Summary
Quick Answer
Programmatic SEO still works in 2026, but only when each generated page answers a question a real person asked. Google does not ban templated pages. It bans pages generated mainly to rank, which it calls scaled content abuse, and pages that funnel visitors somewhere more useful, which it calls doorway abuse. The difference is whether the page carries information that only you have.
- Google ran four spam updates in 2026 alone, in March, June, August and September.
- Scaled content abuse is about intent and value, not about word count or whether a human typed it.
- Location pages that differ only by city name are the single most common way Canadian service businesses trip the doorway policy.
- The viable programmatic page sets are the ones sitting on proprietary data: your inventory, your pricing, your coverage area, your measurements.
- If you already published hundreds of thin pages, consolidation beats deletion in most cases.

Table of contents
- What is programmatic SEO?
- Does programmatic SEO still work in 2026?
- What does Google actually prohibit?
- Four tests to run before you generate a single page
- Where programmatic pages earn their keep
- How to build a page set that holds up
- Common mistakes
- What to do if you already published hundreds of thin pages
- Frequently asked questions
What is programmatic SEO?
Programmatic SEO is the practice of building many pages from one template and one structured data source, so each page targets a predictable variation of a search query. One template plus 200 rows produces 200 pages.
You have used dozens of these pages this week without noticing. Every flight route page, every “homes for sale in” page, every software integration directory and every currency conversion page is programmatic. It is a legitimate and old technique. The reason it has a reputation problem is that the same template that produces a genuinely useful page set also produces a worthless one at identical cost, and the worthless version is faster to ship.
Programmatic SEO is not the same as AI content
These get conflated constantly. Programmatic SEO is about structure: a repeated page shape filled from a dataset. AI content is about production method: who or what wrote the prose. You can build a programmatic page set with zero AI involvement, and you can write a single hand-crafted essay entirely with AI. Google’s policies address both, but separately, and understanding which one you are doing changes which policy applies to you. If you want the wider map of how these disciplines fit together, our guide to the types of SEO and the order to tackle them puts programmatic work in context.
Does programmatic SEO still work in 2026?
Yes, with a narrower definition of “works” than in 2021. A programmatic page set that holds rankings in 2026 has three properties: each page contains data the searcher cannot get elsewhere in one place, the page set maps to real query demand rather than every possible permutation, and the pages are navigable as a hierarchy rather than existing only as sitemap entries.
What stopped working is volume for its own sake. Google shipped spam updates in March, June, August and September 2026, plus core updates in March, May and November. That cadence matters more than any single update, because it means the window between publishing a weak page set and getting evaluated is now measured in weeks. The old game of shipping 5,000 pages and harvesting traffic for eighteen months before the hammer fell no longer has eighteen months in it.
There is a second pressure that most programmatic advice has not caught up with. AI answer engines summarise rather than list. A page that exists to be one of ten blue links for “plumber in Airdrie” is competing for a slot that may not be shown. A page that holds a specific, citable fact has a different job and a better chance of being the source an answer engine quotes. We cover the mechanics of that in our answer engine optimization playbook. For most Canadian businesses the practical conclusion is the same either way: fewer pages, more substance per page, and a clear SEO and AI search program behind them rather than a one-off page dump.
What does Google actually prohibit?
Two policies in Google’s spam documentation govern programmatic page sets. Neither mentions templates, automation or page counts as the problem. Both are about purpose.
Scaled content abuse
Google defines it as the case where “many pages are generated for the primary purpose of manipulating search rankings” rather than helping users. The examples Google lists are worth reading literally:
- Using generative AI tools or similar tools to generate many pages without adding value for users
- Scraping feeds, search results or other content to generate many pages, including through automated transformations such as synonymizing, translating or other obfuscation techniques
- Stitching or combining content from different web pages without adding value
- Creating multiple sites with the intent of hiding the scaled nature of the content
- Creating many pages where the content makes little or no sense to a reader but contains search keywords
Note the phrase that repeats: “without adding value.” The policy is not anti-automation. A page assembled by a script from your own licensed dataset adds value. A page assembled by a script from someone else’s dataset with the adjectives swapped does not.
