PageForge 1.3: check every run before it writes, update generated pages, spintax and resumable batches. See what changed
Account Get it free

Programmatic SEO keyword research: find patterns that scale

How to find a head term and modifiers that scale, check demand without trusting inflated volumes, size and cluster the page set, and build a keyword matrix.

Programmatic SEO keyword research is a different job from the keyword research you do for a blog post. You are not hunting for one good phrase but for a pattern, a head term plus a set of modifiers, where every combination is a real search that one useful page can answer. This guide shows how to find that pattern, check it without trusting inflated numbers, size and cluster the page set, and build a keyword matrix in a spreadsheet that tells you which combinations to build.

Key takeaways

  • Research a pattern, not a list: a head term, one primary set of modifiers, and a decision for every combination.
  • Start from what the business sells and where it delivers. Tools fill gaps. They should not set the scope.
  • Long-tail volume estimates are weak. Validate the pattern with a sample of combinations, Search Console data and the results page itself.
  • Size the page set with a funnel: raw combinations, then coverage, intent, data and a pilot. The number that survives is usually far smaller than the multiplication.
  • Cluster by results-page overlap, not by wording, so one page serves one intent.
  • Some combinations should never become pages: ones you cannot serve, near-duplicates, modifiers that do not change the answer, and anything that only exists to catch a search.
  • Keep the whole process in one spreadsheet, so the matrix becomes your import file.

What programmatic SEO keyword research is for

Classic keyword research ends with a list of phrases and a page for each. Programmatic keyword research ends with three things:

  1. A pattern that becomes a title and a URL, such as {Service} in {City} and /drain-cleaning-round-rock/.
  2. A modifier list that fills the pattern: the cities, the industries, the integrations.
  3. A decision for every combination: build it, merge it into another page, hold it until you have the data, or drop it.

The unit you evaluate is the pattern. If “drain cleaning in [city]” is a real kind of search with a consistent intent, you do not need to prove demand for each of 40 cities separately. You need to prove the pattern, then decide row by row whether you can serve that row well.

That shift changes which evidence matters. Keyword tools are built to rank individual phrases by estimated volume. For a pattern, the useful questions are different. Does the intent stay the same across combinations? Can you supply true, specific facts for each one? Does the business actually deliver there? A pattern that passes those questions is worth building even when every combination looks small. A pattern that fails them is not worth building however large the head term looks.

If you want the wider picture of how programmatic pages work on WordPress, from templates to publishing, read the programmatic SEO guide for WordPress. This post stays on the research step, which is the part that decides whether the rest is worth doing.

Anatomy of a programmatic SEO keyword pattern

Every pattern has the same parts. Naming them makes it easier to see which part is weak when a pattern does not work.

The head term

The head term is the thing the searcher wants: a service, a product category, a type of software, a course. “Drain cleaning”, “CRM”, “maths tuition”, “office cleaning”. A good head term for programmatic work has three qualities:

  • Stable intent. People who search it want roughly the same outcome, wherever they are and whoever they are.
  • A page you can deliver. You can sell it, book it or answer it on one page, with a clear next step for the visitor.
  • Many meaningful modifiers. It combines with places, audiences or attributes in ways that change the answer, not only the wording.

“Plumbing” is too broad to be a useful head term. It covers drain cleaning, water heaters, leak detection and remodels, and each of those has its own intent, price and process. “Drain cleaning” is a better head term. “Hydro jetting for restaurant grease traps” is probably too narrow to carry many modifiers, though it may deserve one carefully written page of its own.

