Seed keywords from customer conversations

A keyword tool expands whatever you type into it. Type an industry term and you get an industry-shaped list. Type a customer sentence and you get the searches around that sentence. Nothing later in the process repairs a bad starting point, which is why the first keyword deserves more attention than the four hundredth.

The source is not a suggestion box inside a tool. It is the language your customers already used when they described the problem: sales calls, support tickets, reviews, intake forms, chat logs. This guide covers how to pull those phrases out, how to strip a sentence down to a seed, and how to check which of them anybody types at all.

What the first keyword decides

Expansion is arithmetic on the input. One seed becomes a few hundred related queries, and every one of those queries inherits the assumptions built into the seed. A seed drawn from internal vocabulary returns queries that describe your org chart. A seed drawn from a customer's sentence returns queries that describe the customer's situation.

Everything after this step — expansion, intent labels, clustering, page plans — works on the set you started with. Fix the set and the rest gets easier. Patch the set later and you are editing hundreds of rows instead of a dozen.

Where the language is already written down

You do not need to interview anyone to begin. The words are already recorded.

• Sales calls and demos. A prospect describes the problem before anyone reframes it. The first minute usually holds the most useful sentence in the call.
• Support tickets and chat transcripts. Customers write the failure in their own words, at the moment those words matter to them.
• Reviews, yours and your competitors'. A reviewer names what they were trying to do rather than the feature that did it.
• Intake and onboarding forms. A free-text field is a seed list with a submit button.
• Community threads and question sites. Where buyers ask each other, nobody has polished the phrasing.

Take the sentences verbatim. "We cannot tell whether the blog is doing anything" carries more than the note "wants reporting", because the note has already been translated, and translation is where the searcher's words get lost.

Stripping a sentence down to a seed

A verbatim is a sentence, not a keyword. The seed hides inside it.

"We cannot tell whether the blog is doing anything" contains "does my blog work", "blog traffic reporting", and "how to measure blog performance". Three seeds out of one sentence, each pointing at a slightly different search.

Two things to hold onto while you strip.

• Both halves of the vocabulary. For every customer phrase, write the industry term that means the same thing. The customer's words find the demand. The industry's words are what a searcher expects to land on.
• The verb, not only the noun. People describe what they are trying to do. Set up, compare, cancel, and fix each carry a different intent than the noun on its own, and that difference is what intent mapping sorts later.

Ten to twenty seeds is a normal result from an afternoon of reading. Past that you are expanding, which is a later step's job, not this one's.

Checking which seeds have demand behind them

This is the step that turns a vocabulary list into a keyword list. A phrase people say on a call is a candidate. A phrase people type is a keyword, and the two do not always match.

Run each seed through keyword research and read three things: whether there is measured volume, what intent the tool reports, and how difficult the surrounding terms look. The goal is not to rank the seeds. It is to split them into two groups.

The first group has demand. Those go forward into expansion and become the material for pages.

The second group has none, and it is the more useful of the two, because "nobody searches this" has two separate causes. Either the market describes the problem in different words, in which case the related terms with volume will show you exactly which words, and you have found better seeds than the ones you brought. Or the phrase is genuinely unused, and it belongs on a watch list rather than in a content plan. Customer language frequently runs ahead of search language, and a watch list tells you when that changes.

Zero volume on a term you use to describe your own product is a positioning finding, not a keyword finding. That case has a guide of its own.

Do it with an agent

Connect the Doverity MCP and the agent can read your project context, expand the list, and check demand in one pass. This prompt runs the whole thing; the keyword research skill spells out the same workflow from the agent's side.

Here are sentences customers used on recent calls and tickets:

[paste 8-15 verbatims]

1. Pull one or two seed keywords out of each sentence. Give me the
customer's phrasing and the standard industry term for each.
2. Run research_keywords on every seed, then get_keyword_metrics on
the results so I have volume, difficulty, and intent per keyword.
3. Split the output into "has measured demand" and "no demand yet".
4. For each no-demand seed, name the three closest terms that do carry
demand, so I can see what the market calls this instead.
5. Save the demand group with a topic:<topic> tag once I confirm.

The two-group split is the deliverable. The demand group is a content plan. The no-demand group is usually a better vocabulary than the one you started with.

Seed keyword FAQ

How many seeds should I start with?

Ten to twenty per topic. Seeds are inputs, and each one expands into hundreds of queries, so a long seed list mostly produces duplicate rows. If you cannot reach ten, the topic is narrow, which is worth knowing before you write anything.

Are seed keywords the same as long-tail keywords?

No, and they point in opposite directions. A seed is what you expand from and is usually short. A long-tail keyword is a specific multi-word query that comes out of that expansion. Seeds set the territory. The tail is where a smaller site competes inside it, which is the next guide.

Can I do this without paying for a data provider?

Half of it. Conversations, autocomplete, People Also Ask, and your own Search Console cost nothing, and this guide runs on the first of those. The paid part is the demand check, because volume and difficulty come from a provider. Doverity gives you credits when you sign up, the paid tier is $10 a month for $10 of usage, and self-hosting costs nothing at all.