I mined 636 Google Autocomplete queries before building a niche site — here's what people actually search for

Most niche-site builders pick a keyword, check a volume number, and start writing. Before building my latest project, I tried something cheaper and more honest: I scripted Google's free Autocomplete API and harvested every real query pattern around one product niche — car laptop desks.

The method (free, ~300 requests)

Google's suggest endpoint (suggestqueries.google.com/complete/search) returns what real users type. For each seed term (car laptop desk, car laptop stand, laptop desk for car...), I appended a–z and 0–9 suffixes and collected everything — 296 queries later I had 636 unique search phrases, each a real thing people type.

Simple scoring: +3 if it contains a year (fresh demand), +2 for question words (content opportunity), +1 for 4+ words (long tail).

What surprised me

1. The top query wasn't about products at all. "Is it legal to have a laptop mounted in your car" appeared across 17 different seed expansions. Nobody selling desks answers it — that's a content gap you could drive a truck through.
2. "3D print" showed up 12 times. A meaningful slice of this market wants to print a stand, not buy one. A product-only site would never serve them; a guide on materials that survive hot cars (PLA softens at dashboard temperatures) does.
3. Intent clusters map to site architecture. Position words (passenger seat / back seat / steering wheel / cup holder) and type words (stand / table / tray / mount) each had full suggest trees — so the site's URL structure became exactly those clusters. The keyword data designed the sitemap.
4. Navigational noise is a filter, not a target. "near me", "amazon", "officeworks" queries tell you what NOT to build pages for.

The result

The data became CarLaptopDesk.com — position pages, type pages, and guides answering the legality/heat/DIY questions the autocomplete surfaced. Whether it ranks is now Google's call, but at least every page maps to a query someone actually typed.

Takeaway: before trusting a paid tool's volume estimates, spend 20 minutes with the autocomplete API. It won't give you numbers, but it gives you the shape of demand — and shape is what decides your information architecture.

Happy to share the harvesting script if anyone wants it — it's ~80 lines of Python with rate limiting.