Modifiers

A modifier is the part of the query that varies. Most programmatic patterns use one of these types:

Modifier typeExample queryPage type it needsFacts each page needs
Locationdrain cleaning round rockLocal service pageAreas covered, response time, local phone, price range
Audience or industrycrm for dental clinicsSolution pageFeatures that matter to that industry, workflows, proof
Integrationinvoicing app quickbooks integrationIntegration pageWhat syncs, setup steps, limits
Attribute24 hour plumber austinA section, sometimes a pageHours, call-out fee, what counts as an emergency
Comparisontool a vs tool bComparison pageVerified feature and price differences
Formatmaths tuition onlineVariant page or sectionHow the online version differs

The last column decides whether a pattern is feasible. A modifier only earns its own page when you can fill that column for every row, with facts that are true for that row and not for its neighbors.

Primary and secondary modifiers

Most patterns have one primary modifier that defines the page, and several secondary modifiers that describe it. In “emergency drain cleaning in Round Rock”, the city is primary: it changes who you send, how fast you get there and which phone number rings. “Emergency” is secondary: on most plumbing sites it is a section on the drain cleaning page, not a separate page for every city.

A simple test: a modifier deserves its own page when it changes the answer. If the emergency service has different hours, a different price and a different crew, it may be a pattern of its own. If it changes one sentence, it belongs in the template as that sentence.

Step 1: Find head terms from the business, not from a tool

The most common way to waste a programmatic project is to start with a keyword tool’s export. Tools are good at finding words. They know nothing about what the business sells, where it delivers or which jobs make money. Start with sources that do:

  • The service or product list. Write down every distinct thing a customer can buy or book. Split anything that has a different price, process or team.
  • Sales and support conversations. The words customers use on calls and in quote forms are often closer to real queries than the names on your services page.
  • Lead and order data. If the CRM records the service and the customer’s town, you already know which combinations bring in business.
  • Site search. If your site has a search box, its logs show what visitors expected to find and did not.
  • Search Console queries for the pages you already have. These are real queries that already showed your site, with impression counts.
  • Paid search terms, if the business runs search ads. The search terms report lists the exact queries that triggered ads.

For each candidate head term, write one line: what the page sells or answers, and what the visitor does next. “Drain cleaning: book a same-day visit by phone.” “CRM for dental clinics: start a trial with the dental appointment template.” If you cannot write that line, the head term is not ready for a template.

Then cut the list to one or two head terms for your first pattern. One pattern done well teaches you how your pages get indexed, what visitors do on them and where the data is thin. Five patterns started at once teach you very little.

Step 2: Collect modifiers that scale

With a head term chosen, build the modifier list. The rule is the same as in step 1: start from what is true, then expand.

Locations come from your service area. List the places you actually serve, with the details that change by place: who covers it, how fast, at what price. A census list of every town in the state is useful for spelling and county names, not for deciding scope. If you do not send a technician to a town, a page for it is a promise you cannot keep. The same applies to franchises and clinics: list the locations that exist. The use case for local service businesses shows the kind of columns that make each place distinct.

Audiences and industries come from your customers. A SaaS company building “CRM for [industry]” pages should start from the industries it has real customers, features or documentation for. An industry page with nothing industry-specific on it is the home page with a different heading. See how this works for SaaS page sets.

Integrations come from the product. Only list integrations that exist and that you can document.

Once you have the core list, expand it with sources that show how people phrase things:

  • Autocomplete. Type the head term followed by “in”, “near” or “for” and note the suggestions. Then add letters: “drain cleaning in a”, “drain cleaning in b”. Suggestions reflect real searches, but they are filtered and incomplete, so a missing suggestion proves nothing.
  • Related searches and “People also ask”. These show neighboring intents, which usually become secondary modifiers or page sections rather than new pages.
  • Search Console with a regular expression. In the Performance report, filter queries with a custom regex such as (in|near|for) to see which modifier forms already bring impressions. A filter such as drain|clog|sewer shows how people describe one service in different words.

Finally, normalize the list. Decide once how you write each place (“St. Louis” or “Saint Louis”), merge duplicates, note which names are neighborhoods inside a larger city, and mark places that share a name with a town in another state. Normalizing now saves you duplicate URLs later.

