A keyword is worth what its buyers pay, not what its searchers click.

Semrush labels “walking pad” an informational query. Then it reports a $1.62 cost per click on the same term. The market is pricing this keyword like a purchase; the tool is filing it under questions.

Search volume counts lookers, not buyers. A keyword’s value comes from the purchase behavior behind the search — how many searchers actually buy, and what those purchases are worth. On Amazon, that behavior is directly observable. On Google, it has to be estimated.

We’ll run a real category through the calculation in a moment. First, the trap that makes most keyword research start in the wrong place.

The volume trap, revisited.

The first article on this site argued that most search markets should never become websites — and put a four-question framework on the table for telling the difference. The Search Opportunity Framework treats search volume as a starting hypothesis, never a verdict. Volume looks like a go signal. It isn’t one.

That raises a question the framework deliberately leaves open: if search volume doesn’t decide value, what does? This article answers it. A keyword is worth what its buyers pay, not what its searchers click. Volume tells you how many people type a phrase. It doesn’t tell you how many of them buy, what they pay, or what margin is left after the cost of being found.

The people asking this question are searching for answers, and the answers they find are telling. The query “keyword value calculation” returns a first page of blogs, tool pages, and forum threads — no dominant framework, no locked-in incumbent, a modest 21% difficulty score. The demand for a better method is real, and the SERP hasn’t claimed it yet.

Marketplace operators have made the same point from the other side of the funnel: a high search count doesn’t equal market size. The count only shows how many listings compete for the term — never how many of those searchers end up buying.

Volume measures attention. Value measures money. The two rankings disagree more often than keyword research admits, and the worked example ahead will show exactly how much.

To see why, you have to compare what Amazon can observe with what Google can see.

Amazon knows who buys. Google only knows who clicks.

### What Amazon can see

Amazon is a marketplace, and marketplaces keep the receipt. In Brand Analytics, purchase behavior per search term is directly observable: search, click, add-to-cart, and purchase are all in the data — not modeled, not estimated.

Run 36 weeks of ABA data on a category you operate and you can compute, per term, the share of searchers who become buyers. The direction that surprised nobody who watched the ads: the highest-volume term in the category converted among the fewest searches into purchases.

On Amazon, you don’t estimate buyer proportion. You read it.

### What Google can’t

Google has no equivalent. The observable signals stop at the click, and the two proxies Google-side tools offer — intent labels and cost per click — are both unreliable in commercial categories. Semrush calls “walking pad” informational while reporting a $1.62 cost per click and two active ad creatives, one from a DTC brand and one from Amazon’s own shopping ads.

The market is bidding like the query is a purchase. The label says it’s a question.

The classification fails on its own terms. “Commercial intent keywords” — a phrase that literally names what it is — is labeled informational in Semrush, with a $0 CPC and a 0.01 competition score.

These are the tool’s own labels, not our interpretation of them. A label system that files that query under informational isn’t broken; it just wasn’t built for purchase economics.

The structural difference in one table:

None of this is a knock on Google. It’s a fee structure. Amazon charges per sale; Google charges per click — a marketplace prices transactions, an ad platform prices attention.

The platforms observe what they’re paid for, and they structure their data accordingly. The consequence for keyword research is not philosophical; it’s practical. One platform hands you the buyer behavior for free, and the other sells you a proxy for it.

So what do you do with the fact that Google can’t show you buyers? You estimate — and you need a tool that knows what it’s estimating.

The keyword value formula (it is a tool, not a score).

The asymmetry has a practical consequence: the two platforms require different value calculations.

“text

Amazon side (directly observable):

Keyword Value = Purchase_Count × Average_Order_Value × Profit_Margin

Google side (indirect estimate):

Keyword Value = Search_Volume × est_Buyer_Proportion

× Average_Order_Value × Profit_Margin − Content_Cost

`

On the Amazon side, every input is a data point you can pull from your own operation. On the Google side, one input — the proportion of searchers who actually intend to buy — is an estimate by construction, and the rest are ranges you'll fill from your own numbers: order value, margin, content cost. The formula doesn't remove the uncertainty. It gives the uncertainty named slots, which is the only way to keep the reasoning honest.

Which is exactly what a tool should do — and why this formula is not a scoring system. This formula is a boundary-checking tool. It tells you whether a keyword's commercial signal merits further research. It does not score the keyword, and it does not replace the framework's sequential judgment. Where inputs are uncertain, you establish a direction, not a number. The framework — the four questions laid out on the framework reference page — decides whether a market is worth entering at all. The formula decides whether a single keyword's commercial signal justifies the next step of research. Different jobs, different tools, and confusing them is how operators end up ranking keywords by numbers they invented.

