Keyword Research at Scale: The 21-Query Method
Open a keyword tool. Type your product name. Sort by volume. Grab the top 50 results. Maybe run a competitor domain. Call it done.
That process can miss buyer questions outside obvious product terms; a fixed percentage is not established here.
We have applied this research process across 85+ e-commerce accounts. Treat the 1,300-2,500 keyword range as a planning illustration, not a guaranteed count of viable terms or a typical first-week result.
The difference between 5 queries and 21 isn't incremental. It expands the questions you inspect; keyword count still cannot establish the share of a market you can profitably reach.
What Standard Keyword Research Can Miss
Standard keyword research has a structural problem. It's built around one type of query: plug in a seed keyword, get related keywords back. That's one dimension of discovery.
Your actual market has at least five dimensions of demand. Each one requires a different query type, a different tool configuration, and surfaces a different type of keyword. Skip any dimension and you leave entire audience segments untapped.
Here's what gets missed:
Competitive intelligence keywords. Your SERP competitors rank for terms you've never considered. Query types 6-12 pull rankings from competitor domains and surface entire keyword categories that seed-based research can't reach. In one skincare account, competitor ranking analysis uncovered 14 distinct audience angles the brand wasn't targeting. Not 14 keywords. 14 angles, each representing 50-150 keywords.
Question-based keywords. People Also Ask questions reveal how customers describe their problems in their own language. Query 13 mines these two levels deep. Level 2 PAA is where the insights live - questions about competing product categories, specific objection patterns, and comparison angles that don't show up in standard keyword suggestions.
Long-tail expansion keywords. The alphabet explosion (Query 14) is the single most tedious step in the protocol. Take your top 5 product types, append every letter A through Z. 130 sub-queries. We've found entire market verticals from this method that never surfaced in any other query type.
Persona and problem queries can reveal a distinct buyer question. Their CPC and conversion quality depend on the market, offer and destination; specificity does not establish a fixed cost or conversion advantage.
Domain intersection can expose terms absent from the initial seed set. Inspect their relevance and metric definitions; the share worth retaining varies by account and source.
The 21-Query Protocol: Category by Category
The protocol runs in sequence. Each category builds on the previous one's output.
Category 1: Core Queries (Queries 1-5)
Purpose: Build the baseline keyword universe. Expected output: 300-500 keywords.
Start with your seed terms. Not product names - product types. "Running shoes" is a product type. "Nike Air Zoom Pegasus 40" is a product name. Seeds should represent how customers search, not how you organize your catalog.
Query 1 collects available search-volume estimates for the seed list; Query 2 explores related language. For example, hydrophobic and ceramic-coating terms may overlap, but not every hydrophobic product is a ceramic coating. Validate the actual product distinction before clustering.
Query 3 goes long-tail with autocomplete-style suggestions. Query 4 runs bulk expansion by feeding all seeds simultaneously to surface cross-pollination between product types. Query 5 enriches everything with CPC, competition level, SERP features, and trend direction.
The critical checkpoint: if you're under 200 keywords after Category 1, your seeds are too narrow. Go back and add product type variations.
Category 2: Extended Queries (Queries 6-12)
Purpose: Competitive intelligence and advanced metrics. Expected output: 200-400 additional keywords.
Query 6 identifies your true SERP competitors. Not who you think you compete with - which domains actually rank for your top 20 commercial keywords. This distinction matters. Your perceived competitors and your SERP competitors often differ.
SEO difficulty and domain-authority measures describe a provider's organic-search model, not your paid-auction cost or ability to convert. Use them only for their stated purpose; prioritize paid tests from relevance, product fit and the actual economics.
Queries 8-9 pull historical trends and SERP analysis. Keywords showing steady upward trends are emerging opportunities where early entry gives you cost advantages before competition catches up.
A pocket with modest reported demand may be worth investigating. Record the geography, date and source; provider CPC estimates and a summed volume do not establish the cost or availability of this account's traffic.
Checkpoint: 500-900 keywords cumulative. Under 400 means you need to run Query 10 against more competitor domains.
Category 3: PAA Extraction (Query 13)
Purpose: Question-based keywords from People Also Ask. Expected output: 100-250 question keywords.
