Keyword Research

Companies name their product in their own words. Pagedeck described itself as a "collaborative document solution," and every page title and blog headline used that phrase. Almost nobody types it into a search box. The people looking for exactly this product search for "meeting notes app," "meeting minutes template," or "how to organize team docs."

Search engines work by matching the words a person typed against the words in a document. When the wording is off, everything built on top of it stops paying out. That is what happened at Pagedeck. Ravi published twenty blog posts over six months and never saw more than ten search visitors in a week. The posts weren't bad. The headlines were phrases nobody was searching.

Meetings about wording ran into the same wall every time. The only evidence anyone could bring was personal experience, so "every product manager I know calls it a meeting notes tool" collided with "that's insider vocabulary, nobody outside uses it," and the decision went to whoever was most senior in the room. If that choice was wrong, it took half a year for the mistake to surface.

Keyword research replaces that argument with a record. Search engines and ad platforms have been logging how often each phrase gets searched for years, and pulling those numbers turns a matter of taste into a comparison. Once "meeting notes app" shows 2,900 searches a month and "collaborative document solution" shows fewer than ten, there is nothing left to debate. Search engine optimization (SEO) splits into four areas of work, and this one sits at the front of the line for that reason. Target the wrong phrases and polishing pages or fixing the site's technical plumbing buys you nothing.

Five branches sit under this term. Analysis reads what the person behind a query wants. Metrics measure the candidates. Types are the two groups candidates fall into once measured. Structures place the chosen phrases on the site. Risks describe what goes wrong when that placement is careless. The five aren't a flat list — they show up in order inside a single pass of the work.

The pass itself has four steps.

1. Collect. Start by writing down roughly ten seed words that describe the product and its customers — for Pagedeck, meeting notes, shared docs, team wiki. Feed each seed into a keyword tool to pull related queries, then add whatever the search box autocompletes and whatever phrasing shows up in support tickets. A tool can also show which phrases send traffic to a competitor's blog, and the candidates nobody on the team would have invented usually come from there. Ten seeds become 300 to 500 candidates, and nothing gets filtered out yet.
2. Measure. Attach two numbers to every candidate: how often it is searched in a month, and how hard it is to rank near the top for it. Then draw a line through both and cut the list once. A site that just launched might keep only phrases searched at least 100 times a month with a difficulty under 30, which takes 400 candidates down to roughly 40. The survivors are the raw material for the next step.
3. Read. When two candidates have similar numbers, the choice comes down to what the searcher wants. This is the step where you run the query yourself and look at what fills the results page.
4. Group. Cluster the surviving phrases by intent and assign one page per cluster. One phrase, one page is the rule, and breaking it produces the failure described in the last branch.

Analysis is the branch step 3 relies on, and it holds one term. Search intent is what the person expects to find after typing a query. Someone searching "meeting minutes template" wants a file to download right now. Someone searching "best meeting notes app" wants a comparison list. Point a product page at the first query and the search engine will not lift it, because the ten results already ranking are all downloadable files and yours is a different kind of document. Intent falls into four rough groups: learning what something is, navigating to a specific site, comparing options, and buying now. The last two sit closest to signup, while the first two carry far more volume. The fastest way to check intent is to search the phrase and look at what occupies the first screen.

Metrics are the two rulers used in step 2.

1. Search volume — how many times a phrase was searched in a period, usually a month. The number a tool reports is an estimate derived from advertising data rather than a measurement, so it can be off by a factor of two or three, and seasonal phrases average out into a figure that matches no individual month. It's safer to use volume for ranking candidates against each other than as an absolute count.
2. Keyword difficulty — a 0 to 100 score for how hard it is to reach the top results for a phrase. Tools calculate it from the links and site size of whatever currently ranks in the top ten. The formulas differ, so the same phrase can score 40 in one tool and 62 in another; pick one tool and compare only inside it. A site built last month might cap its targets at 30, while a site with a year of history can reach for 50.

Types are the two groups those numbers split candidates into.

