How to find content ideas your keyword tool can’t show you
Keyword tools show what people type into Google. Public discussions show what they actually ask, in their own words, and which questions nobody answers. Here is how to use both, on a real example.
Most content teams find ideas the same way: open a keyword tool, sort by volume, filter by difficulty, write. It works, and everyone in your market has the same list. That is why the top ten results for so many queries read like the same article.
There is a second source that most teams skip: the places your readers ask each other questions. It gives you three things a keyword tool can’t: the exact words people use, how many different people have the same problem, and which questions are left unanswered.
Below is the method, worked through on a real 325-comment Reddit thread where someone reviewed every meal kit they had tried and invited questions. Then we checked each idea in Semrush. The results were not what you would expect.
1Find where your readers ask
Look for threads where people ask for advice in your topic, not threads about your brand. Good signs: a post that invites questions (“ask me anything about…”), comparison threads (“X vs Y?”), and complaint threads (“is it just me, or…”). A few hundred comments from the right thread beat thousands of loosely related ones.
Our thread qualifies on all counts. The author reviewed about seventeen services and wrote You can ask me about any service and I’ll do my best to answer.
People did, for weeks.
2Pull out every question
Read every comment, including the replies three levels down, and copy out each question and each problem stated as a question. Keep the exact words. Skip the thank-yous and the recommendations for now; you want what people needed to know.
In this thread, that gave us questions about price, diets, allergies, living alone, cancelling and portion sizes. The phrasing is worth keeping because it is the phrasing your reader will recognise: nobody in the thread asks for “single-serve meal kit solutions.” They ask which kits are the most single-friendly?
(link).
3Count who asks
Group the questions and count the different people behind each group, not the comments. One person asking the same thing three times is one person. In our thread:
- What will it really cost? 6 people asked about price, budget or value, including
How much per meal after the discounts run out?
(link) - Cooking for one. 4 people asked about kits for a single person.
- No onions or garlic. 3 people can’t eat them, including
I can't do onions or garlic and that killed the usefulness of Sunbasket for me.
(link) - Low fat, specifically. 2 people, one with a prescribed limit:
50 grams per day total fat with the goal to gain weight
(link) - Cancelling. 2 people, one of whom wrote
It was kind of a hassle. Took about 15 minutes for each cancelation.
(link)
These numbers are small, and that is normal for one thread. They are not demand estimates. They tell you which problems are real and shared, and they give you the language to describe them.
4Mark what nobody answers
For each question, read the replies. Some get a confident, specific answer. Others get a guess, a “try X” with no reason, or nothing. Those are your content gaps: people asked, and the discussion couldn’t help them.
The low-fat question is a clear case. The person explained that of the plans they had looked at, none seem to really have a low fat choice
(link), and the most concrete advice in reply was to read every ingredient list. That is an article waiting to be written.
5Check search demand last
Only now open the keyword tool. Checking demand last stops it from deciding your topics; it decides how you frame and title them. We checked each idea in Semrush’s Keyword Magic Tool (US database, October 2026):
| Question from the thread | Keyword we checked | Semrush result |
|---|---|---|
| Cancelling | how to cancel hellofresh | 5,400/month, KD 34 |
| No onions or garlic | low fodmap meal delivery | 720/month, KD 21 |
| Low fat | low fat meal delivery | 140/month, KD 29 |
| Keto | keto meal kit delivery | 40/month, KD 32 |
| Real cost | how much is hellofresh after discount | No data |
| Cooking for one | best meal kit for one person | No data |
| No onions or garlic | meal kit without onion and garlic | No data |
| Cancelling | easiest meal kit to cancel | No data |
Three lessons came out of this table.
Customers and search tools use different words. Nobody in the thread said “FODMAP” when they asked; they said they can’t eat onions or garlic. The keyword tool had no data for that phrasing, but 720 searches a month for the clinical term. The discussion tells you the problem; the tool tells you the label to put on it. Use both: the search term in the title, the customer’s words in the opening paragraph.
“No data” doesn’t mean no demand. Four of our checks came back empty, including the cost question that six people raised in one thread. Keyword tools estimate from samples and miss long, specific questions. A topic with no reported volume but repeated real questions is often easier to win, because few pages are written for it.
The thread tells you what a winning page must say. “How to cancel HelloFresh” has real volume and competition. The thread tells you what searchers actually ran into: cancelling through chat, waiting for an agent, and the option not being where the help pages said it would be. A page that addresses those specifics beats one that restates the official steps.
A template for your idea list
Keep one row per idea, so you can sort by evidence and demand together:
| Column | What goes in it |
|---|---|
| Question | The exact words, with a link |
| People | How many different people asked it |
| Answered? | Well answered, poorly answered, or not at all |
| Search term | The closest keyword, and its volume (or “no data”) |
| Angle | What your page will say that the thread and the current results don’t |
Prioritise ideas with several people, a poor answer, and either measurable volume or a clear match to a buying decision. Drop ideas where one person asked once and the tool shows nothing.
Where else to look
- Google’s “People also ask” and question tools like AnswerThePublic show how questions are phrased at search scale. They don’t show who asked, or whether anyone answered well.
- Your support tickets and sales calls are the best source of questions from people close to buying. They miss everyone who never contacted you.
- Public discussions catch people comparing options before they ever reach you, and show the answers they got. That is what this method uses.
Reading a 325-comment thread for questions takes a while. Valtaris does the reading: export a thread with our Chrome extension and the report groups what people asked and said, counts each person once, and links every quote to its source. If you want to see the counting step in more detail, we walked through it in Voice of customer analysis: a worked example, and sorted the kinds of things customers say in Voice of customer examples.
About the data. Quotes are verbatim from public Reddit comments and link to their source; we don’t show usernames. The thread was captured on 27 September 2026 with 325 comments. People counts are from our own reading of the thread. Search volumes and keyword difficulty are Semrush estimates for the US, checked on 10 October 2026; “no data” means Semrush returned no result for that exact phrase. Valtaris is not affiliated with Reddit, Semrush or any company mentioned.