What is programmatic SEO? How it works, with real examples
Learn what programmatic SEO is, how the pipeline works step by step, what Wise, Zapier and Tripadvisor publish, and when you should not build it.
Maciej WiatrRead as markdownWhat is programmatic SEO? It is a way of building many search pages from one template and one dataset, where each page targets one variation of a repeating keyword pattern. Wise’s currency pages, Zapier’s integration pages and Tripadvisor’s “things to do in” pages are the standard examples. This post covers how the pipeline works, what those three sites actually publish, and when you should not build this at all. One rule applies throughout: every page needs data that no other page on the site has.
What is programmatic SEO?
Ahrefs defines it as “the creation of keyword-targeted pages in an automatic (or near automatic) way.” Each page comes from one row of a dataset.
The method has three parts:
- A keyword pattern. A head term plus a modifier that repeats. Backlinko’s example is “convert [currency] to [currency].”
- A dataset. One row per modifier, with the facts that page will show.
- A template. The layout that turns a row into a page.
What does programmatic SEO mean?
“Programmatic” means code generates the pages from data, so nobody writes them one by one. “SEO” means each page is built to match a real long-tail query that people search.
What are programmatic SEO pages?
They are the output: one page per dataset row. Three examples are Wise’s /currency-converter/usd-to-eur, Zapier’s “Google Sheets + Slack” integration page, and Tripadvisor’s “Things to do in Orlando.”
How is programmatic SEO different from traditional SEO?
Traditional SEO means writing unique pages by hand, often for competitive head terms. Programmatic SEO generates templated pages for many low-volume long-tail terms. Mangools points out that it needs database and engineering skills, and that page quality comes from the data more than the writing.
Siege Media gives two tests:
- Volume. If you have fewer than 100 target keywords, use traditional content instead. Siege doesn’t source that number, so treat it as a rule of thumb.
- Result type. Programmatic pages fit search results that are full of directories, list pages and landing pages. They don’t fit results dominated by in-depth editorial articles.
How does programmatic SEO work?
These are the steps, in the order you would run them.
1. Choose a head term and modifiers
Backlinko says to check that search intent stays the same across modifiers. Few guides explain how to check it. My method: search 10 to 20 modifiers from your list and look at what ranks. If “best CRM for startups” returns comparison tables and “best CRM for nonprofits” returns vendor pages and discounts, the two queries have different intent. One template won’t serve both.
2. Gather the dataset
Backlinko ranks data sources by risk:
- Proprietary data is the safest.
- Public data needs added value on top.
- Scraped data is the riskiest.
Ahrefs adds that you should avoid the same data sources your competitors use. If everyone builds on the same public feed, the resulting pages all look alike. The data hub covers sourcing and enrichment in more depth.
3. Build the template
Ahrefs says to decide three things for every section of the template:
- which parts stay fixed,
- which parts come from the data,
- which parts are written for each page.
On length, Backlinko recommends 500 to 1,000+ words per page. Google’s policies set no word count, and the pages at Wise and Zapier are mostly data and tools. I treat word count as optional and data as required.
4. Test a small batch
Backlinko suggests building 3 to 5 pages by hand. Ahrefs suggests testing about 10 rows before you scale. Either way, read the test pages as a user would before you generate thousands more.
5. Generate in batches
Backlinko suggests no-code tools for up to about 100 pages. For 1,000+ pages it suggests a custom app (it names Next.js, Vercel and PostgreSQL), rolled out in stages. If you build with Next.js, my guide to ISR, 404s and sitemaps at scale covers the rendering side.
6. Get the pages indexed
Submit sitemaps in Google Search Console and link the pages internally. Google’s large sitemaps documentation gives the limits: up to 50,000 loc entries per sitemap index, and up to 500 sitemap index files per site.
Don’t expect full indexation. In one vendor-published case study of 512 pages, these were the shares indexed within 90 days:
- glossary pages: 94%
- city pages: 89%
- integration combination pages: 73%
Siege Media expects indexing and traction to take 3 to 6 months.
7. Measure, then prune
Ahrefs says to check the page folder after about 3 months, then improve, redirect or remove weak pages. Backlinko lists four metrics to watch: indexation rate, crawl stats, how traffic spreads across pages, and cannibalization. I would track all four per template, because a sitewide average hides a template that is failing.
