides and service pages, two intents coexist and you probably need both pages, well linked to each other.
Building one page per variant of the same intent: "renovations valencia", "renovation company valencia", "full renovations valencia". They are the same question with different vocabulary. Three pages competing with each other perform worse than one done properly, and they split your internal links as well.
For a couple of years now there has been a demand source that appears in no classic keyword tool: the long questions people put to a conversational assistant.
They differ in three concrete ways. They are longer and carry conditions inside them ("which renovation company in Valencia works with residents' associations on a fixed quote"). They have no measurable volume in the traditional sense, because they do not pass through a search box with public data. And they are answered by citing sources, which means the competition is not for a position but for being cited.
Ignoring this layer is a reasonable decision if your business does not feel it. Measuring it costs little: the AI Analytics module lets you research which questions exist around your category and who appears in the answers. In local categories we still find clearly commercial questions where no city business appears convincingly, which is a cheap opportunity while it lasts.
The practical way to fold this into research without duplicating work: once the page map exists, review each transactional page and ask whether it explicitly answers the conditions appearing in those conversational questions. Usually it does not, and adding two concrete paragraphs — areas, timescales, client type, conditions — is the entire job.
There is a dimension generic research ignores and that determines half the outcome for a city business: how geography enters the query.
There are three forms and they are worked differently. The query with an explicit place name — "dentist Benimaclet", "employment lawyer Valencia" — is the easiest to identify and the most competitive. The query with implicit local intent — "dentist near me", "emergency locksmith" — never mentions the city, Google resolves it by user location, and it will not appear in your lists if you only search for terms containing place names: you detect it by looking at where your impressions come from. And the query with no local component — "how much does a root canal cost" — which attracts people from the whole country, of whom only a tiny fraction are potential customers.
The practical consequence is that prioritising by aggregate volume systematically pushes you towards the third category, the one that generates the least business. A local business should concentrate most of its map on the first two, and use the third only as supporting content linked to the service.
On neighbourhoods, a useful clarification. Creating one page per area only works if each page contains something that applies exclusively to that area: real project references, particularities of the housing stock or the commercial fabric, travel times. If it is the same template with the place name swapped, you are not creating pages but duplicates, and the net effect on the site is negative. Our rule: one area page for each area where you genuinely have work to show. The rest wait.
Sorting by search volume is the fastest way to build an editorial calendar that generates no business. We score each candidate on four criteria and order by the total.
This is the practical difference between research that gets used and research that gets filed. The final document has one row per page, not per word, and every row carries:
URL (existing or to create) · a single intent · the primary query · the secondary queries the same page resolves · what it must contain to answer them properly · what it links to and where it receives links from · priority and status.
With that document, whoever writes knows exactly what to do without asking again, and whoever manages knows what is being done and why. A sheet of eight hundred keywords sorted by volume allows neither.
"Which existing page could cannibalise this one". Filling it in forces a review of the current site before creating anything, and prevents the most expensive problem with new content: competing against yourself without noticing.
Complete research does not expire monthly. What does change is the market and your own data, so we work at three rhythms.
Full, once a year, or whenever the business changes services or service area. Quarterly review of your new impressions: queries that have started appearing and were not on the map; the cheapest and most profitable update. Light monthly review of the priority list based on what has been published and what worked.
And a warning about the opposite: redoing the research from scratch every quarter is a common way of never executing anything. Research is 10% of the work; the other 90% is writing the pages and linking them properly.
Before paying for third-party estimates, look at the real queries you already appear for. It is free, it is first-party data, and there are usually ten obvious pages waiting in there.
Sign in to Semalt See AI AnalyticsBetween 40 and 80 well-chosen ones cover the real market of most city businesses. Five-hundred-keyword lists are inventory, not strategy.
They are useful for discovering vocabulary you did not know and for order-of-magnitude estimates. They are not useful for prioritising: that needs intent, competition and your own data.
Consolidate into one and redirect the other, almost always. Optimising both separately preserves the problem and splits internal authority.
Yes, when they link naturally to a commercial page and you can publish consistently. As the only strategy, they generate traffic that does not convert.
Worth measuring, and in practice they are resolved by adding concrete conditions to pages you already have rather than creating new content. It is one of the best effort-to-result jobs available right now.
Keyword research stops being a box-ticking exercise the moment its deliverable is a page map with one intent per row. Everything else falls into place from there: you know what to write, in what order, and why; you catch cannibalisation before creating it; and every published page has a clear function inside a system rather than being a loose contribution to the blog.
If you want the highest-return piece of work available today, it is this: open your queries with impressions and no clicks for the last ninety days and look for the ones with clear commercial intent. Connecting the domain and extracting that list takes an hour, costs nothing, and for most of the businesses we work with that exercise produced the first three pages of the plan.
And if you would rather have the full map built and prioritised — cannibalisation column included — get in touch: it is part of the free initial audit we deliver within 48 hours.
Free analytics for Google Search, SERP rankings and AI answers - plus fast indexing for your new pages.
Sign in to Semalt semalt.comGet in touch for a free audit
