Market and keyword research

The keyword tool is the most tempting place to start and the worst. Type a service into it and ten thousand rows come back, sorted by a number that describes none of your constraints. Not your market. Not the cases you are licensed to take. Not the four pages already covering half of what those rows describe. The list looks like research. It is a record of what other people search for.

A research brief is the rulebook the keyword work has to obey, written before any rows exist. It names who the research serves, the task they are trying to finish, what the business can actually deliver, where it operates, which pages already do part of the job, and what gets thrown out on sight.

The payoff arrives later, when somebody asks why a high-volume term is not in the plan. With a brief you point at the rule it broke. Without one you are arguing from taste, and the biggest number usually wins that argument.

Write the rules before the rows, or the rows will write the rules for you.

What the brief has to contain

Write the audience task as a sentence a real person would recognize. Horizon Legal's is a person in Austin comparing injury counsel after an accident. That sentence does work later. It rules out people researching how adjusters calculate offers, who are a genuine audience with a genuine need and no intention of hiring anyone this month.

Record only coverage the organization genuinely has. Aspirational markets are the most common contamination in a brief, because nobody enjoys writing down that the firm cannot take a case in Dallas. Six months later the plan contains twelve Dallas pages and nobody remembers deciding to build them.

Search competitors are not business competitors

The firm you lose clients to and the site you lose the result page to are frequently different organizations. Horizon competes commercially with other Austin injury firms. On the result page it competes with legal directories, a state bar resource, a national publisher running an accident guide, and a settlement calculator somebody built as a lead magnet.

This distinction decides what you are supposed to build. When the market is held by tools and guides, a service page walks into a fight it cannot win however good the copy is. Reading the result as though only other firms were present produces a plan that loses on page type before a word is written.

List them separately in the brief. Business competitors on one line, search competitors on another. When the two lists barely overlap, that is the most useful finding of the whole research pass, and it usually arrives before you have generated a single keyword.

Reject rules are the field people skip

Include rules are easy and mostly useless, because almost everything looks includable at the start. Reject rules do the work, and they are uncomfortable to write because each one closes a door somebody was hoping to leave open.

Horizon's run: outside Texas, criminal defense, family law, anything implying a guaranteed outcome, and any term whose searcher is plainly a law student. That last rule removes several hundred rows with respectable volume and no possibility of a client, which is exactly the kind of row a volume sort promotes to the top of the list.

Write the brief in one sitting

Twenty minutes, before any tool is open. A field you cannot fill is a finding worth recording, not a reason to start researching anyway.

1. Name the audience and their task.One sentence each. If the task sentence could describe three different people wanting three different things, it is not finished.
2. Write down what the business can actually serve.Offer, market, and language, restricted to today. Anything aspirational goes on a separate line marked as such, where it cannot quietly become scope.
3. List the URLs that already do part of this job.This one field prevents most duplicate-page decisions later. A keyword that maps to an existing URL is an improvement rather than a new page, and that difference is worth thousands of dollars of writing.
4. List search competitors separately from business competitors.Open the result page for the obvious query and write down who is actually there. Directories and tools count. This is where your page-type decisions will come from.
5. Write the reject rules.Aim them at the terms that will look tempting in three weeks. A vague rule loses to a big number; a specific one does not have to argue.
6. End with the research question.One sentence naming what this run has to settle. Horizon's: which Austin injury tasks deserve an improvement to an existing page, and which reveal a genuinely missing one.

Before asking which queries are worth having, look at who is already getting them. The organic market is the set of sites and pages currently capturing demand around your audience task. It exists whether or not your own site does, which is the part that surprises people running this for a domain that launched last month.

This is a different question from keyword research and it comes first. Market work asks where demand is already being captured and by what kind of page. Keyword work asks which exact queries are worth evaluating. Run them in the other order and you generate a thousand rows with no idea which page type any of them wants.

Study the pages that are winning, not only the domains. The page type is the finding.

A new site is not a failed analysis

Analyze a domain nobody has linked to yet and Market Map will show close to nothing. That is a correct result rather than a broken one, and it is worth recording as a dated baseline before anyone concludes the tool is wrong.

The evidence you need sits on the competitor side regardless. Three Austin firms earning visibility through settlement calculators and explanatory guides tells you what this market rewards, and it tells you that whether your own site has four visitors or four hundred thousand. A brand-new project can finish this lesson completely, and often produces a sharper plan than an established one whose owner keeps staring at their own line.

The trap is copying the portfolio. Competitors ranking with calculators does not make building a calculator the strategy. It means the market rewards a tool for that task. Whether you can build a genuinely useful one, and whether that task is the one your business wants, are separate questions this lesson does not answer.

What the estimates are, and what they are not

Scope is what makes the comparison valid. Country and subdomain handling have to match across every domain you analyze, or you are comparing a national profile against a city one and calling the difference a competitive gap. Turn subdomains on only when the subdomain genuinely belongs to the business being studied.

When the winning page type is one you cannot build

This is the uncomfortable finding and it arrives often. The market rewards an interactive calculator and you are a two-person firm with no developer. Or the top four results are directories with a decade of listings, and no amount of writing turns a service page into a directory.