One Canadian wrinkle. Translation appears in that list as an obfuscation technique, which worries bilingual businesses. The distinction is purpose. Machine-translating 400 English pages into French to double your indexed page count is the behaviour the policy targets. Publishing a properly localised French version of your site because Quebec language rules require it, or because a third of your customers read French, is a service decision. Have it translated properly by a person, mark it up with correct hreflang, and it is not what Google is describing.
Doorway abuse
This is the policy that actually catches Canadian service businesses, and almost nobody reads it before building location pages. Google defines doorway abuse as the case where “sites or pages are created to rank for specific, similar search queries” and “lead users to intermediate pages that aren’t as useful as the final destination.” Its examples include:
- Having multiple websites with slight variations to the URL and home page to maximize their reach for any specific query
- Having multiple domain names or pages targeted at specific regions or cities that funnel users to one page
- Generating pages to funnel visitors into the actual usable or relevant portion of a site
- Creating substantially similar pages that are closer to search results than a clearly defined, browseable hierarchy
Read the second one again. “Pages targeted at specific regions or cities that funnel users to one page” is a precise description of the standard service area page set sold across Canada: forty pages for forty towns, each one a find-and-replace of the last, every call to action pointing at the same contact form. If your Calgary page and your Okotoks page differ only in the three places the town name appears, that is the policy, quoted.
This does not mean service area pages are forbidden. It means a service area page has to be about the area. Our work with trades and contractors is built on the version that holds: real coverage boundaries, response times from that depot, permit and inspection details for that municipality, the jobs actually completed there, local pricing conditions. That page is not a doorway. It is the final destination.
The two neighbouring policies worth knowing
Two more show up in programmatic projects. Thin affiliation covers publishing affiliate content “where the product descriptions and reviews are copied directly from the original merchant without any original content or added value,” which is the fate of most programmatic product comparison sites. Expired domain abuse covers buying a lapsed domain and repurposing it “primarily to manipulate search rankings by hosting content that provides little to no value to users,” which is what happens when a programmatic project is bolted onto a purchased domain to borrow its history.
Four tests to run before you generate a single page
Run every proposed page set through these four questions. If you cannot answer all four, the page set is not ready, and no amount of word count will fix it.
- The unique data test. What is on this page that is not on any competitor’s page and not available from a public API? If the answer is “the city name,” stop.
- The demand test. Does anyone search this variation? Pull the query data before you build. Most permutation sets have real demand for the first fifteen rows and none for the remaining four hundred.
- The destination test. If a searcher lands here from Google, can they finish their task on this page? Or does the page exist to push them to a different page? The second case is a doorway by Google’s own definition.
- The hierarchy test. Can a human browse to this page from your navigation in three clicks or fewer, through a category structure that makes sense? If the only path in is the sitemap, Google’s “closer to search results than a clearly defined, browseable hierarchy” language applies to you.
The demand test is the one that saves the most money. Businesses routinely commission 500 page sets where 40 pages would capture 90 percent of the available search volume, then pay to maintain 460 pages that will never be visited. Build the 40. Add rows later when the data says to.
Where programmatic pages earn their keep
Programmatic SEO pays off when you hold a dataset nobody else can copy. It fails when the dataset is public. Here is how that breaks down across the business types we see most in Canada.