Step 3: Validate demand without trusting inflated numbers

Here is the uncomfortable part of programmatic SEO keyword research: the long-tail numbers in keyword tools are the least reliable numbers those tools show, and the long tail is exactly where programmatic pages live.

Why long-tail volume misleads

  • Estimates, not counts. Third-party tools model search volume from their own data sources, and two tools can disagree widely about the same long-tail phrase.
  • Ranges and grouping. Google’s Keyword Planner often shows broad ranges rather than exact figures unless the account spends on ads, and it groups close variants together.
  • Zero is not zero. Combinations such as “water heater repair leander” often show no volume in tools even when people search them, because the numbers are too small or too scattered to estimate.
  • Inflation at the head. The head term can look large while the combinations you would actually build are tiny. Adding head-term volume across a page set, or multiplying an average by the number of pages, overstates the opportunity.

So use volume to compare items inside your list, not as a forecast. “Austin is our largest market” is a fair reading. A monthly traffic projection for 40 pages is not.

Signals that hold up better

SignalWhat it tells youHow to checkWeakness
Search Console impressionsReal queries that already showed your sitePerformance report, filtered by service and placeOnly covers what you already appear for; rare queries are hidden for privacy
Results-page compositionWhether pages like yours are what searchers getSearch a sample of combinations and note the page typesResults vary by location and over time
AutocompleteThat a phrasing existsType the head term plus a modifierFiltered and incomplete
Google TrendsRelative interest between terms or regionsCompare the head term across regions or variantsRelative only, no absolute numbers
Your own lead dataWhich combinations bring businessCRM, call tracking, quote forms by town or serviceBiased toward places you already serve well
Paid search termsExact queries with commercial intentSearch terms report on a small campaignCosts money; ad results differ from organic ones

None of these is a forecast either. Together they answer the question that matters: is this a pattern real people search, with an intent your page can satisfy?

Validate the pattern with a sample

You do not need to validate every row. Pick 10 to 15 combinations that cover the range of your list: your largest market, a few mid-sized ones, and a few small ones at the edge of your service area. For each, record:

  1. Whether autocomplete or Search Console shows any sign of the phrasing.
  2. What kind of pages rank: local service pages, directories, guides, map results.
  3. Whether the results look alike in a large market and a small one.

If the sample is consistent, the pattern is real and the long tail inherits the intent. You can build “drain cleaning in Leander” even though no tool shows a number for it, because Round Rock and Cedar Park behave the same way and Leander is a town you serve.

If the small markets show completely different results, such as only directories and map listings, you have learned where the pattern stops before building pages for those places.

Step 4: Check intent on the results page

Intent is the reason a pattern works or fails, and the quickest way to read it is the results page itself. For each sampled combination, look at the top results and ask what kind of page Google shows:

  • Mostly local service pages and a map: a local service pattern fits.
  • Mostly guides and articles: searchers want an explanation. “Roof repair cost in Austin” usually wants a pricing guide, not a booking page.
  • Mostly lists and comparisons: “best CRM for dentists” wants an evaluation. A page that only describes your own product will struggle to match it.
  • Mostly map results and directory listings: business profiles answer the query. A location page can still help visitors who find you, but expect it to play a supporting role.

Two traps are worth naming.

“Near me” is not a modifier you can fill. “Plumber near me” depends on where the searcher is. You cannot build a page per “near me” query. Your city pages, your business profile and an accurate service area serve those searches together.

Results shift with location. When you check a local query from far away, you may see different results than a local searcher would. Treat the check as a sample rather than a measurement.

Write the result for each sampled head term in one or two words, such as “service pages” or “guides”. You will use that label in the matrix later.

Step 5: Estimate the size of the page set

The raw size of a page set is a multiplication: head terms × primary modifiers × any second modifier. Five services in 40 cities is 200 combinations. Add “emergency” as a second modifier and it is 400. The multiplication is where many projects go wrong, because it treats every combination as a page.

