Start here
- Programmatic SEO isn’t banned — thin programmatic SEO is. Google formalized this in March 2024 as the “scaled content abuse” policy, and it’s enforced sitewide, not page-by-page.
- The public data (Zapier’s own case, plus several third-party reconstructions) doesn’t even agree on Zapier’s current traffic within a factor of three — which is itself the most useful data point in this article.
- In the one detailed, page-level dataset available publicly, three template types built on identical infrastructure produced wildly different outcomes: 94% vs. 73% indexing, 51% vs. 28% page-1 rate.
- The distribution of results is a power law, not a plateau: the top 10% of pages typically carry roughly 40%+ of the traffic. Plan for that from day one, not month twelve.
The ceiling nobody puts in the deck
establishedBefore the tactics: there is a hard governance layer sitting on top of everything else in this article, and it can erase the whole project.
In March 2024, Google formally renamed what it used to call “spammy automatically generated content” to scaled content abuse, and broadened the definition. The current wording, from Google’s own spam policy documentation, is that it covers cases where many pages are generated for the primary purpose of manipulating search rankings and not helping users — and critically, the policy is explicit that this applies regardless of the method: AI, no-code automation, or a team of human freelancers cranking out near-identical pages all fall under the same rule.
That “method-agnostic” framing is the part most programmatic SEO guides gloss over. It means “we had a human editor review it” is not, on its own, a defense. The defense is whether the page would still be worth publishing if you could only make one of it.
Google’s own John Mueller has reportedly pushed back hard on the category in public commentary that SEO teams frequently cite — describing the “programmatic SEO” label itself as something that too often just dresses up spam, according to secondary reporting from agencies tracking his public statements. I can’t independently verify the exact wording of an unrecorded comment, so treat the framing rather than the phrasing as the reliable part: Google’s team is openly skeptical of the category by name, which should reset how much credit you give a “the algorithm will figure out my pages are good” plan.
Three conditions tend to show up together in a real violation: the page is one of many near-identical pages, its primary purpose is ranking rather than serving a specific user need, and it adds little or no value you couldn’t get from any of the other pages in the set. High page count alone is not the trigger — Zillow’s individually-sourced property listings and a ticketing site’s boilerplate city pages get treated very differently by the same policy, because the underlying data differs.
Zapier’s traffic numbers don’t agree with each other — and that’s the real lesson
establishedZapier is the example every programmatic SEO article opens with. It’s a genuinely good example. It’s also a case study in how unreliable secondhand SEO reporting gets, even for the most-cited case in the niche.
Zapier connects app-integration demand to landing pages using two page types: one profile page per app, and one combination page per pair of apps (URL pattern /apps/[app-a]/integrations/[app-b]), each describing the specific triggers and actions available between that pair. That structure is confirmed directly on Zapier’s own blog, and the pairwise-page count (50,000+ live integration pages, drawn from a catalogue of several thousand supported apps) shows up consistently across independent write-ups.
What doesn’t show up consistently is the traffic number attached to it. Here’s what I found simply comparing sources published within the same 18-month window:
Reported Zapier organic traffic, by source
Compiled from publicly available third-party SEO write-ups, not from Zapier’s own disclosures — Zapier does not publish organic traffic figures directly.
| Source | Reported figure | Published |
|---|---|---|
| Practical Programmatic | ~2.6M / month | 2025 |
| MADX Digital | ~1.6M/mo blog + ~16% via pSEO pages | 2026 |
| GrackerAI | ~5.8M / month | 2025 |
| GrackerAI (case study page) | 5–6M / month | 2026 |
| Salt Agency | “quadrupled,” no absolute figure given for current state | 2024 |
| ViewEngine AI | 9M+ / month | 2026 |
That’s not a rounding error — it’s more than a 3x spread across sources published in roughly the same period, none of which cite Zapier’s own disclosed numbers because Zapier doesn’t publish them. This is the least-discussed risk in programmatic SEO content: almost none of the “proof” articles in this niche are using primary data. They’re estimates built on third-party traffic-estimation tools (which themselves disagree with each other), repeated and rounded until they read like established fact.
Treat that as the default state of any pSEO case study you read, including the ones in this article that I couldn’t independently verify — and treat it as a specific reason to build your own tracking (Search Console + a consistent attribution model) from day one rather than benchmarking yourself against someone else’s estimated numbers.
What one page-level dataset actually showed
probableAlmost every pSEO case study reports an aggregate number and stops. One recent published account broke a single 512-page, 18-month SaaS program down by template family and by individual page performance, which is rare enough to be worth examining closely — with the caveat below.
