Why You Can Buy a Generator and Not a Climbing Bike

Current taxonomies store what your product is for as a single word. Your buyer describes it in a phrase. Across twenty queries captured on one day, that one difference decided every result.

A buyer asks for a bike. The answer comes back in degrees and tooth counts.

Someone types “best bike for climbing hills.” Google’s AI Overview answers with precision. A 50/34T crankset paired with an 11-32T or 11-34T cassette. A sub-compact 38T or 42T chainring with a cassette up to 50T for loose terrain. A seat tube angle around 76 degrees to stop the front wheel lifting on steep pitches. A high stiffness-to-weight ratio.

That is a complete specification.

Now look at the filters Google Shopping offers on that query: Type, Department, Features, Price, Age Group, Product Rating, Number of Speeds, and Tire Width. No facet for gearing ratio. None for frame geometry, weight, or stiffness. Zero of the four needs the AI just named are filterable. The buyer was handed a specification, then handed a control panel that cannot express any part of it.

Look closer at that list and the real problem is narrower than eight missing facets. Only one of those fields says what a bike is for, and it is Type: mountain, road, hybrid, touring, cruiser. Five words. Climbing, commuting, gravel racing, and loaded touring are four different purposes competing for those five slots. “Climbing hills” is not among them, so there was nothing for the buyer’s sentence to land on.

It gets sharper. The buyer said “climbing hills.” The listings read Specialized Crux Expert, Terrel CF, Pinarello X1 105, S-Works Tarmac, Canyon Grail CF SLX 8 AXS. Brand, model, groupset, material, speed count. Not one product title on that page contains the word “climbing.” The right product. The overlap is zero.

The commercial consequence is sharper. Sponsored Products for that query include a $299 step-through city bike and a $399 Guardian. Factor’s $11,299 climbing bike sits in an organic result, and a $13,499 S-Works Tarmac appears only in the Shopping tab. The best-fitting, highest-value inventory on the page is reachable only by paying for placement or by already knowing the model name.

Four queries where none of that happened

The temptation is to conclude that search is broken. It is not. Of twenty queries captured on 2026-08-04, four worked cleanly, and each worked for the same reason.

“best generator to run a whole house.” Google Shopping carries a Power Output facet measured in kilowatts, banded under 4.65, 4.65 to 10, 10 to 18, and over 18. A Power Source facet too: gasoline, propane, solar, natural gas. The AI Overview names the Generac Guardian 22kW and 24kW as the top choice. The Generac Guardian 24KW is directly purchasable at $6,875.10. Product titles read “Whole Home Standby” against a buyer who typed “whole house.” This works because every generator publishes its wattage, so a numeric facet could exist.

“dog food for sensitive stomach.” The AI names Purina Pro Plan Sensitive Skin & Stomach, and the product is right there in the results. The buyer’s exact phrase appears in the product titles themselves, carried as part of the product name across multiple manufacturers.

“crib mattress non toxic.” A vague, non-technical phrase. It resolves to two live facets: GREENGUARD Certified and Organic. A third-party certification standard exists, so the buyer’s worry has somewhere to land.

“best kayak for ocean paddling.” The AI says to get a dedicated sea kayak, 14 to 18 feet, sit-on-top for casual use. The facets are Type, with “Sea” as a value, plus Length, Rider Position, and even Tracking System. Three of three named needs are filterable. The use case itself became the product type.

Three of those four escaped the single word. A number cannot collapse into a word, because wattage is continuous. A certification is a flag standing beside the word rather than competing for it. And the dog food category skipped the taxonomy altogether by putting the phrase in the product name, so the match happened on the title.

The kayak did not escape. It got lucky.

The kayak did not have a better taxonomy. It drew a better word.

Same kind of product. Same shape of buyer. Same kind of need.

Kayak Type offers fishing, recreational, sea, whitewater. Bike Type offers mountain, road, hybrid, touring, cruiser. Four words against five. Neither taxonomy is richer than the other on the axis that decides the outcome.

The kayak buyer said “ocean paddling,” and “Sea” was sitting there waiting. That is the entire reason that query worked.

The bike buyer said “climbing hills,” and no word was waiting, even though the concept is completely standard everywhere else. BikeRadar publishes “Best climbing bikes 2026: 22 lightweight road bikes.” The Pro’s Closet sells a collection called “Climbing Road Bikes For Sale.” Factor Bikes has a page titled “Climbing Bikes.” Canyon publishes “Best lightweight climbing road bikes.” Editorial, merchants, and buyers all use the term.

A category always holds more purposes than the taxonomy has words for. The purposes left without a word are unsearchable, however established they are elsewhere. That is not a bug anyone introduced. It is what happens when a phrase has to fit in a word.

The pattern repeats when the need is described, not specified

“headphones that do not hurt ears” produces an AI Overview naming the Shokz OpenRun Pro 2, the Bose QuietComfort, and the Sony WH-1000XM6. None is purchasable in any Google commercial module on the page. The need is articulated. Nobody is standing in front of it.

“office chair for lower back pain” names the X-Chair X4, the Herman Miller Sayl, and the Branch Verve. None appears in the six paid slots. The titles that do appear carry Active Lumbar, Ergonomic, and Massage. Zero say “back pain.” The buyer describes the symptom. The merchant writes the clinical term. Both are correct. Neither can find the other.

That is the whole thing in one line. Buyers search in one language. Products are listed in another.

What this means for your catalog

This is not something you got wrong. It is whether the word for what your product does happens to exist, and that was settled long before you priced your first product.

On all four queries here that worked, the purpose escaped the single word: as a number, as a certification, or by being written into the product title instead. Publish the wattage. Publish the length. Get the certification. Where the purpose can leave the word behind, the retrieval layer carries your product to the buyer.

If what your product is for takes a phrase to say, and there is no number to publish and no certification to earn, nothing in the retrieval layer will supply that phrase for you. Not the shopping taxonomy, which has one word and has already spent it. Not the AI Overview, which states the need in terms no filter can express. Not the ad auction, which hands the slot to a $299 bike while an $11,299 climbing bike sits in organic.

You have to supply the language yourself. The buyer’s words go in your title, your description, and your collection names, before anyone builds a facet for them. Your listing is not wrong. You had to name the model, the groupset, and the material, because that is what the product is. But a merchant with a genuine climbing bike in stock has a buyer who described the need in detail, an AI that already specified the answer, and a path between them that is theirs alone to build.

That is what the intersection is for. Reach Dog maps buyer vocabulary against product vocabulary so you can see which of your products sit on the far side of the gap, and which words move them across. Open the graph and see the intersection.

Methodology

Twenty queries across three types, captured from live Google search results on 2026-08-04. Type A is a product plus a qualifier, Type B a situation, Type C a performance requirement on high-ticket goods. Each capture recorded the AI Overview, People Also Ask, sponsored and shopping modules, and the available filter facets.

Conditions, stated plainly because they affect what a reader will see if they repeat this. All captures used one browser profile on a United States connection, signed in to a personal Google account, so results carry whatever personalization that account attracts. A shopping browser extension was active and injected product panels into several pages; those injected modules are not Google results and were excluded from every count here. Search results change constantly, and this is a dated snapshot rather than a permanent property of any platform. Anyone can retype these queries and check, and should expect their own results to differ at the margins.