Run SERP analysis with PAA extraction on your top 30 informational keywords. Two levels deep.
Level 1 PAA gives direct questions: "How long does ceramic coating last?" "Is it worth the money?" These are useful but predictable.
Level 2 PAA is where real insights surface. You click into a PAA answer and Google shows you the next layer of related questions. Here you find category comparison angles ("ceramic coating vs PPF - which is better?"), objection patterns, and questions that reveal customer language you'd never write yourself.
A People Also Ask question can reveal language worth investigating. It does not establish an unserved information gap, commercial demand or a profitable ad group.
Category 4: Expansion Queries (Queries 14-18)
Purpose: Systematic expansion across angles most brands never discover. Expected output: 275-600 additional keywords.
Query 14 is the Alphabet Explosion. Take your top 5 product types. Append every letter: "[product type] a" through "[product type] z." That's 130 sub-queries. Yes, it's tedious. Run them anyway. From 26 letters on one product type, we've surfaced new market verticals, product format variants, and content angles that no other query type reaches.
Collect relevant low-competition candidates as hypotheses. Test conversion quality with a compatible destination rather than assuming that a provider's competition label predicts a higher conversion rate.
Query 16 generates comparison queries. "[Your brand] vs [competitor]," "[product type A] vs [product type B]." People actively comparing are close to buying. Most brands don't target these terms at all, which means low competition for high-intent traffic.
Query 17 builds problem-solution queries. "How to [problem your product solves]," "why [symptom your product addresses]," "best [product] for [problem]." Lower CPC, earlier journey stage, critical for scaling beyond bottom-funnel.
Use supported buyer language and a useful destination to make the hypothesis concrete. Persona specificity alone cannot establish higher Quality Score or lower CPC.
The 900-1,800 cumulative count is an illustrative workload range. Review relevance and source coverage rather than diagnosing a fixed-count shortfall.
Category 5: Shopping Queries (Queries 19-21)
Purpose: Shopping-specific intelligence. Expected output: 150-300 shopping keywords.
Organic visibility can reveal a relevant page or query to investigate. It does not automatically raise paid Quality Score or guarantee a combined click-through gain. Retain separate organic and paid reporting definitions.
Some combinations will have reported demand; others will be missing, below a provider threshold or irrelevant. Preserve unknown states and investigate useful terms rather than treating a missing estimate as confirmed zero demand.
Query 21 is Domain Intersection. Compare relevant domains and retain query, geography, period and provider definitions. Organic ranking gaps suggest research candidates; they do not identify competitor revenue or prove an uncovered paid opportunity.
The earlier 1,300-2,500 count is an example workload range. A smaller relevant set may be sufficient; keyword quantity alone does not diagnose narrow research.
The Scoring System That Prevents Keyword Bloat
A list of 2,000 keywords without prioritization is a spreadsheet that makes you feel productive. It's not a campaign architecture.
Every keyword gets scored using four weighted factors:
A prioritization score is an editorial planning rubric. State its factors and weights, retain missing observations and validate the next test against actual economics. SEO difficulty and estimated CPC do not identify paid profitability.
Then a volume tier multiplier adjusts the final score:
| Tier | Monthly Volume | Example multiplier | Decision context |
|---|---|---|---|
HIGH | 500+ | 1.0x | Review relevant demand and actual account evidence |
MEDIUM | 100-499 | 0.8x | Preserve uncertainty and test useful intent |
LOW | 50-99 | 0.5x | Low query volume alone does not prohibit Smart Bidding |
REVIEW | Below 50 | Not prescribed | A missing or small estimate does not prove no useful demand |
A keyword scoring 85 with 250 monthly searches (MEDIUM tier) gets a prioritized score of 68. A keyword scoring 70 with 800 monthly searches (HIGH tier) gets 70. The higher-volume keyword deploys first despite the lower base score because it generates actionable learning faster.
The weighted score can suggest an initial allocation. Validate it against actual contribution, uncertainty and test needs; a subjective score does not make the resulting budget an objectively optimal one.
Funnel Classification: Where Each Keyword Belongs
Every keyword gets classified into one of three funnel stages. This classification determines budget allocation, bid strategy, landing page type, and deployment priority.