The dataset below comes from a single agency’s self-published, anonymized client case (theStacc, May 2026). It’s a vendor publishing its own results to sell its own product — the same conflict of interest as every case study in this space, including several cited elsewhere in this article. I’m using it because the internal structure of the data (how three template types diverged on identical infrastructure) is more useful than the topline ROI number, and because it’s unusually granular for a public source. Treat the pattern as directionally credible and the specific dollar figures as unverified.
Same site, same internal linking, three template families
Source: theStacc, 512-page B2B SaaS programmatic build, 18-month tracked period. Single client, self-reported.
| Template type | Pages | Indexed in 90d | Page-1 rate | Decay rate |
|---|---|---|---|---|
| Glossary / definitional | 186 | 94% | 51% | 7% |
| Location / city | 198 | 89% | 34% | 12% |
| Integration combo | 128 | 73% | 28% | 19% |
Fig. 1 — Indexing and page-1 rate by template family, one 512-page program. Lighter bar = indexed within 90 days, darker bar = reached page 1. Source: theStacc, 2026.
The gap between glossary pages (94% indexed, 51% page-1) and integration combo pages (73% indexed, 28% page-1) is the real headline, and it holds a pattern that shows up across the other sources I checked too: informational, single-entity pages index and rank more reliably than combinatorial pages, because a definition of a stable concept doesn’t depend on two independent things both having search demand. A “Tool A + Tool B” page is only as strong as its weaker half.
Traffic distribution across the same 512-page set
Sorted by month-18 click volume and bucketed into quartiles. Source: theStacc.
| Cohort | Pages | Share of clicks |
|---|---|---|
| Top 10% | 51 | 42% |
| Next 20% | 102 | 31% |
| Next 40% | 205 | 22% |
| Bottom 30% | 154 | 5% |
That’s a power law, not a bell curve, and it’s consistent with how organic traffic distributes almost everywhere at scale — the specific split will vary by site, but the shape won’t. If your plan treats every one of your 10,000 pages as an equally valuable asset, you’re going to spend equal maintenance budget on the 30% of pages producing 5% of the return. Build the review-and-cull step into the plan before you launch, not as a rescue mission in month twelve.
A page-economics framework, with my assumptions shown
my model, not a data sourceI’m not going to hand you a fabricated “average cost per pSEO page” statistic — nobody has good aggregate data on that across the industry, and anyone who states one with confidence is guessing. What I can give you is the framework I actually use to decide whether a template is worth building, with every assumption labeled so you can swap in your own numbers.
Break-even framework for a programmatic page template
Original model. Every input below is a variable you should replace with your own data — none of these are claimed as industry averages.
| Variable | What it represents |
|---|---|
| Fully-loaded cost per page (C) | Data sourcing + template engineering (amortized) + generation + human review + hosting/monitoring share |
| Index probability (I) | Use your own template family’s historical rate, or the low end of a comparable public case until you have 90 days of your own data |
| Page-1 conditional probability (P) | Conditional on indexing — i.e., of the pages that index, what share reach page 1 |
| Value per page-1 page (V) | (monthly organic clicks at that position) × (your visitor→signup rate) × (LTV or ACV-weighted value per signup) |
Expected value per page reduces to EV = C × I × P × V − C. The variable growth-stage teams underweight is P — the conditional page-1 rate — because it’s tempting to only track raw indexing. The dataset above shows exactly why that’s a mistake: integration combo pages indexed at 73% but only converted 28% of indexed pages to page 1, versus glossary pages converting 51/94 = 54% of indexed pages. A template can look “fine” on indexing and still be a bad investment on the number that actually pays your bills.
The practical use of this framework isn’t to produce a precise dollar figure — it’s to force you to write down I, P, and V as explicit numbers before you commit to 2,000 pages of a template, instead of discovering them by publishing all 2,000 pages first.
My unpopular take: most 10,000-page plans are a vanity metric
Here’s my take, based on the pattern across every dataset above, not a claim that everyone else has missed something obvious: the page count in most programmatic SEO pitches is optimizing for the wrong audience. A 10,000-page number is compelling in a board deck. It is not, on its own, correlated with the 40%+ share of value that — per the only granular dataset available — concentrates in roughly the top tenth of pages.