Estimate the real size with a funnel instead. Say you have 6 services and 40 cities, 240 combinations:

FilterQuestionExample result
RawEvery service in every city240
CoverageDo we deliver this service here?28 cities get all 6 services, 12 get 3: 168 + 36 = 204
IntentDoes the results page want a service page for this service?One service returns guides, so its 34 rows merge into one guide: 170
DataCan we fill every column with true, specific facts?40 rows lack response times or prices: 130 ready, 40 on hold
PilotWhich rows go first?The 20 rows that already show demand

The numbers are hypothetical, but the shape is typical: the set you can build well is a fraction of the multiplication, and the first batch is a fraction of that.

The estimate also tells you what the project costs. Say collecting and checking facts takes about 15 minutes per row. Then 130 rows is roughly 32 hours of work before any page exists. That is the real budget, and no generator can do that part for you.

How big is too big for search engines? Google’s sitemap documentation describes sites of about 500 pages or fewer as small (Google: what is a sitemap). Its crawl budget guide is aimed at sites with more than 10,000 pages that change daily, or a million or more pages (Google: managing crawl budget). Most service-business page sets are nowhere near those levels. For them, the limit is quality and review time, not crawling. A sitemap and a hub page that links every generated page help discovery at any size.

Step 6: Cluster keywords so one page serves one intent

Clustering decides which keywords share a page. In programmatic work it has a specific job: making sure the pattern produces one page per intent, not one page per phrasing.

Group by wording first

Start with the easy merges. Normalize plurals, word order, state abbreviations and filler words:

  • “plumber round rock”, “round rock plumbers”, “plumbers in round rock tx”
  • “drain cleaning round rock”, “round rock drain cleaning service”

Each line is one cluster. In a spreadsheet, a helper column that lowercases the keyword and removes words such as “in”, “the”, “service” and “tx” puts most of these next to each other once you sort on it.

Use results-page overlap for the hard pairs

Some pairs look different but mean the same thing, and some look alike but do not. For those, compare the top 10 results for each query and count the URLs they share. A common rule of thumb is that queries sharing three or more of the top 10 URLs belong on the same page. Pick a threshold and apply it consistently.

A hypothetical check for one city:

Query AQuery BShared top-10 URLsDecision
drain cleaning round rockclogged drain round rock6Same page
drain cleaning round rockhydro jetting round rock4Same page, hydro jetting as a section
drain cleaning round rocksewer line repair round rock1Separate pattern
water heater repair round rockwater heater installation round rock3Same page for now, revisit later

Run this check on your sample cities, not on every row. If “sewer line repair” splits from “drain cleaning” in Austin and Round Rock, it will almost certainly split everywhere. The relationship between two phrasings is a property of the pattern, so a sample settles it.

Name each cluster after its page

Each cluster gets a label that becomes the page’s title pattern, such as “Drain Cleaning in Round Rock, TX”. The other phrases in the cluster go into the template as headings, FAQ questions and body copy, so one page covers the whole cluster naturally. “Clogged drain” becomes a heading. “Hydro jetting” becomes a section with its own short explanation and price. Clusters that share a head term roll up to a hub page, such as “Drain cleaning across the Austin area”, that links to every city page in the set.

Step 7: Choose the patterns you will build

By now you may have several candidate patterns. Score them on the factors that predict whether the pages will be worth publishing. A 1 to 3 scale is enough:

PatternIntent fitData availableBusiness valueRowsTotal
{Service} in {City}3331309
Emergency {Service} in {City}2131306
{Service} cost in {City}1221305
{Service} for {Property type}221245

The scores are hypothetical. The method is what matters: a pattern with perfect intent and no data is not ready, and a pattern with great data but the wrong intent will not match what searchers want. In this example the plain service-and-city pattern goes first. “Cost in city” turns out to be a guide topic, and “emergency” becomes a section in the main template.

