The best argument we've shipped came from a sales call, not a keyword tool
Every Friday, we spend 30 minutes rereading that week's sales and pilot call notes, hunting for the exact words a buyer used, not a summary of what they meant. One of those phrases became the entire premise of a post we shipped a few weeks later: "we don't want to give them GitHub access" turned into an argument about why companies buy a GitHub seat just so someone in ops can edit a workflow twice a year, and what removes that cost. A keyword tool would never have found that phrase, because nobody had searched it enough times yet to register.
What a keyword tool can't see yet
A keyword tool tells you the volume and difficulty of a term that already exists. Useful, as far as it goes. But a phrase only shows up there once enough people are already searching it, which means everyone with the same tool sees it at the same time. By the time it's visible, it's already crowded.
Where the real words live first
Before anyone searches anything, they say it out loud to another human: a colleague, a rep on a call. That version is clumsier than the polished, category-standard phrase a vendor's homepage eventually settles on, and it's exactly the thing worth writing down. "We don't want to give them GitHub access" is a real line from a real call. Nobody types that into Google. It's the sentence underneath the problem, months before the problem gets a tidy name.
Thirty minutes, every Friday
The habit is small enough to survive a bad week. Whoever ran calls that week drops the raw notes into an LLM, not to summarize them, specifically to pull out the exact phrases a buyer used. A phrase that shows up once might just be how one person talks. A phrase that shows up three times, from three unrelated calls, earns a spot on next week's list of topics worth writing about.
Running it through an LLM instead of a person skimming from memory isn't about speed, it's about honesty. A person recalling a call tends to reach for the tidy version of what someone meant, which quietly turns their actual words into marketing language before anyone decides to. Extraction has one job: the sentence exactly as said, not as improved.
Proof, not just a nice ritual
"We don't want to give them GitHub access," said independently on more than one call, became the entire premise of a post we shipped a few weeks later: why companies pay for a GitHub seat just so someone in ops can edit a workflow twice a year, and what removes that cost. Not a loose inspiration, the actual argument the piece was built on. "Everyone sets it up differently" is where our rollout and provisioning posts came from the same way. Neither phrase had enough search volume to show up in a keyword tool the week we needed it.
This is just talking to customers, and that's fine
Conversion copywriters have said versions of this for years. Copyhackers built an entire practice on writing ad copy in a customer's own words, down to a stated rule that no more than 10 percent of a landing page should be the writer's own language. We haven't added a new idea, just a schedule. Voice-of-customer research that only happens when someone remembers to reread old notes mostly doesn't happen. Thirty minutes, every Friday, fixed, is the only reason it survives past the first busy week.
The actual payoff
A post built from a phrase someone already said out loud doesn't need to teach the reader new vocabulary first. It just uses the words they already reached for, so they recognize their own problem in the first five words instead of the first paragraph. We're not smarter than a keyword tool at finding volume. We're just closer to the conversation than a tool ever gets to be.