How do I know when two questions are really the same question?
Ask whether the correct answer changes. If the ideal answer to both is the same paragraphs in the same order, they are one question wearing two phrasings and belong on one page. If answering one well requires facts the other does not need, they are two. The search results are a useful second opinion: when the top results for both are the same URLs, searchers are treating them as one intent.
The answer-identity test
Write the answer to the first question as a rough outline. Now try to answer the second from that outline. If you find yourself adding nothing, or only changing the wording of the opening sentence, you have one question. If you have to introduce a new concept, a new number, or a different decision rule, you have two.
This test is better than measuring word overlap because questions that share almost no vocabulary can have identical answers, and questions that differ by one word can have completely different ones.
Where the test gets slippery
Comparison phrasings and how-to phrasings often look like duplicates and are not. A reader asking which of two options to choose wants a decision rule and the conditions that flip it. A reader asking how to set one of them up wants a sequence. Same subject, different jobs, two pages.
The reverse trap is cost and time questions. Those tend to be genuinely one question no matter how many ways people phrase them, because the real answer is the same set of variables every time.
What to do with the phrasings you did not pick
They still have a job. Use them as section headings inside the surviving page, as the exact wording of the short answer, or as an explicit line in the body that mirrors how a person would ask it. What you should not do is spin up a page per phrasing with the words swapped. That is the pattern search engines classify as scaled abuse, and it is also just a worse experience.
For local variants specifically, the same rule applies with an extra condition — see how the hub structure handles that rather than generating a page per city.
Why this matters more once drafting is automated
A human writer naturally refuses to write the same article twice. A generator fed a list of forty keyword phrases will produce forty pages even when only twelve distinct questions exist, because nothing in the pipeline objects. The deduplication step has to be an explicit stage in the workflow, before drafting, with a human confirming the merge decisions. That stage is cheap and it prevents the most expensive failure mode in automated content production.
Topics: question discovery · deduplication · content planning · intent
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.