Start with one pattern. Once it is live, the second pattern is much easier to judge, and you reuse most of the spreadsheet.

Keyword combinations that should not become pages

This is where programmatic SEO keyword research protects the site. Some combinations look like opportunities in a spreadsheet and turn into liabilities once published.

Google’s spam policies describe doorway abuse, which includes pages aimed at specific regions or cities that funnel visitors to one page, and substantially similar pages that look more like search results than a clear, browsable hierarchy. The same document describes scaled content abuse: many pages made mainly to manipulate rankings, with little value for readers, however they are produced (Google spam policies). Google’s guidance on helpful content asks a related question in plain terms: is the content made mainly to attract visits from search engines (Google: creating helpful content)?

Drop or merge these combinations before they reach a template:

  • Places or audiences you cannot serve. If no technician goes to Kyle, a page that says you serve Kyle is wrong, not ambitious.
  • Near-duplicates. “Plumber Austin” and “plumbers in Austin” are one search. A page for each splits your own signals and gives visitors nothing new.
  • Modifiers that do not change the answer. “Best”, “affordable”, “top rated” and “cheap” are claims, not page types. Put them in your copy if they are true. Do not multiply pages by them.
  • “Near me”, ZIP code and street-level explosions. Unless each ZIP code has different facts that matter to the visitor, ZIP pages are the same page many times over.
  • Neighborhoods that are not separate markets. A city with distinct districts that people search by name may deserve district pages. A small town’s subdivisions rarely do.
  • Intent mismatches. If the results page wants a guide or a comparison, a service page will not fit, however much volume the phrase has.
  • Combinations that make no sense. Snow removal in Miami. Pool opening in a town with no pools. A spreadsheet will generate them without blinking, and a template will publish them.
  • Competitor names you cannot write about factually. “[Competitor] alternative in [city]” pages need verified, current facts about the competitor, and still risk reading as bait.
  • Claims you cannot stand behind. A license in a state where you hold none, an office in a city where you have none.

For the content side of the same problem, see programmatic SEO without thin pages, which covers the swap test and how to handle weak rows once the pages exist.

Worked example: build a keyword matrix in a spreadsheet

Here is the whole process in one sheet. The business is a hypothetical plumbing company north of Austin, Texas. It offers five services and works in eight cities, so the raw pattern has 40 combinations. The goal is a matrix that marks each combination as build, hold, merge or drop, and then becomes the import file.

The formulas work in Google Sheets and recent versions of Excel unless noted. To keep them readable, the input lists are named ranges, which you create with Data > Named ranges in Google Sheets or the Name Box in Excel.

The input lists

Put the inputs on their own tab:

Named rangeContents
ServicesDrain Cleaning, Water Heater Repair, Leak Detection, Sewer Line Repair, Water Softener Installation
Stemsdrain, water heater, leak, sewer, softener (one per service, in the same order, used to match queries)
CitiesAustin, Round Rock, Cedar Park, Georgetown, Pflugerville, Leander, Kyle, Buda
NotOfferedKeys for combinations the business does not serve, written as `Service
IntentService, IntentFitOne row per service with the result of your results-page sample: “service pages” or “guides and retailers”
GscQueries, GscImprA Search Console query export: the query text and its impressions for the last three months

In this example, sewer line repair is not offered in Kyle, Buda, Leander or Georgetown, and water softener installation is not offered in Kyle or Buda. That is six keys in NotOffered. The results-page sample found service pages for four services, and a mix of guides and retailer pages for water softener installation.

Generate every combination

On the matrix tab, put headers in row 1 and start the first combination in row 2. These formulas produce every service and city pair when you fill them down to row 41:

A2  Service:  =INDEX(Services, INT((ROW()-2)/COUNTA(Cities))+1)
B2  City:     =INDEX(Cities, MOD(ROW()-2, COUNTA(Cities))+1)
C2  Stem:     =INDEX(Stems, INT((ROW()-2)/COUNTA(Cities))+1)

Row 2 gives Drain Cleaning in Austin, row 3 Drain Cleaning in Round Rock, and row 10 moves on to Water Heater Repair in Austin. The service index goes up by one every eight rows, and the city index cycles from 1 to 8.