A boundary check is a low bar by design. You are not computing a score to rank keywords against each other; you're answering one question — does this keyword's commercial signal merit further research — and attaching a direction. Weak signal: deprioritize. Strong signal: spend the deeper research there. The bar stays low on purpose, because the framework's questions carry the decision weight, not this formula.

The formula needs one input you can't get directly from Google: the proportion of searchers who actually intend to buy. Here are two ways to estimate it.

Two ways to estimate buyer proportion on Google.

### Path 1: SERP intent distribution

Read the first page as a market record. Classify the top ten results by page type — commercial product pages, brand storefronts, retailer category pages, marketplaces, media, forums, video — and the mix tells you who is competing for the click and why. The commercial share of the page is a rough proxy for buyer proportion: the more pages built to sell, the more searchers the market believes are buyers.

Run it on "walking pad": a brand storefront, Amazon product and live-shopping pages, two retailers, two media roundups, two videos — roughly six of eight results exist to sell something.

That is a direction signal, not a number. Page type is not purchase intent; a video review can sit next to a product page for the same query. The method is free and fast, and its resolution stops at "strong signal, weak signal, or noise." A page of forums and definition posts means a research audience; a page of product grids means a buying audience — the boundary moves by category, so the method stays rough by design.

Read the walking pad page as a record and the signal is clear: media roundups exist because readers are choosing between products, videos because shoppers compare before they commit, retailer category pages because the retailer expects the traffic to convert. Every page type answers the same question — who is searching, and what are they about to do?

### Path 2: CPC back-calculation

The second path starts from money. Cost per click is what the market will actually pay to be seen for a term, and advertisers only bid that way when the click carries commercial value. A high CPC is a market vote for buyer intent.

The walking pad family shows the ladder: the head term "walking pad" at $1.62 per click, "walking pad with incline" at $0.73, "walking pad under desk" at $0.56.

The shortest, most generic query draws the highest bid — advertisers compete hardest where buyers are most likely to land. The feature-modified terms cost less per click and signal more precise intent, which is exactly the pattern Amazon's own purchase ratios will show in the worked example.

Honest limits on both paths. CPC is not a conversion rate; it's an upper bound on what the market believes a click is worth. Without PPC conversion data, buyer proportion on Google remains an indirect estimate — you are reading market behavior, not measuring it.

Neither path produces a verified buyer percentage, and anyone quoting an "industry average buyer proportion" is inventing a number. What the two paths produce is a band: SERP structure gives the composition, CPC gives the intensity. Together they answer one question — strong, weak, or undetermined commercial signal.

When real conversion data eventually appears — your own ads, a client's account, a paid-search report — use it to calibrate the band, not replace the method. The band narrows with evidence; it never collapses into a single verified number.

Now let's run both paths on a real category — and see what Amazon's directly observable data says about whether the estimates hold.

A worked example: a fitness equipment category.

The category I operated is fitness equipment — a walking pad category, anonymized on purpose, because the method matters more than the brand. It ran long enough to accumulate two data layers: Amazon's directly observable purchase ratios, and my own advertising and session metrics.

Here is the first layer — 36 weeks of Amazon Brand Analytics data in the category, presented as ratios because absolute numbers would describe one store, not the search behavior:

Search term	ABA rank	Weekly search volume range	Search → purchase ratio	Cart → purchase ratio
walking pad	#1	48K–55K	0.47%	8.9%
under desk treadmill	#2	52K–54K	0.67%	12.2%
treadmill	#3	182K–192K	0.38%	8.9%
walking pad treadmill under desk	#4	19K–20K	0.54%	9.5%
desk treadmill	#5	17K–18K	0.56%	11.2%
treadmill clearance sale	#6	51K–57K	0.37%	6.4%

Read the table by rows and the ranking story flips. The #1 term by search frequency, walking pad, doesn't carry the highest purchase ratio; the #2 term by volume, under desk treadmill, does. Volume ranks popularity. The ratios rank purchase behavior — and the two rankings disagree at almost every row. That disagreement is the whole point of this article, sitting in one table.

The cart-to-purchase column shows where the difference lives. Under desk treadmill holds 12.2% of carts through to purchase — the highest in the table — while treadmill clearance sale holds only 6.4%, and their search-to-purchase ratios sit in the same order: 0.67% versus 0.37%. The gap between the two columns is the drop-off: the first says how many searchers are serious, the second says how many of those finish. Serious, finishable intent is exactly what a buyer proportion is trying to measure on Google — and on Amazon it's just a column in the data.