1. Head keyword — short, high volume, heavily contested. "Meeting notes" gets 50,000 searches a month at a difficulty of 78, and the first screen belongs to large publishers and long-established sites. A one-year-old site almost never takes the top spot there.
2. Long-tail keyword — longer, lower volume, and far clearer about what the searcher wants. "Free team meeting minutes template" gets 210 searches a month at a difficulty of 12. One of those looks trivial, but twenty of them together deliver as much traffic as a single head term, and because the searcher already knows their problem precisely, more of them sign up. A new site always starts here.

Structures are how step 4 places chosen phrases on the site, and the two work as a pair.

1. Topic cluster — one central article plus several supporting ones, all linked to each other. Under the topic of meeting notes you might publish templates, how to write them, tool comparisons, and sharing methods, each linked back to the center. It signals to a search engine that the site covers the topic broadly, and it gives a reader who landed on one article a path to the next.
2. Pillar page — the page at the center of that structure. It surveys the whole topic and routes readers to each supporting article. The contested head keyword usually goes here, while the supporting articles each take one long-tail phrase.

Risk is the branch that shows up when placement goes wrong. Keyword cannibalization is what happens when several pages target the same phrase and split the ranking between them. Pagedeck published three separate posts aimed at "how to write meeting minutes" over about a year, and the search engine never settled on which one to promote — all three hovered between positions 12 and 20. Merging them into one article and linking the other two to it moved the survivor to position 4. The one-phrase-one-page rule exists to prevent exactly this.

The five branches form a line. Metrics measure the candidates, the numbers sort them into types, analysis decides which side to take, structures seat the winners on real pages, and risk is the checkpoint on that seating. None of it pays out immediately: a newly published page usually needs three to four months before it settles into a position, which is why the choice made at the front of this line is expensive to get wrong.

"Volume on this one is great but difficulty is 70. Our domain won't crack page one in six months."

The site isn't strong enough to win that phrase yet, so pick a different one.

"Let's take this as ten long-tail pieces and add the head term later as a pillar page."

A sequencing proposal: win the easy phrases first, build authority, then go after the hard one.

"That phrase is already covered by the post from February. Update it instead of writing a new one."

Heading off two pages competing for the same query and eating each other's ranking.

At Pagedeck this work starts three weeks before the quarterly content plan. Ravi assembles the candidate table with volume and difficulty attached, and Mia lays it over the product roadmap. If automatic meeting summaries ship next quarter, the phrases that lead to that feature get claimed early. Sofia joins the session too, because the language customers use on sales calls overlaps with what they type into search more often than anyone expects. The candidate "meeting notes management" came straight out of a support message that read "our meeting notes never pile up anywhere." By the end of the session each row in the table carries a phrase, its volume, its difficulty, the intent behind it, the page that will own it, and the person writing that page.

The table never becomes the plan untouched. Twenty phrases sorted by volume go in, and about half get cut — either they overlap with something the site already ranks for, or what the searcher wants is not something Pagedeck sells. "Free meeting minutes template" has serious volume, but most people who type it take the file and leave. Whether to chase it anyway depends on the goal set in the content strategy: drop it when signups are the pressing need, keep it while the job is still getting the name known. The table is a set of options, not a schedule, and the cut is where a quarter's worth of writing effort actually gets allocated.

The most common mistake is assuming the big-volume phrases are the ones worth having. Put 50,000 searches a month next to 200 and the first looks 250 times more valuable. But volume counts the people who searched, not the people who reach your site. Ranking 15th for a 50,000-volume phrase brings in about ten visitors a month; ranking 1st for a 200-volume phrase brings in around sixty. A team that only chases the big phrases spends a year watching neither rankings nor traffic move, and usually concludes that search doesn't work for their category.

The second mistake is treating this as a one-time exercise. Building the first table takes days, which makes it feel finished. Queries keep shifting, though. New tools create phrases that didn't exist, a competitor moving onto your terms pushes difficulty up, and rankings you already won slip when the pages above you get rewritten. That is why the phrases you hold and the candidates that have appeared since deserve a review every quarter. A team still working from last year's table keeps writing toward positions somebody else took months ago. The first build takes days; every rebuild after that takes an afternoon, so the quarterly habit costs less than the catch-up does.