In Google Sheets you can also produce both columns with one formula: =ARRAYFORMULA(SPLIT(TOCOL(Services&"|"&TRANSPOSE(Cities)),"|")). The fill-down version is easier to audit and to extend with more columns, so it is used here.

Build the keyword, title and slug

D2  Keyword:  =LOWER(A2&" "&B2)
E2  Title:    =A2&" in "&B2&", TX"
F2  Slug:     =LOWER(REGEXREPLACE(A2&"-"&B2&"-tx","[^A-Za-z0-9]+","-"))

REGEXREPLACE turns every run of spaces and punctuation into one hyphen, which mirrors how WordPress cleans slugs: “Round Rock” becomes round-rock, and “St. Louis” would become st-louis. In versions of Excel without REGEXREPLACE, use =LOWER(SUBSTITUTE(SUBSTITUTE(A2&"-"&B2&"-tx"," ","-"),".","")) and check the results by eye. If a title reads badly here, it will read badly on the page too, so fix long ones in the sheet.

Add the checks

G2  Offered:      =IF(COUNTIF(NotOffered, A2&"|"&B2)>0, "no", "yes")
H2  Duplicate:    =IF(COUNTIF($F$2:$F$41, F2)>1, "duplicate", "")
I2  Impressions:  =SUMIFS(GscImpr, GscQueries, "*"&C2&"*", GscQueries, "*"&LOWER(B2)&"*")
J2  Intent:       =INDEX(IntentFit, MATCH(A2, IntentService, 0))
  • Offered looks the pair up in the coverage list.
  • Duplicate flags any slug that appears twice. It catches “St Louis” and “St. Louis” in the same list, which clean up to the same URL.
  • Impressions adds up Search Console impressions for every query that contains both the service stem and the city name. It is rough on purpose: “clogged drain round rock” counts toward Drain Cleaning in Round Rock, which is what you want.
  • Intent pulls the result of your results-page sample for that service.

Add the data columns and the decision

Columns K to N hold the facts each page needs: Phone, Areas (neighborhoods covered), Response (typical response time) and Price (a “from” price). Fill them from the business, not from guesses. Then add the decision:

O2  Data complete:  =COUNTBLANK(K2:N2)=0
P2  Decision:       =IF(G2="no","drop",IF(H2="duplicate","drop",IF(J2<>"service pages","merge or skip",IF(O2,"build","hold"))))

The order of the tests matters. Coverage comes first, because a combination you do not serve is dropped whatever its data looks like. Intent comes before data, because there is no point collecting facts for a page type searchers do not want.

Add a summary block anywhere on the sheet:

Build:          =COUNTIF(P2:P41, "build")
Hold:           =COUNTIF(P2:P41, "hold")
Merge or skip:  =COUNTIF(P2:P41, "merge or skip")
Drop:           =COUNTIF(P2:P41, "drop")

What the matrix shows

Here are eight of the 40 rows, with hypothetical impression counts:

ServiceCitySlugOfferedImpressionsIntentData completeDecision
Drain CleaningAustindrain-cleaning-austin-txyes410service pagesTRUEbuild
Drain CleaningLeanderdrain-cleaning-leander-txyes35service pagesTRUEbuild
Water Heater RepairCedar Parkwater-heater-repair-cedar-park-txyes60service pagesTRUEbuild
Leak DetectionKyleleak-detection-kyle-txyes0service pagesFALSEhold
Sewer Line RepairRound Rocksewer-line-repair-round-rock-txyes22service pagesTRUEbuild
Sewer Line RepairBudasewer-line-repair-buda-txno0service pagesFALSEdrop
Water Softener InstallationGeorgetownwater-softener-installation-georgetown-txyes15guides and retailersTRUEmerge or skip
Water Softener InstallationKylewater-softener-installation-kyle-txno0guides and retailersFALSEdrop

