The tempting misreading
Once you accept that AI search answers questions, the obvious move is a page per question. Teams generate hundreds of near-identical URLs, each answering one narrow query, and end up with a library nobody would read and search engines classify as scaled thin content.
The underlying error is treating retrieval as if it worked at page granularity. It doesn’t — models retrieve passages. A single strong page can be quoted for a dozen questions if it is structured so each passage stands alone.
When a question deserves its own page
Give a question its own URL when someone would plausibly search for it as their whole problem, when the answer needs more than a few hundred words to be genuinely useful, and when you can say something a competitor hasn’t. All three, not one.
When it belongs inside an existing page
Variations in phrasing, follow-up questions, and narrower cases of a broader topic belong as sections of the page that owns the topic. Consolidating them concentrates the authority signals that decide whether you get recommended rather than merely listed.
Ten questions answered well on one page beat ten pages answering one question each.
The test that settles it
Draft the page. If it can only reach useful length by restating the same idea in different words, it should have been a section. Padding is the symptom that the decision was wrong, and no amount of editing fixes a page that had nothing extra to say.
What consolidation buys you
Fewer, stronger pages are easier to keep current, easier to link to, and easier for a model to treat as the canonical source on a topic. They also fail more gracefully: one excellent page that drifts slightly out of date is less damaging than forty thin ones nobody maintains.
If you already have the sprawl, the fix is a merge rather than a delete — fold the useful passages into the topic page and redirect. You keep the substance and lose the dilution.