Why "programmatic SEO" became a slur

Programmatic SEO used to mean any SEO strategy that generated pages from templates and data. Marketplaces, directories, listing sites, city-by-city landing pages — all programmatic. Some of it was great. A lot of it was garbage.

Google's Helpful Content Update in 2022, followed by the March 2024 core update targeting "scaled content abuse," made the garbage strategy non-viable. Sites that had 30,000 pages from a template + city loop got deindexed en masse. The programmatic SEO subreddit filled up with post-mortems. Consultants started advising clients to delete programmatic pages preemptively.

The blanket advice was wrong, though. What died wasn't programmatic SEO. What died was low-uniqueness programmatic SEO — the pattern where the template is 95% of the page and the variable data is 5%. What survived — and still ranks — is programmatic SEO where each page has enough real, unique content that a human reading it gets something the reader couldn't get from any other page on the site.

The difference is quality controls. This guide is about those controls.

The four controls that separate survives-from-gets-killed

Control 1: Uniqueness thresholds

Every page you generate needs a minimum unique-content threshold — measured in words or characters, not just percentage. Percentage-based similarity checks miss the case where a 400-word template shares structure with 200 words of unique content per page and reads as substantially duplicated.

A working threshold for most verticals:

  • At least 400 words of per-page unique content, minimum
  • Less than 40% of total page content shared with any other page on the site
  • No two pages sharing more than 60% n-gram overlap on 3-5 word sequences

Sites that hold up under algorithmic scrutiny sit well above these numbers — often 1,000+ words of unique content per page. The threshold is the floor, not the target.

Control 2: Real data sources per page

The unique content has to come from somewhere. It can't be AI-generated variations on the same theme. That's the trap — AI can produce 10,000 different-sounding versions of the same idea, but they're semantically the same and Google's evaluating semantics.

Real data sources look like:

  • Actual product specifications, prices, reviews per product page
  • Actual demographic, economic, and geographic data per city page
  • Actual practitioner credentials, service specifics, availability per location page
  • Actual case data, timelines, outcomes per case-type page
  • Actual customer reviews, testimonials, project photos per service page

If you can't identify a real data source that makes each page genuinely different, you can't generate that page programmatically without risk. Which is fine — sometimes the answer is "generate 50 pages instead of 500."

Control 3: Editorial guardrails

Programmatic doesn't mean unedited. A page that ships fully automated is a page nobody looked at. That's where the failure modes live — the "$0.00" price display when a value is missing, the pronoun that says "her" on a page about a male practitioner, the city name that got misspelled in the source data, the sentence that reads like nonsense because a template variable didn't fill in.

The guardrails that work:

  • Every generated page passes automated QA checks (no empty variables, no lorem ipsum, no obviously broken sentences) before publish
  • Every generated page in the first 10% of a rollout gets a human eyeball pass
  • Every generated page gets sampled — 5% ongoing manual review post-launch — to catch drift

The point isn't that a human writes each page. The point is that no page ships without at least one path where a human could have caught something obviously wrong.

Control 4: E-E-A-T signals per page

Every page needs experience, expertise, authoritativeness, and trustworthiness signals — not just the site level. The signals that count on a programmatic page:

  • A named author or authoring team, per page, with credentials that match the topic
  • First-person or observational language where it applies ("we've handled 400 of these cases," "in our clinic we see this most often")
  • Specific evidence — statistics with sources, examples with detail, case data with dates
  • Fresh timestamps that reflect actual updates, not automated re-publish dates

Pages that pass E-E-A-T at the site level but fail at the page level get lower rankings than they should. See the E-E-A-T for local service pages guide for the deeper treatment.

The crawl-budget math nobody explains

Programmatic sites hit crawl-budget limits fast. Google allocates a fixed crawl budget per site — the number of URLs Googlebot will crawl in a given period. A site with 40 pages has plenty of budget for each. A site with 20,000 programmatic pages competes with itself for crawl attention.

This matters because uncrawled pages don't rank. And crawl frequency signals importance — pages Google crawls weekly rank better than pages Google crawls quarterly.

The math:

  • If your crawl budget supports 500 URL fetches/day and you have 20,000 pages, Google will re-crawl each page roughly once every 40 days
  • If your programmatic pages update frequently (prices, availability), 40-day crawl cycles mean users see stale data in the SERP
  • If a subset of your programmatic pages are low-quality, Google spends budget crawling them and skips your good pages

The practical implication: fewer, better pages beat more, thinner pages. If you have 5,000 programmatic pages and only 800 of them actually rank, deleting or noindexing the 4,200 non-performers usually improves the ranking of the 800 performers. Because now the crawl budget concentrates on them.

When programmatic works well

Sites where programmatic SEO reliably works, when the quality controls are in place:

  • Marketplaces with real inventory. Product pages driven by real product data, reviews, and availability. Each page has genuine unique content sourced from the listing.
  • Directory sites with verified data. Business listings with verified NAP, hours, services, reviews. Each page is a real business with real per-listing content.
  • Multi-location service businesses. Location pages for a service business with real per-location detail — the specific staff, the specific service area, the specific reviews. Not "we serve [city]!" templates.
  • Comparison sites with real data. "X vs Y" pages backed by actual feature comparisons, actual pricing, actual review data, not AI-generated pro/con lists.
  • Vertical-specific service pages. Local trade services generated from a skill pack, where the skill knows what a plumber page needs vs an electrician page. See [for/plumbers](/for/plumbers) and [for/electricians](/for/electricians) for examples of this pattern.

When programmatic is a bad idea

Categories where programmatic SEO tends to fail regardless of quality controls:

  • Editorial or opinion content — programmatic can't fake perspective
  • YMYL content (health, finance, legal) where individual expertise per page is required
  • Topics where semantic depth per page matters more than data coverage
  • Long-tail content clusters where each page needs a distinct argument

If your subject matter is one of those, spend the effort on individually-authored pages. Programmatic isn't the answer.

A test for whether your programmatic strategy will survive

Before you generate 500 pages, generate 5. Give them to someone who doesn't work at your company and isn't an SEO. Ask them: "Are these pages genuinely useful? Would you have gotten value from any of them if you searched for them?"

If the answer is yes on 4 of 5, your quality controls are probably working. Scale up.

If the answer is "they're... fine? I guess?" on all 5, your quality controls aren't working. Fix them before you generate the other 495.

Google is running the same test on your pages, at scale, and answering it faster than you can.

The takeaway

Programmatic SEO isn't dead. Low-uniqueness programmatic SEO is dead. The difference is the four controls — uniqueness thresholds, real data sources, editorial guardrails, per-page E-E-A-T — and the discipline to scale slowly enough to verify each cohort before shipping the next.

Done well, programmatic still produces some of the best organic ROI in SEO. Done badly, it produces sites that get deindexed. The controls are what separate them.