The wrong response is to build the closest thing you can and hope. A weak version of the winning format loses to the strong version, and it also loses to the formats you skipped in order to attempt it. The right response is to narrow. Take the slice of the task the dominant format handles badly and check whether a different page type serves that slice completely.

Directories are the clearest case. A directory ranks on breadth and cannot be beaten on it. What it does badly is depth on any single entry: no real case detail, no local specifics, no answer to the question a person asks once they have the list. That gap is a page you can build, and it is a different page from the one you were about to attempt.

Record this in the observation you carry into the next lesson. A market note saying the winning format is out of reach and here is the slice that is not will change which seed you expand. A note saying competitors rank with calculators will not.

Read the market in one pass

The output is not an export. It is one written observation strong enough to change which seed you expand in the next lesson.

1. Set the scope first.Country and subdomain handling, matching the brief. Set it once and hold it constant for every domain, or nothing you compare afterward means anything.
2. Record your own profile, including an empty one.An established site gets a directional starting point. A new site records that it has no modeled visibility yet, dated. Both are valid baselines and only one of them is embarrassing.
3. Analyze at least two search competitors.From the brief's list, which means a directory or a publisher qualifies. Two is the minimum that lets you tell a market pattern from one company's preference.
4. Drill from totals into pages.Overview shows the shape, Ranking Keywords shows the queries behind it, and Traffic Pages shows which page types earn it. The third view is the one that changes plans, and it is the one people skip.
5. Write the one observation that changes the next step.Not a summary of the market. One sentence about what the winning pages help someone do, added to the brief before you choose a seed.

Expansion is where the row count explodes and the thinking usually stops. A tool returns ten thousand ideas in four seconds, the column headers look quantitative, and sorting by volume feels like analysis. It is a sort. It answers a question nobody asked: which of these strings is typed most often by anyone, anywhere in the market, for any reason at all.

The job here is narrower and harder. Take one focused seed, learn the language patterns inside that topic, then reduce a large idea set to a small number of tasks you could act on. The brief you bring is not the one from lesson 2.1 either. It now carries the market observation from 2.2, and that observation is what tells you which seed is worth expanding.

One focused seed, expanded and then reduced against the brief. Volume sizes demand and decides nothing.

Choosing the seed

A focused seed represents a topic family the brief actually covers. For Horizon that is personal injury lawyer austin. Not lawyer, which returns divorce and immigration and criminal defense. Not a list of forty variations, which returns the same clusters forty times and spends the run.

Volume is a signal, not a score

Difficulty scores carry the same warning with more confidence attached. They model competition, and they cannot see that the top three results are a directory, a government page, and a tool. That is a market a good service page will not enter, whatever the number says.

Grouping variants without merging tasks

Spelling, word order, and singular against plural are almost always the same task. Personal injury lawyer austin and austin personal injury attorney are one decision wearing two costumes, and treating them as two rows is how a site sets out to cannibalize itself.

The harder call is a modifier that looks cosmetic and is not. Spanish speaking personal injury lawyer austin is not a wording variant. It is an access requirement that changes who can be served, what the page has to prove, and possibly who writes it. Group by the task the person is trying to finish, and when you are unsure, open the result page for both and see whether the same pages rank.

Keywords and prompts are different instruments

A keyword is a compact expression you can compare against demand estimates and a live result page. A prompt is a fuller request carrying context and constraints: I was rear-ended in Austin, have medical bills, and received an offer from the insurer, what should I compare before hiring a lawyer.

Both describe the same person and they answer different questions. Keyword work tells you about search demand and page fit, which is this module. Prompt work tells you how answer engines describe your category and whether they name you, which is modules 10 and 11. Substituting one for the other produces confident conclusions about the wrong system.

Expand, then reduce

The reduction is the lesson. A run ending in two hundred rows nobody has judged is the same failure as the ten thousand you started with, arrived at more slowly.

1. Reopen the brief and read it.Including the market observation from 2.2. Working from memory is how the reject rules quietly stop applying halfway down the list.
2. Enter one focused seed with matching scope.Same country and language as Market Map. A scope mismatch here invalidates every comparison you are about to make against the market work.
3. Apply the brief before you look at any number.Reject on audience, offer, market, language, and legal limits first. Doing it afterward means reading a metric attached to a row you were always going to cut.
4. Group rows into task clusters.Collapse the wording variants that describe one job. Ten thousand rows becomes a few hundred clusters, and clusters are the unit you can actually reason about.
5. Record what each cluster needs.Modifier or question pattern, likely page experience, current or planned coverage, business value, and what the live result looks like. This is the part that survives into lesson 2.4.
6. Read the metrics as separate signals.Volume, difficulty, cost per click, competition, intent, and trend each answer one narrow question. None of them ranks the list for you.
7. Mark every cluster Keep, Review, or Reject with the reason.A cluster with no reason attached will be re-argued from scratch in a month. Twenty-five to sixty decisions is a normal landing point rather than a target to hit.

Intent labels are where keyword research usually stops and where the real decision starts. Informational, commercial, transactional, navigational: a first sort, useful for about thirty seconds. None of them tells you whether to build anything.