| Business type | Page set that works | Page set that fails | Why |
|---|---|---|---|
| Trades and contractors | Service plus municipality, carrying permit rules, inspection timelines and completed jobs for that municipality | Service plus every town within 300 km | Coverage data is yours. Town names are not. |
| Real estate | Neighbourhood pages with current listing counts, days on market and school catchments | “Best realtor in” pages per city | Live MLS data refreshes the page. Opinion copy does not. |
| Short-term rental operators | Property pages with real availability, pricing and local guides written from actually staying there | Generic “things to do in” pages | Availability is proprietary and changes daily. |
| SaaS and technology | Integration pages, each documenting the actual field mapping and setup steps | “Alternative to competitor” pages generated from feature lists | Your integration docs exist nowhere else. |
| Professional services | Nothing programmatic. Twelve hand-written pages beat 200 generated ones. | Practice area plus city matrices | The buying decision rests on trust signals a template cannot carry. |
| E-commerce | Faceted category pages with real inventory, filtered counts and in-stock status | Pages for filter combinations with zero products | Empty results pages are the textbook thin page. |
The pattern across every row: the page sets that work are the ones where the data changes without anyone writing anything. That is the tell. If your programmatic pages are static the day after launch, they are content, not data, and they will be judged as content.
How to build a page set that holds up
Seven steps, in order. Skipping step two is the most expensive mistake available.
- Find the dataset first, not the keyword. Start from what you actually hold: inventory, coverage, pricing, measurements, availability, case history. If the dataset does not exist yet, the real project is building the dataset.
- Validate demand row by row. Export your Search Console queries, pull a keyword tool, and mark which rows have searches. Cut the rest. Ship the survivors.
- Design one page by hand, properly. Write the best possible single page for the strongest row. No template, no variables. If the hand-written version is not compelling, the templated version will be worse 200 times over.
- Extract the template from that page. Work backwards. Mark which blocks come from data, which are genuinely static, and which need human input per page. Expect at least one human-input block. A page set with zero human input blocks is the thing Google describes.
- Build the browseable hierarchy. Category index, subcategory index, then detail pages, all linked in the navigation and in body copy. This is the step that converts a doorway set into a site section.
- Ship in batches of 20 to 50 and watch. Check indexation, impressions and engaged sessions per batch. If batch one earns nothing in six weeks, batch two will not either. Fix the template, not the volume.
- Put a refresh cycle in writing. Decide who updates the dataset and how often, before launch. Pages that have not changed in a year are the ones that quietly lose rankings after a core update.

What a page set costs to run
Budget in hours, not pages, because the per-page marginal cost is close to zero and that is exactly what makes these projects go wrong. A realistic split for a 40-page set looks like this, and it is worth noting how little of it is the build:
- Dataset assembly and cleanup. Usually the largest line. If your coverage areas, pricing tiers or job history live in someone’s head, this is where that gets fixed.
- Query validation. A few hours, and it decides the size of everything downstream.
- The hand-written reference page. Treat it as a full page build, because it is one.
- Template and generation. Smaller than clients expect, often the cheapest line in the project.
- Per-page human input. The line that gets cut first and should never be. Even two original sentences per page changes the character of the set.
- Ongoing refresh. A standing monthly or quarterly commitment, which is why this work belongs on a website care plan rather than in a one-off project.
If a proposal quotes you a flat per-page price with no dataset line and no per-page human input line, you are buying the version Google penalises.
Common mistakes
- Building the full permutation matrix. Services times cities times modifiers produces an impressive number and a thin site. Demand is not evenly distributed and your page set should not be either.
- Treating word count as the fix. Padding a thin page to 1,200 words makes it a longer thin page. Google’s policy language is about value, not length.
- Orphaning the set. Pages reachable only through the XML sitemap match Google’s description of pages “closer to search results than a clearly defined, browseable hierarchy.”
- One call to action for 300 pages. If every page funnels to the same form, you have built the funnel Google’s doorway examples describe. Give each page something to do on the page.
- Reusing the same FAQ block everywhere. Identical FAQ markup across hundreds of URLs is the easiest duplicate signal to spot, and it wastes the one block that could have carried per-page specifics.
- No canonical discipline on faceted pages. Filter combinations generate near-infinite URLs. Decide which are indexable before launch, not after Search Console reports 40,000 pages.
- Publishing all of it at once. A thousand new URLs in a day is a pattern. Batched launches also let you learn something before you have spent the whole budget.