The programs that actually work in the sources I reviewed all shared a smaller-than-advertised starting point: 100–200 pages in the first phase, a hard cap on weekly publishing velocity (the one detailed case study capped at 25–30/week and saw indexing collapse from 87% to 41% the one week they tried 80), and an explicit review-and-cull cycle before scaling the template further. “10,000 pages” is a target for month eighteen, if the first 200 pages prove the template works. It is a bad target for month one.
Counter-argument, for fairness: Zapier’s own scale — tens of thousands of pages, built over years against a catalogue that grows on its own as new apps integrate — shows the model does work at real scale eventually. The disagreement isn’t about whether scale works. It’s about whether you should build the scale before or after you’ve proven the unit economics on a template.
Early in working with pSEO builds, I greenlit a combinatorial template — the “Tool A + Tool B” pattern — without setting a search-volume floor on either side of the pair. The logic seemed sound: more combinations, more long-tail coverage. What actually happened matches the integration-combo numbers above almost exactly: a large share of the low-volume pairings never indexed, and the ones that did rarely reached page 1, because a page about two things is only as strong as its weaker half. It’s the direct reason I now treat a minimum independent search-demand threshold on each side of a combinatorial page as non-negotiable, not a nice-to-have.
What could be wrong with this article
In the interest of the same honesty standard I’m applying to the sources above:
- The most granular dataset in this piece is a single, anonymized, vendor-published client case. I’ve treated its internal structure (how template families diverge) as more reliable than its topline ROI figure, but it’s still one unaudited data point, not a controlled study.
- The Zapier traffic table demonstrates that secondary sources disagree — it does not resolve which figure, if any, is accurate. I don’t have primary access to Zapier’s Search Console.
- Google doesn’t publish exact numeric thresholds for scaled content abuse enforcement. Everything in the “ceiling” section is a synthesis of Google’s public policy language plus practitioner reporting on enforcement patterns, not a leaked rulebook.
- The economics framework is a decision tool, not a benchmark — it will only be as accurate as the numbers you put into it.
Frequently asked questions
Is programmatic SEO against Google’s rules?
No — the practice itself isn’t prohibited. Google’s scaled content abuse policy targets pages generated primarily to manipulate rankings without adding real value, regardless of whether they’re built with AI, no-code tools, or human writers. Programmatic pages built on genuinely unique, per-page data (verified listings, first-party pricing, real product data) are explicitly treated differently from templated shells.
How many unique data points does a programmatic page need?
There’s no official minimum Google has published. The one granular case study reviewed here used a working internal rule of at least 3 unique, dynamically-pulled data points per page as a quality gate — pages below that threshold were held back and sent for data enrichment rather than published. Treat that as one practitioner’s internal QA bar, not a documented ranking factor.
Should I publish all my pages at once?
The evidence points against it. Sudden, large batches appear to strain crawl and indexing systems for newer or lower-authority sites — one documented case saw its indexing rate drop from 87% to 41% during a single week it tried publishing at roughly 3x its normal pace, and reverted to a capped weekly cadence afterward. A phased rollout with an internal linking hub in place before scaling appears to be the more consistent pattern across sources.
Which page type indexes and ranks most reliably?
Across the sources reviewed, single-entity, informational pages (glossary/definition-style) consistently outperformed combinatorial pages (pairing two variables, like tool-to-tool integrations) on both indexing and page-1 rate. Combinatorial pages are only as strong as their weaker input, which is why several practitioners apply a minimum search-demand threshold to each side of a pairing before generating that page.
How long before a programmatic SEO program shows results?
Every source reviewed described a slow first several months — commonly cited as roughly 3-6 months before visible ranking movement, sometimes longer — followed by a faster compounding phase once a meaningful share of pages reach page 1 and internal link equity has matured. This is consistent enough across independent sources that “it’s not working yet” at month three shouldn’t, on its own, be treated as a verdict.
Do I need to refresh programmatic pages after they’re published?
The pattern across sources says yes. Traffic decay on a meaningful share of pages (double digits in the case reviewed here) appears normal, not a failure state, and a periodic audit-and-refresh cycle on underperforming pages is described as materially more efficient than letting the same pages decay and eventually pruning or merging them.
Is a smaller programmatic SEO program (under 200 pages) worth doing?
It’s arguably the more defensible starting point based on everything above. Several sources explicitly frame large page counts as a scaling decision to make after a smaller template has proven its index rate, page-1 rate, and conversion path — not as a starting move. If your team can’t clearly define the conversion path for a template before publishing, that’s a stronger signal to hold off than the page count is.
Ten thousand pages isn’t a strategy. Knowing which two hundred are worth publishing first is.