What makes a page’s data unique
This is the part most guides skip. Zapier’s own write-up says the data unique to each integration page is mostly the apps’ triggers and actions, which change often. That is why a “Google Sheets + Trello” page differs from a “Google Sheets + Slack” page in ways a user cares about.
The same theStacc case study found that pages with at least 3 data points from the dataset got indexed and ranked, while pages with only 1 or 2 struggled. It is one anonymized site, so treat it as a signal and not a law.
The test I would run on every row before publishing: compare the page with its closest sibling. Count the facts that differ, ignoring the swapped name. If fewer than three differ, hold the page back until you have more data for it.
Real examples of programmatic SEO and their page patterns
Page counts for these sites vary by 10x or more between sources. Most are third-party tool estimates, so the table gives ranges and the method behind each number.
| Site | Pattern | Unique data per page | Page count (method) |
|---|---|---|---|
| Wise | “[currency] to [currency]” | Live rate, converter, historical chart | 537,135 indexed, whole site (Omnius, Mar 2026) to ~8.5M converter pages (Backlinko, sitemap count, Jan 2026) |
| Zapier | “[app] + [app] integrations” | Triggers and actions | 508K (Mangools, Oct 2024) to 590K+ (Backlinko, Jan 2026) |
| Tripadvisor | “Things to do in / Best [X] in [city]” | Reviews, ratings, photos, attraction links | 71M (Mangools, unsourced) to 700M+ indexed (Gupta, unsourced) |
Traffic estimates are just as spread out. Wise ranges from about 6.4M organic visits a month (Ahrefs, 2021) to “100+ million visits every month” (Backlinko). Backlinko puts Zapier’s /apps/ folder at 610K+ organic visits a month, and counts almost 100K keywords containing “things to do in.” It also reports Zillow at 7.5M recently-sold pages and about 243M organic visits a month.
The examples include failures too. Backlinko reports that G2 fell from almost 12M monthly visits in 2021 to “less than 1 million,” and ties the drop to Google updates, AI Overviews and Reddit’s presence in results.
Is SEO dead now with AI?
No, but clicks are shrinking for some page types. In February 2026, Tripadvisor’s CEO reported “ongoing declines in flyby visitors to our site,” and linked them to AI Overviews, according to Skift. The CFO expects SEO to bring “less than 10%” of Experiences bookings value by the end of 2026. When you model returns for a new template, assume fewer clicks per ranking than older case studies show.
Is programmatic SEO bad?
Not by definition. Google’s spam policies target pages “generated for the primary purpose of manipulating search rankings,” however they are made. Two parts of the policies hit templated sites directly:
- Doorway abuse. Google’s own example is city- or region-targeted pages that funnel users to one destination. That is the classic programmatic failure.
- Scaled content abuse. Since the August 2026 update to the policy page, this also covers attempts to “manipulate generative AI responses in Google Search.”
Google’s AI content guidance, updated in October 2026, asks you to fact-check titles, meta descriptions, structured data and alt text, and to consider telling users how automation was used. Advice from 2024 to fill templates with ChatGPT text no longer holds up.
John Mueller has been blunt about it. In 2023 he called programmatic SEO “often a fancy banner for spam.” In September 2026, replying about one specific site, he said Google’s systems may have “lost faith in your site providing good value to users based on the old pages,” and that recovery “tends to take time and significant effort” (TechWyse). His point is that weak templated pages can drag down the whole site, including the pages that were fine. My post on what Google’s 2026 updates removed covers the update timeline in detail.
When is programmatic SEO the wrong approach?
Don’t do it if:
- You have no proprietary or enriched data. If your rows hold the same facts as everyone else’s, your pages will too (Backlinko, Ahrefs).
- The keyword pattern is too small. Siege Media’s line is about 100 keywords.
- Intent changes across modifiers. Run the sample search test from step 1.
- The results favour editorial content. Templates lose to in-depth articles there (Siege Media).
- Pages would only swap a city name. That is a doorway page by Google’s own example.
- You can’t keep the data fresh. Zapier’s triggers and Wise’s rates have value because they are current.
- The site has little authority. Backlinko flags this as a risk, and Mueller’s comment shows the downside reaches beyond the new pages.
If two or more of these apply, I would write 20 good pages by hand instead. If none apply, start with 10 rows, check that each passes the three-fact test, and scale only the templates that get indexed.