The question this lesson answers is narrower and far more expensive to get wrong. For this task, on this site, what happens next? There are five possible answers and four of them are not create.

A modifier is not a page type and a keyword is not a page. Decide the action, then attach the evidence to it.

The five actions

The distinction carrying the most weight is between improve and create. Building a page for a task an existing URL already half-serves is how a site starts competing with itself, and that damage takes months to appear and longer to unwind.

The reason this is worth slowing down for is that create is the expensive option and it is also the one that feels like progress. Improving a page nobody celebrates produces a status update nobody reads. It is still usually the right call on an established site, and the three questions are what let you defend it.

What the live result page tells you

Open the actual result in the actual market before deciding anything. Not the keyword tool's intent label, which is inferred from the words in the query. The result page is the market's own answer to what this query means, and it is regularly at odds with the label.

Read the page types, not the brands. When eight of the ten results are guides and tools, the market has decided this is not a provider-selection query, whatever the modifier suggested. Sending a service page into that result is a decision to lose, taken before anyone writes a word.

Watch for competing interpretations. A query returning four results about one meaning and four about another is not one task and it is not one page. Split it, take the half you can serve, and record the other half as rejected with its reason.

Qualify one candidate at a time

Slower than it looks, and it does not batch. Each candidate needs its own look at a live result page.

1. Write the searcher's task in one sentence.In their situation, not your category language. If you cannot write it, you do not understand the query well enough to decide anything about it.
2. Open the live result in the target market.Record the dominant page types, the result features, and whether more than one interpretation is present.
3. Check what you already have.Search the site and inspect current ranking URLs. Half-coverage is the common case, and it is what turns a create into an improve.
4. Choose exactly one of the five actions.One task, one action, one owner. A task marked both improve and create is an unresolved argument rather than a decision.
5. Save the evidence with the decision.Target URL, what the result page showed, your confidence, and the rejection reason where there is one. Without it the next person repeats this lesson from the beginning.

Most keyword research dies at this step. The thinking was real, then it gets flattened into a spreadsheet column of keyword text, which discards the market, the result-page evidence, the reason a term was kept, and the reasons forty others were not. Three weeks later somebody asks why a term is in the plan and the honest answer is that nobody remembers.

The fix is to keep two records rather than one, because two different questions are being asked and a single list cannot answer both of them.

Save decisions, not wording variants. The evidence travels with the decision or the decision cannot be reviewed.

Two records, two questions

The saved research run answers what you saw. It holds the query set, the market and scope, the date, who ran it, the notes, and every Keep or Reject with its reason. It is a snapshot of evidence, and reopening it costs nothing, because it reads stored data instead of asking a provider for fresh numbers.

The keyword list answers what you committed to. It holds only rows that survived qualification, each with a plan type, status, priority, owner, target URL, and due date. It is a work record, and it should be dramatically shorter than the run that produced it.

Collapsing the two is the most common version of this mistake. Every provider result becomes planned work, the list grows to several hundred rows, and the team stops trusting it, because a list where everything is planned is a list where nothing is prioritized.

Where each decision goes

Two of the five actions have no destination, and that is deliberate. A consolidate decision is not keyword work. It is a redirect and a content merge, and it belongs in the technical queue with the pages named, not in a keyword list where it will sit unread. A monitor decision belongs in the research record with a review date, because promoting it to the plan would mean committing to work you explicitly said you were not ready to commit to.

The rule underneath all of this is that the keyword list holds work somebody has agreed to do. Anything else, however well researched, stays in the record it came from. Lists that mix committed work with maybe-later rows are the reason teams stop opening them.

Leave the rejects where they are

Deleting rejected rows feels like tidying and is closer to destroying evidence. The rejection reason is research output. It is what stops the same tempting term being re-proposed every quarter by somebody who was not in the room, and it is the fastest way to hand this project to whoever picks it up next.

The same holds for rows you could not resolve. Unresolved is a legitimate state with a review date attached. Dropping them silently makes the plan look more decided than it is, which is comfortable right up to the point where somebody asks about a term you never actually judged.

Hand it off without rebuilding it

The test is whether a colleague could open what you saved and act on it without asking you a single question.

1. Reopen the exact run from saved research.Confirm the topic, market, author, date, and decisions before planning anything. Reopening reads stored evidence at no cost; a fresh search asks the provider again and returns different numbers.
2. Promote only qualified rows.A row without a completed page-fit decision from lesson 2.4 is not ready, however good the keyword looks sitting there.
3. Set the work fields.Plan type, status, priority, owner, target URL, due date, and the reason it was kept. The reason is the field people skip and the one that gets asked about.
4. Tag for a decision you will actually make.Topic, audience, location, or page group, chosen because you will filter or report on it. Tags nobody filters by are maintenance with no payoff.
5. Route each decision to its destination.New pages to Content Plan, improvements to their existing URL, and only the decision-changing queries to Rank Tracker.
6. Leave rejected and unresolved rows in place.With their reasons and, where relevant, a review date. The record of what you turned down is worth as much as the list of what you kept.