BOFU - Bottom of Funnel (60-70% of budget)
Classify the query by the actual buyer question, product facts and destination. A layer name cannot establish competition, CPC, conversion rate or a prescribed budget share.
Bid strategy: Target ROAS or Target CPA. Landing pages: Product detail pages with direct purchase path.
MOFU - Middle of Funnel (25-35% of budget)
Keywords indicating active evaluation include comparisons, reviews and alternatives. Their conversion rates and appropriate launch timing depend on the offer, exposure, economics and measurement; no universal 2-5% rate or 30-conversion prerequisite is established.
Bid strategy: Smart Bidding with longer learning periods. Landing pages: Collection or category pages with comparison capability.
TOFU - Top of Funnel (5-10% of budget)
Research-stage queries can support useful education, but costs, conversion rates and later contribution require actual evidence. Neither a universal 0.5-2% conversion range nor a 60-80% CPC discount follows from the label.
Bid strategy: Maximize Clicks or manual CPC. Landing pages: Educational content or advertorial pages.
The deployment sequence matters as much as the classification. Start with BOFU. Prove the bottom of the funnel works. Then add MOFU. Then TOFU. Each layer gets its own advancement criteria:
The retained stage sequence is an illustrative planning idea: start with the clearest buying question, then test adjacent questions when contribution and usable evidence support it. Fixed elapsed days, conversion counts, budget splits and attributed ROAS thresholds do not validate expansion for every account.
Define the actual advancement criteria before testing. The example’s counts and splits are planning assumptions; using a gate does not itself prove sufficient evidence or prevent every expansion mistake.
The 7-Layer Angle System
Beyond funnel classification, every keyword maps to one of 7 demand layers. Each layer represents a different type of audience, intent pattern, and competitive landscape.
Layer 3: Conquest. Competitor brand captures. "[Competitor] alternative," "[competitor] vs [your brand]." Most brands leave the most money here. These searchers haven't committed yet. Budget: 10-15%.
Layer 4: Temporal. Seasonal and event-driven. "Back to school [product]," "best [product] 2026." Requires a quarterly activation calendar. Budget varies by season.
Question-led research may support useful content and permitted follow-up. Its contribution needs actual measurement; audience selection can confound a comparison with cold traffic.
Layer 6: Persona. Demographic segments. "[Product] for seniors," "professional [product]." Each persona needs its own messaging and often its own landing page. Budget: 10-15%.
Layer 7: Shopping Micro-Intent. High-intent modifiers. "Buy [product] online," "best deal on [product]." Pure bottom-funnel. Highest conversion rates. Budget: 10-15%.
The layer system gives you a strategic view. You can instantly see which layers you're covering and which have gaps. Most accounts we audit are 80%+ concentrated in Layer 1 with virtually nothing in Layers 3, 5, and 6.
The Negative Keyword Connection
Keyword research and negative keyword architecture are two sides of the same coin. The 21-query protocol finds what you should target. A parallel process determines what you should exclude.
Review three possible negative-keyword scopes. None is mandatory for every account or campaign type; apply only supported controls justified by actual relevance and the campaign objective:
Irrelevant-query negatives: Identify genuinely irrelevant searches in the actual account. Jobs, free offers, education, DIY and complaint queries can be relevant to some businesses; do not apply them as universal waste lists.
Competitor-query controls: Review whether competitor searches belong in the particular campaign and whether supported exclusions improve interpretability. The presence of a competitor name does not prove that a click cannot convert.
Brand-query controls: Separate capture and prospecting reporting where useful, using supported settings in the intended scope. Excluding brand terms does not guarantee the cheapest click, highest ROAS, new-customer delivery or a particular routing outcome.
Retain negative terms only when actual relevance evidence supports their scope. A larger count does not establish lower CPA and can exclude useful demand.
Weekly search term mining adds 10-20 new negatives per week. The process takes 30 minutes: export last 7 days of search terms, cluster by theme, add irrelevant clusters as negatives, promote high-converting new terms to exact match campaigns.
Review actual search terms at a cadence appropriate to volume, conversion delay and risk. Keep the reason and scope for each proposed exclusion; four, eight or twelve weeks do not guarantee that waste is captured or that the structure is effective.