The summary for all 40 rows:

  • Drop: 6. The six combinations the business does not serve.
  • Merge or skip: 6. The remaining water softener rows. The results-page sample showed guides and retailer pages, so one water softener guide serves that intent better than six city pages.
  • Hold: 3. Leak detection in Kyle, Buda and Leander, where the business has no confirmed response time yet.
  • Build: 25. Every remaining combination, with complete data.

Two things stand out. First, impressions did not decide anything. Drain cleaning in Leander has a small count and is still a build, because the pattern is validated and the data is complete. Impressions only set the order of the pilot. Second, the 40-row multiplication became 25 pages, and every one of them has a reason to exist that you can explain in one line.

Turn the matrix into an import file

Filter the Decision column to “build”, copy the columns a page needs to a new tab, and export it as CSV in UTF-8:

Service,City,State,Phone,Areas,Response,Price
Drain Cleaning,Austin,TX,512-555-0110,"Hyde Park, Mueller, Travis Heights",Same day,From $129
Drain Cleaning,Leander,TX,512-555-0114,"Crystal Falls, Bryson",Next morning,From $139
Water Heater Repair,Cedar Park,TX,512-555-0112,"Buttercup Creek, Twin Creeks",Same day,From $149

Quote any cell that contains a comma, as the Areas column does here. Check that phone numbers and prices survived the export unchanged, because spreadsheet apps sometimes reformat them. The CSV preparation guide lists the usual export traps.

Keep the matrix tab. When a held row gets its data, or the business adds a city, you change one cell and the decision updates. The CSV is a snapshot. The matrix is the plan.

Where PageForge fits after the research

Everything above works with any way of building pages. If you build in WordPress with PageForge, the research maps directly onto its settings.

Brainstorming columns. The free plugin’s AI Site Planner uses your own AI provider key. Its Local SEO plan quick prompt can suggest page patterns and the CSV columns each page needs. Its documentation says to treat keyword ideas as suggestions to check, because the AI does not crawl your site. Use it to stress-test your matrix, not to replace it.

From matrix to template. Upload the CSV on Data Sources, then set the template’s Title Pattern to {Service} in {City}, {State} and the Slug Pattern to {service}-{city}-{state}. Column names become placeholders, so the cluster label you chose in step 6 becomes the title pattern directly. The pattern docs explain how slugs are cleaned, which matches the slug formula in the matrix.

Catching what the matrix missed. On the Generate screen, Check and preview reads the file and writes nothing. For the 25-row file it would report something like this:

Ready: 25 new, 0 to update, 0 skipped.
First page: “Drain Cleaning in Austin, TX” at /drain-cleaning-austin-tx/

If two rows clean up to the same URL, it skips the second with a message such as Same URL as row 3., and a title placeholder with no matching column stops the run. What it cannot do is notice that two different URLs serve the same search. The quality checklist says so plainly, which is why the clustering in step 6 stays your job.

The free plugin runs up to 100 rows per run, so a 25-row pilot fits in one run. Larger sets are split into files of up to 100 rows each, and Pro has no per-run row limit.

