Start with a real customer problem, expand it into the language people search, filter for fit and effort, then read the results page before you write. A low difficulty score helps—but it is never the whole decision.
For a newer site, a useful keyword is not simply one with a low difficulty number. It is a query with a real reader need, a result page you can credibly improve on, and a natural next step your site can help with. If any one of those is missing, the keyword can be cheap on paper and expensive in writing time.
| Signal | What it can tell you | What it cannot tell you | Useful decision |
|---|---|---|---|
| Search demand | Whether enough people may use the phrase. | Whether those people are your readers. | Keep only terms connected to a customer problem you can answer. |
| Keyword difficulty | A directional way to sort a list. | Whether your page format or evidence will beat what already ranks. | Use it as a filter, not a promise. |
| Search results | The kind of answer Google currently rewards. | Whether your draft is genuinely better. | Decide article vs. tool page, then find a specific angle. |
| Your site data | What Google already shows for your pages. | The whole universe of demand; some queries are omitted or aggregated. | Look for clues, then expand beyond them. |
Search Console’s own documentation explains that impressions, clicks, and average position are different measurements—and that average position is an average, not a fixed rank. Read Google’s metric definitions ↗
The common trap is to brainstorm ten clever variations, look each one up, and end with ten unrelated numbers. Start smaller. Pick a seed that describes the job a customer is trying to do, then let expansion data show the vocabulary around it.
Use the question behind the search, not your internal product language. “How do I check whether AI mentions my brand?” is a better starting point than a feature label. A local business, ecommerce store, or SaaS team will use different words for the same underlying concern.
Look for related and suggested phrases together. Mark which ones describe the same reader job, which are useful sections for one page, and which are merely adjacent. RankTopAI’s AI keyword research tool is built for that expansion-and-clustering step; do not treat every returned phrase as a page idea.
Set a minimum demand threshold appropriate for your site, put very hard head terms aside, and discard anything off-topic. The relevance filter is the one people skip: a phrase can have attractive metrics and still lead to the wrong audience, wrong product, or wrong action.
Open the results for the survivors. If the page is full of tools, a blog post is likely the wrong asset. If it is full of generic listicles, a specific walkthrough, comparison method, or firsthand example may have room. If it is all major brands with deep resources, save the phrase for later.
Do not scan only titles. Ask what the searcher is trying to accomplish when they land. A page full of free checkers signals an immediate-action intent; a page full of explainers signals a learning intent. The right topic can become the wrong page if you ignore this distinction.
Multiple small publishers, dated pages, forum discussions that reveal unanswered questions, or several articles that cover the same topic vaguely. These are not guarantees, but they suggest the existing answer may be incomplete.
Brand navigation, a wall of tool landing pages when you only plan to write a guide, or results dominated by heavyweight publishers with highly specific assets. A low score does not cancel these warning signs.
Use Search Console’s query and page views ↗ to connect a query to the page Google already shows from your site. Then decide whether to improve that page, create a complementary page, or leave the query alone. This matters because two pages that answer the same intent can compete with each other instead of giving Google a clearer result to choose.
In a July 2026 RankTopAI planning run, we started with two broad seeds around LLM SEO tools and AI visibility audits. The expansion returned a mixed list: some terms described a plausible customer job, while others were only loosely related.
The useful result was not a magic number. It was the ability to separate one LLM-tools article from one audit walkthrough, then keep a tool-intent phrase assigned to a landing page rather than forcing it into a blog post.
This is an internal planning example, not a prediction of rankings. For the resulting guide, see our LLM SEO tools article; for a related diagnostic workflow, see the AI visibility audit walkthrough.
This routine is intentionally modest. It makes a defensible next-content decision without pretending you can research the entire web in an afternoon.
Google notes that query tables do not show every possible query and that some queries are anonymized or subject to data limits. That is another reason to use your own data as a starting clue rather than the entire research process. See Google’s query-data limitations ↗
A well-written guide will not satisfy someone who wants an instant calculator, a category page, or a branded login. Match the asset to the job before you spend time on the draft.
Close variations often belong together. One focused page can cover a main question, a natural synonym, and a few related sub-questions better than a cluster of thin posts that repeat each other.
Difficulty, volume, and position are useful signals; none is a guarantee. Google’s own SEO guidance is blunt on this point: there is no secret that automatically puts a page first. Use research to reduce wasted work, then earn the result with a genuinely useful page. Read the Google SEO Starter Guide ↗
Use the free keyword research tool to expand a few seeds, compare related questions, and organize the results before you decide what deserves a page.
Try the AI keyword research tool →If your next question is whether a brand appears in AI answers, start with the free AI visibility checker. For a broader diagnostic checklist, read the free GEO audit guide.
There is no universal cutoff. Start with a range that reflects your site’s current authority and resources, then use the live results to decide whether the existing pages are realistically beatable. A lower score is useful only when the phrase is relevant and the page type matches your plan.
Yes, if expansion data and the results page support the idea. Search Console shows what Google has already associated with your site; it cannot be the complete list of opportunities for a new topic.
Target one reader problem. If several phrases describe the same problem naturally, they can live in one article. Split them only when the reader’s task, expected format, or next action changes.