How to do keyword research

Where do the seed terms come from?

The first common mistake is starting by opening a tool. A tool only expands what you feed it; if the input is wrong, the output is a thousand wrong keywords.

Seeds come from the business itself. What customers say on the phone, the question that keeps coming back in your messages, the sentence your salesperson answers every day. There is a detail here we have watched play out many times: the word you use inside your industry is usually not the word a customer types. They write the plain name of the thing, not the formal one on your brochure.

Three more free sources that get ignored: the queries report in Search Console, which tells you what you are already shown for and which of those you did not know about; a competitor's category list, which leaks the structure of the market; and Google's own autocomplete, which is in effect a list of things people really type.

A ten minute exercise worth more than an hour with a tool: open your last three customer conversations and write down the exact words they used, without tidying them up. Those clumsy, unprofessional phrasings are what gets typed into Google.

Ten to twenty seeds is enough. More than that, at this stage, is just noise.

How do you grow the list?

Now the tool comes in. Feed every seed to a keyword suggestion tool and collect the output without judging it; deleting is the next step's job. Our free tool on the keyword research page does this for Persian and pulls the number using the rule explained in the next section.

Collect three more sources by hand, because tools tend to miss them. The related searches at the bottom of the results page, the related questions block, and the phrasings repeated across the titles of the first ten results. The third one matters most: a phrasing several competitors put in their titles is a phrasing that has worked in this market.

At this stage your list is probably two hundred to a thousand rows, full of duplicates and junk. That is normal. Keep one rule only: never hand-write a keyword you invented yourself. The list should hold only what came from a source or from a customer's mouth.

Which number do you trust? The rule that differs for Persian

Search volume is the number of times a phrase is searched in a month, on average. Two things to take from Google itself: these numbers are rounded, and the real figure fluctuates constantly. So read two keywords with a small gap between them as equal, not as one beating the other.

Now the rule that differs for Persian. Almost every global tool asks you to pick a country, and for Persian that choice damages the number. Iran does not exist in location based products at all, and if you pick another country you have picked the exit country of a VPN, not the user. What you get back is a slice of the same Iranian audience under the wrong label.

The right move: pick no country. For a Persian phrase the number with no country filter is the real number, not a rough worldwide estimate, because the only other country with native Persian speakers is Afghanistan and its share is small. The measurements behind this are in the field-note block further down the page.

One more fact about the numbers themselves: two different tools give two different figures for the same keyword, because each makes its own estimate. Do not hunt for the correct number; stay inside one tool and compare keywords against each other. The ratios hold, the absolute figures do not.

Two columns that simply do not exist for Persian: keyword difficulty and search intent. Any number a tool shows you under those two headings did not come from Persian data. Intent has to be read off the results page, and the method is in what search intent is.

Where does the monthly number mislead you?

Search volume is a monthly average, and an average hides the shape of the curve. A phrase that peaks for one month and is dead for eleven can carry exactly the same average as one searched steadily all year. On paper they are equal; in practice they are two completely different jobs.

That is why our own tool keeps the last twelve months beside the number. The result reader in pack-keywords.php slices the month list down to twelve on purpose, because a seasonal decision is not made from a single figure.

Two practical consequences. If your business is seasonal, publish the page two or three months before the peak; a new page needs time to settle, and publishing in the peak week means arriving after the party. And if a phrase rose because of a one-off news event, that demand does not come back, and the page you build for it comes down with it.

Group the keywords, do not just list them

This is where keyword research turns from a useless spreadsheet into a map. Group the keywords by the job the user wants done, not by how similar the letters look.

The practical test is simple: if the results pages for two phrases look nearly the same and the same sites are on top, Google treats them as one thing and one page should answer both. If the results pages differ, you need two pages even when the keywords look alike.

Every group has one head phrase that the rest orbit. That head phrase is what goes in the page title and the heading, and the rest of the group fits naturally inside the text with no forced repetition.

Do not build two pages for one group. That is the mistake that comes back later under the name keyword cannibalization: two weak pages standing where one strong page belonged, neither of them ranking properly.

Which page does each group go to?

The last step is the one that usually does not happen: next to every group, write down which page answers it. There are only three cases. A page you already have, a page you need to build, or no page at all because the group is not worth writing.

Take the first case seriously. If you have a page written for that same intent, improve it and do not build a second one. A strong page that has been collecting links and history for years is almost always ahead of a fresh one. A new page is justified when the intent genuinely differs.

For each group, open its head phrase in Google and look at the kind of pages on top. If the first ten results are all guides and you were planning a sales page, either change the page type or drop the group. Those ten minutes prevent writing a page that goes nowhere.

Add one column nobody usually adds: the date of the decision and its reason, in one line. Six months later, when a page moves in the rankings, that note is the only thing that helps you. On a site with no record of its own changes, every drop stays a mystery.

The final output is a plain table: group, head phrase, intent, target page, and whether that page is new or existing. That table is what people call a content map, and everything else is built on it.

When should you drop a keyword?

Dropping keywords matters as much as finding them, and it is taught far less. Four cases go without hesitation.

First, when the results page is full of sites you will never replace: a government site for a government service, or a large marketplace for a product name. Even rank two on a page like that gives you nothing. Second, when the intent does not match your work, however large the number is.

Third, phrases that are simply a competitor's brand name. Fourth, phrases that bring traffic but not customers; a web design company does not need a page for every general phrase about the internet.

And one case that runs the other way and should stay: the very low volume phrase. For a service business, a phrase searched a few dozen times a month that describes exactly what you do is usually worth more than a high volume generic one. A big number is not the goal in itself.