Common programmatic SEO keyword research mistakes

  • Starting from a tool export. The tool decides the scope, and the scope fills up with places you do not serve and phrasings that mean the same thing.
  • Multiplying every modifier. Services × cities × “emergency” × “best” is not a strategy. Each added modifier should change the answer, or it stays in the copy.
  • Treating volume as truth either way. Zero long-tail volume is not zero demand, and high head-term volume is not a forecast. Validate the pattern with a sample and your own data.
  • Clustering by wording alone. Two phrases that look different can share one results page, and two that look alike can need separate pages. Check overlap on the hard pairs.
  • Building before the data exists. A row with a blank response time and no local detail becomes a page that says nothing. Hold it until the facts arrive.
  • Forgetting the structure around the pages. Generated pages need a hub and links between related pages to be found and used. Plan the hub while you plan the pattern. See internal linking for programmatic SEO and hub pages with the sitemap block.
  • Changing slugs after launch. A new URL pattern on live pages means new pages and a redirect for every old URL. Settle the slug format in the matrix.
  • Launching everything at once. A pilot of 20 to 30 pages shows you whether the template and data hold up before the mistakes are multiplied by 200.

FAQ

How many keywords do I need before starting programmatic SEO?

You need one validated pattern and enough rows with real data to make a pilot worthwhile, usually somewhere between 10 and 50. The number of keywords matters less than the consistency of the intent and the quality of the facts behind each row. A pattern with 20 strong rows is a better start than one with 500 rows that share the same three sentences.

Can I trust search volume for long-tail combinations?

Not as a forecast. Long-tail volume in keyword tools is an estimate, often rounded or grouped with close variants, and many real combinations show zero. Use volume to compare items within your list, then validate the pattern with Search Console impressions, the results page and your own lead data. Treat any traffic projection built from long-tail volume with suspicion.

What is a head term in programmatic SEO?

A head term is the part of the pattern that stays the same on every page: the service, product, software category or course people are looking for. Modifiers such as city, industry or integration vary from page to page. A good head term has a stable intent, a page you can actually deliver, and modifiers that change the answer rather than only the wording.

How do I know if two keywords need separate pages?

Compare their results pages. If the top 10 results for both queries share several URLs, a common threshold is three or more, one page can serve both. If they share almost none, searchers want different things and separate pages make sense. Check a few sample locations rather than every row, because the relationship between two phrasings is usually the same across a pattern.

Should I create pages for “near me” keywords?

No. “Near me” searches depend on where the searcher is standing, so there is no single page to build for them. Your location pages, your business profile and an accurate service area work together to serve those searches. Create pages for the places you actually serve, with facts specific to each, and leave “near me” out of titles and URLs.

How many pages should the first batch have?

Start with a batch small enough to review properly, typically 20 to 30 pages, chosen from rows that already show some demand. Read every page in the first batch, check the metadata and layout, and watch how the pages are indexed and used for a few weeks. Expand once the template and data hold up, and fix the template first if they do not.

Are programmatic pages against Google’s guidelines?

Not by themselves. Google’s spam policies target pages made mainly to manipulate rankings with little value for visitors, including scaled content abuse and doorway pages, whatever method produced them. A page built from a template is fine when it gives each visitor specific, accurate information. The keyword research step is where you remove combinations that would only exist to catch a search.

Which tools do I need for programmatic SEO keyword research?

A spreadsheet does most of the work. Add Search Console for real query data, a browser for checking results pages and autocomplete, and a keyword tool for comparing relative interest if you already have one. Your own business data, such as leads by service and town, is often the most useful input of all. The matrix in this guide needs nothing more than formulas and named ranges.

Conclusion

Good programmatic SEO keyword research ends with fewer pages than the multiplication suggests, and a reason for every one of them. Find a head term the business can deliver, collect modifiers from where you actually work, validate the pattern with a sample instead of trusting long-tail volumes, and let coverage, intent and data cut the set down. Keep it all in one matrix so the decisions stay visible and easy to change.

Your next step: build the matrix for one pattern, mark every row, and export the “build” rows as a CSV. If you are building local pages, the walkthrough on creating location pages in WordPress takes that CSV the rest of the way.

Programmatic SEO notes, once a month

New guides, templates and PageForge releases. No spam, unsubscribe in one click.

Build the template once. Let the data do the rest.

PageForge is free on WordPress.org. Upload a CSV, preview the first page, and generate your first 100 pages today.

↑↓ to move↵ to open