The second layer is the cost side. ACOS ranged from 4.5% to 75% with a median of 17.5% across 149 days; session conversion ran from 0.6% to 12.5% with a median of 2.41% across 171 days; ad conversion from 0.56% to 9.59% with a median of 2.89% across the same 149 days.

Products in the category sold between $214.40 and $469.99, and return rates ran 5% to 13%.

The wide ACOS spread is the honest part: buyer proportion varies by keyword, and so does the cost of converting it. The median says the category could work. The range says it depends on which keyword you buy.

Now fill the Google-side formula with real inputs and watch what it does — and doesn't do:

`text

Google side — walking pad, filled with ranges:

search volume = 246.0K monthly (Semrush US) ← observed

est. buyer proportion = indirect estimate; ABA ground truth

in this category: 0.37%–0.67% ← estimated

average order value = $214.40–$469.99 ← observed range

cost ratio (ACOS) = 4.5%–75% ad spend as share

of revenue; median 17.5% (149 days) ← wide range

profit margin = your product margin after all costs ← your number

content cost = your build + maintenance cost ← your number

“

The formula does not output a number here. It outputs a band and a question. Put six-figure search volume, a buyer proportion under one percent, and a $214–$470 order value together, and the band is wide — the ACOS data explains why: the margin input alone swings by more than a factor of ten, depending on keyword and campaign.

The calculation’s job was never to produce a price tag. It was to tell you whether the commercial signal merits the next step of research — and to show you which input is driving the uncertainty. That is a boundary check, not a score.

The where the evidence comes from page records which of these numbers are verified, which are sourced, and which are judgment.

What the calculation reveals.

Take the two rows that matter most. “Treadmill” draws between 182K and 192K weekly searches — the highest volume in the category — and converts 0.38% of them into purchases. “Under desk treadmill” draws 52K to 54K weekly searches, less than a third of the volume, and converts 0.67%.

The most-searched keyword in the category is not its most valuable keyword. Volume is popularity. Buyer proportion is value.

That inverts the ranking keyword research usually trusts. By volume, treadmill is the obvious target. By purchase behavior, under desk treadmill is the better signal — and cheaper to serve, because the longer phrase arrives with intent already attached.

The ad data behaved the same way: spend on the higher-ratio terms stayed closer to the ACOS median, while the big-volume terms wandered toward the top of the range.

The shape matches what operators report across categories: the horror stories come from high-volume terms, the quiet wins from the long tail. A term with massive search volume can run losses for a year while a term a tenth of its size quietly pays for the operation.

The pattern has a boundary. This is one category, observed over 36 weeks — not a law of search.

What generalizes is the habit: check what the searcher becomes before you count how many searchers there are. In practice, the research file changes shape too — the volume column stops being the first column you sort by, and the buyer-proportion column, even an estimated one, becomes the column you read first.

If this changes how you read search volume, it changes what you do before you type a keyword into a research tool.

What this changes about keyword research.

Keyword research changes order, not tools. The first thing to look at is not search volume — it’s whether the market is spending money on the click. Three checks, in order. What does the CPC say? A head term with active ad creatives is a buying signal; a $0 term is not.

What does the SERP’s commercial share say? Count the pages built to sell in the top ten. Who is already paying? Active advertisers are the market’s own demand signal — nobody bids on clicks that don’t convert somewhere.

Only after those three does volume matter, and even then it sets the ceiling, not the value. A high ceiling is nice. A high buyer proportion is what pays for the climb.

The walking pad category shows this in miniature: the highest-volume term carried one of the lowest buyer proportions.

The operators’ rule of thumb agrees — a keyword’s competition is not what makes it valuable; its purchase intent is. Difficulty scores measure how hard a page will be to rank; they say nothing about whether ranking it is worth anything.

The next question is what the SERP itself can tell you about a market — a topic for another article. For now, the takeaway is a habit, not a formula: count the buyers before you count the searchers.

Apply the four questions to your own market — the framework reference walks through Demand, SERP, Evidence, and Asset in order, the same sequence this article kept pointing back to. If you’re new here, start here — the reading path orients you to how Search Asset Lab thinks about search assets.

Data notes: Walking-pad SERP data from Semrush US database, 2026-08-08. Amazon purchase ratios from 36 weeks of Brand Analytics data in an anonymized fitness equipment category. Operational metrics from the same operation (149–171 days), reported as ranges and medians only.