Why Programmatic SEO Built Only for Clicks Is Spam: A Data Report

Programmatic SEO has a beautiful pitch. Take a template, join it to a database, and publish ten thousand pages while your competitors argue about one blog post. For a decade the pitch mostly worked, and a generation of growth teams learned that pages are cheap and clicks are free.

This report is about why that era is closing, with the numbers that closed it. Not because programmatic publishing is inherently bad, but because the version most teams actually ship, pages generated for the click rather than for the person clicking, now matches Google’s written definition of spam, gets enforced against at scale, and is being made economically irrelevant by answer engines. The data tells one consistent story from three directions.

The core position

Google’s scaled content abuse policy defines spam as creating many pages primarily to manipulate rankings rather than to help users. Read that twice: it is not a description of AI content or of automation. It is a description of intent. A programmatic build whose only success metric is clicks is optimising itself into that definition, whoever or whatever wrote the words.

The base rate nobody prices into the spreadsheet

Every programmatic business case starts with a page count multiplied by an assumed trickle of traffic per page. The public data says the second number is mostly zero. Ahrefs studied roughly 14 billion pages in its index and found that 96.55 percent get no traffic from Google at all. Another 1.94 percent get between one and ten visits a month. Meaningful search traffic is concentrated in roughly the top 1.5 percent of the web, and that concentration is the environment your ten thousand templated pages are entering.

Hold those four numbers together and the shape of the argument appears. The web already ignores almost every page on it. Google spent 2024 removing nearly half of the unoriginal layer and manually erasing entire sites. And the interface that increasingly answers the query sends almost nobody to the pages that fed it. Volume, the one thing programmatic SEO produces natively, is the input the system now discounts hardest.

2024: the year the tactic got a name in the spam policy

Until March 2024, mass produced content lived in a grey zone. Then Google shipped a core update and three new spam policies in the same week, and one of them, scaled content abuse, named the tactic precisely: many pages created primarily for rankings, whether they were made by AI, by humans, or by a human pasting AI output into a CMS. The method was explicitly declared irrelevant; the intent is the violation.

The enforcement was not theoretical. Google projected the work would cut low quality, unoriginal content by 40 percent and later confirmed the real figure was 45 percent. During the rollout, Originality.ai tracked 79,000 sites and found 1,446 hit with manual actions, together losing more than 20 million monthly visits. In its sample of fully deindexed sites, every single one showed signs of AI generated content, and half were publishing 90 to 100 percent AI written posts. The pattern continued through the site reputation abuse crackdown that May and another spam update that December. This was not one update; it was a policy regime change.

Case file: the most instructive collapse in the dataset

The clearest single illustration is the one its architect published himself. In late 2023, a founder bragged on X about an “SEO heist”: his team exported a competitor’s sitemap, fed 1,800 of its article titles into an AI pipeline, and published the lot for a software client. It worked, briefly and spectacularly. Reported traffic reached roughly 490,000 visits in October 2023, around 600,000 weekly at the peak, and the thread claimed 3.6 million visits taken from the competitor in total.

Then the mechanism met its counterparty. The brag went viral, Google staff took public note, and across the following updates the site gave back the gains and kept falling, ending below where it started, with the brand permanently attached to the story. The entire arbitrage lasted about four months.

The instructive part is not that it failed. It is the payoff structure it reveals: the upside of click first programmatic publishing is a temporary traffic spike, and the downside is the domain itself. That is a trade you can only rationally take on an asset you do not care about, which is precisely why the tactic clusters on burner sites, and why putting it on a brand you intend to keep is mispricing the risk.

Why it feels like spam even when it ranks

Enforcement explains the traffic graphs. It does not explain the feeling, and the feeling matters commercially, because the visitor having it is your prospective customer. The diagnosis is a concept from information retrieval: information gain. What does this page know that the other results do not?

Run the substitution test on any click first programmatic build. Take the page for one city, swap the city name for another, and ask what else would need to change. If the honest answer is nothing, the page contains no information about the thing it claims to cover, and readers detect that within seconds of arrival. They came asking a specific question; the page answers the template. The result shows up in behaviour long before a quality rater sees it: visitors bounce back to the results, the query gets re run, and the engagement signals the modern ranking systems lean on accumulate against every sibling page in the build.

The commercial version of the problem is quieter but worse. Clicks were never the asset; they were a proxy for demand. A templated page that ranks for a low intent permutation collects visits that convert at close to zero, which is how teams end up reporting traffic growth in the same quarter revenue from organic goes nowhere. The old proxy chain, pages produce clicks and clicks produce customers, broke at both ends: Google removes the pages, and the clicks that survive were never customers.

The 2026 kicker: answer engines do not need your ten thousand pages

Everything above described the enforcement risk. The larger force is structural: the click itself is disappearing from the queries programmatic pages were built to catch. Pew Research tracked 900 adults across 69,000 real searches and found that when an AI summary appears, users click a traditional result on just 8 percent of searches, against 15 percent without one. Clicks on the sources cited inside the summary happen on 1 percent of views. A quarter of sessions simply end there.

Now consider what an answer engine does to a programmatic corpus specifically. Ten thousand near duplicate permutations of one template collapse, in a language model’s synthesis, into a single generic answer that needs one citation, and the citation goes to whichever source holds the underlying data or the genuinely differentiated explanation. The rendered permutations are not just unrewarded; they are redundant by construction. In an answer engine market, the winning position is being the source the synthesis must cite, the canonical dataset, the named methodology, the entity with authority, which is a different discipline entirely, and the one we unpack in SEO vs GEO and build for clients through our generative engine optimisation service.

Where the line actually is

None of this is an argument against programmatic publishing. Some of the most defensible SEO assets on the web are programmatic: they are just built on proprietary data rather than on a thesaurus. The line between a programmatic asset and scaled content abuse is visible in five questions:

• Does each page contain information that exists nowhere else on your site, drawn from data you own or collected, rather than reworded from whatever already ranks?
• Does each page answer a query real people actually make, with search demand you verified per permutation, not inferred for the template?
• Would you show any random page from the build to a customer on a sales call? The wince test has not been beaten as a quality classifier.
• Does the page pass the substitution test, with facts, numbers and specifics that change when the variable changes?
• Is there a maintenance plan, because a dataset that goes stale converts a programmatic asset into programmatic liability on a schedule?

Pass all five and programmatic SEO is what it always should have been: a publishing efficiency applied to genuine information, the version we laid out in our programmatic SEO playbook. Fail them and the data in this report is your forecast: a base rate near zero, an enforcement regime with your tactic’s name written into policy, and an interface that answers the query without you. We have rebuilt sites out of exactly that hole, and the forensic sequence behind recovering 40 percent of lost traffic after the Helpful Content Update is what the cleanup looks like in practice: fewer pages, more information, measured by revenue instead of clicks, the same standard we hold our own Search engine to.

Sitting on a programmatic build and not sure which side of the line it is on? Book 15 minutes and we’ll tell you straight.