- Confusing indexation with success. Indexed and earning nothing is not a win. Measure engaged sessions and conversions per page set, not page count.
What to do if you already published hundreds of thin pages
Do not start by deleting. Start by sorting, because a page earning twelve visits a month is still a page earning twelve visits a month, and the pages worth saving are usually not the ones you would guess.
- Export twelve months of Search Console data at the page level, with clicks and impressions.
- Sort into four buckets. Earning clicks. Earning impressions but no clicks. Indexed but earning nothing. Not indexed.
- Keep and improve the first bucket. These pages have proven demand. Add the per-page substance they never had.
- Fix titles and intros on the second bucket. Impressions without clicks is usually a snippet problem, not a content problem.
- Consolidate the third bucket. Merge groups of near-identical pages into one strong page and 301 the rest to it. Three weak city pages become one regional page that actually ranks.
- Remove the fourth bucket. Pages Google never indexed are earning nothing and cost crawl budget. A 410 or a 301 to the nearest relevant page, consistently applied.
- Rebuild the internal linking after the merges, so no 301 chains and no orphans remain.
Consolidation is unglamorous and it is reliably the highest return work available on a bloated site. We have seen it beat the next five new articles on the same site, which is also the argument for publishing fewer and better pages in the first place. Our note on how many blog posts a month actually makes sense makes the same case for editorial content.
Frequently asked questions
Is programmatic SEO against Google’s guidelines?
No. Google’s spam policies never mention templates or automation as the problem. They prohibit generating many pages “for the primary purpose of manipulating search rankings” and creating pages that funnel users somewhere more useful. A templated page set built on data a searcher needs is compliant. The same template filled with nothing is not.
Will Google penalise AI-generated pages?
Not for being AI-generated. Google’s scaled content abuse policy names “using generative AI tools or other similar tools to generate many pages without adding value for users.” The qualifier is doing the work. Production method is not the test. Whether the page adds value is.
How many location pages can I safely publish?
As many as you have genuinely distinct coverage information for. If you can write three paragraphs about working in that municipality that you could not write about the next one, the page is justified. If you cannot, merge it into a regional page. The number is set by your operations, not by a safe limit.
Do my pages need to be different lengths?
Length is not the signal. Varying word counts on pages that say the same thing changes nothing. What matters is whether the facts on each page differ. A 400-word page with five numbers only you hold outperforms a 1,500-word page of restated generalities.
How do I know if I have a doorway page problem?
Open two pages from the set side by side and read them. If a customer could not tell which one they were on without looking at the heading, and if both push to the same form rather than completing a task, you have one. The second check: can you reach the page from your own navigation without using search or the sitemap?
Does programmatic SEO help with AI search and AI Overviews?
Only the data-backed version does. Answer engines quote specific facts, so a page carrying a number, a boundary, a timeline or a price that exists nowhere else is a citation candidate. A page of restated generalities has nothing to quote. In practice this pushes programmatic strategy in the same direction compliance does, toward fewer pages with more proprietary substance.
Should a small Canadian service business do programmatic SEO at all?
Usually not as a first move. If you have fewer than about fifteen pages, your return is higher on a strong services page, a properly built Google Business Profile and three genuinely useful articles. Programmatic work earns its place once you have a dataset that grows on its own.
The bottom line
Programmatic SEO is a distribution method for data you already own. Treated that way it is one of the most efficient channels available, because the dataset does the differentiating and the template just presents it. Treated as a way to manufacture pages, it now fails faster than it used to, and the cleanup costs more than the build did. Before you generate anything, answer the first of the four tests honestly: what is on this page that nobody else has?
Want a page set that holds its rankings?
We build programmatic page sets the way they should be built: dataset first, demand validated row by row, a browseable hierarchy, and a refresh cycle in writing. If you have a bloated set that needs consolidating instead, that is usually the faster win and we will tell you so. Start with the intake form and tell us what data you hold.