The RSA Connection: From Keywords to Ad Copy
The 21-query protocol doesn't just build keyword lists. It builds the foundation for angle-clustered RSA copy.
Each keyword layer surfaces different buyer psychology. Layer 1 (Product-Defined) buyers are feature-focused. Layer 3 (Conquest) buyers are comparison-focused. Layer 6 (Persona) buyers are identity-focused. Each needs different headlines.
The system works like this: for each ad group created from your keyword clusters, identify 4-6 distinct emotional angles from the keyword language itself. Generate 25 headline candidates across those angles. Score each on a 100-point rubric. Select the top 8-10 that pass the gatekeeper test: every headline must be visualizable, falsifiable, and unique to your brand.
A usable headline must fit the actual product and retained evidence. For a footwear example, 'Compare Fit and Cushioning Options' describes a task; a medical or pain-relief promise needs its own substantiation and appropriate review. A craft rubric cannot approve that promise by itself.
This connection between keyword research and ad copy quality is why the protocol matters beyond just finding more keywords. Better keyword coverage with generic RSAs still underperforms. The 21-query protocol gives you both the coverage and the emotional intelligence to write copy that converts.
Case Study: From "Maxed Out" to Primary Growth Engine
A DTC skincare brand. $25K/month Google Ads budget. 180 keywords across 6 ad groups. Growth stalled for 4 months. CPA stuck at $34. They were shifting budget to Meta because they believed Google had nothing left to give.
We ran the full 21-query protocol.
Category 1 confirmed what they knew - their obvious product keywords were saturated. Category 2 changed the picture. Competitor ranking analysis revealed keyword categories they'd never considered: skin concerns by body area, ingredient-specific searches, age-demographic queries.
By Query 18 (Persona Queries), we'd found 14 distinct audience angles they weren't targeting. Not keywords. Angles - each with 50-150 keywords.
The biggest blind spot was Layer 3 (Conquest). Five competitor brand angles with combined monthly volume of 8,000+ searches. Nobody from their brand was showing up for "[competitor] vs" queries.
Layer 5 (Question) surfaced 6 informational angles that became retargeting audience feeders. Layer 7 found 3 long-tail pockets with 90+ buying intent scores and virtually no competition.
Author-reported observations after seven weeks, without an isolated causal estimate:
- Keywords: 180 to 1,400+
- CPA: $34 to $26.50 (about 22% lower)
- Conversion volume: reported 40% increase
- $18K/month revenue estimate attributed to new angles; incrementality not established
The account retrospective reports a changed acquisition mix after research work. Without a suitable comparison it does not isolate the protocol's contribution.
Building the Architecture From Your Keywords
The keyword list is the input. The campaign architecture is the output. Bridging the gap requires four steps.
Step 1: Classify every keyword. Three dimensions: funnel stage (BOFU/MOFU/TOFU), layer assignment (1-7), and campaign grouping based on shared intent and landing page requirements.
Step 2: Score and prioritize. Run the scoring formula. Apply volume tier multipliers. Sort by prioritized score. This determines which keywords get budget first.
Step 3: Map landing pages. Different layers need different pages. Competitor-capture queries need comparison pages. Persona queries need segment-specific messaging. Question queries need educational content with retargeting pixels. Using your homepage for everything is the most common architecture mistake we see.
Step 4: Deploy in stages. Start with Layers 1, 2, and 7 (highest intent, fastest feedback). Add Layer 3 (Conquest) in week 2. Add Layers 5 and 6 in weeks 3-4. Add Layer 4 (Temporal) as seasonal windows open.
Staged deployment prevents Google from spreading budget too thin across too many new campaigns simultaneously. Each layer gets optimized before the next one adds complexity.
What Changes When You Run This
Three things shift when you move from random keyword picking to systematic intelligence.
Second, your campaign architecture has purpose. Every campaign maps to a specific layer, funnel stage, and audience intent. Budget allocation becomes strategic.
Third, question-led research can support permitted follow-up and new test ideas. A returning-audience conversion difference does not establish a universal 3-5× lift or prove the keyword caused later purchases.
The research takes a week. The architecture takes another week. Results show up in weeks 3-7 as new campaigns gather data and optimize.
Put the framework to work
Open the twenty-one-query checklist