Content Uniqueness Rules at Scale
The rules that decide whether a generated set survives are written before the first page is built. Ivan Sp defines which page elements must genuinely differ before any programmatic build begins. Google penalizes thin and duplicate content rather than automated generation, so the rules deciding what differs are what separate a working build from a liability.
Where the Line Actually Sits
Pages generated from a template perform normally when each one carries real, page-specific information. Pages differing only by a swapped city or product name are the ones at risk.
The rules therefore get written before the build rather than after a traffic drop. Retrofitting uniqueness onto a thousand published pages costs more than defining it at the start.
The Element Checklist
One directory build covering roughly 1,000 city pages required six elements to differ genuinely:
- Introductions referencing real local landmarks and context per city
- Long-tail keyword variants combining service, city, and neighbourhood
- Themes summarized from real customer reviews for each listed business
- Questions and answers specific to that city and that service
- Schema markup written per listing rather than per template
- Internal links to related services and nearby locations
The list moves per project. A catalogue with rich specification data supports uniqueness through specifications, and a service business with thin location data does not.
Grading the Source Material First
Source material gets graded before a build gets quoted. Material stating only positioning language grades lowest, and material carrying measured outcomes with dates grades highest.
Low-grade material stops the project. A page cannot be written more specifically than its source supports, and generating anyway means inventing the specifics, which puts unverifiable claims in front of customers at volume.
Quality Scoring After Generation
Generated pages get scored against user-first indicators and against red flags. The indicators cover first-hand experience, specific methods, likely reader satisfaction, actionable information, and original insight. The red flags cover promotional language, writing to hit a word count, pages existing only to rank, generic advice, and obvious keyword optimization.
Pages get a pass, borderline, or fail status from that balance. Failing batches get corrected at the template level, since a template fault repeats across every page it produced.
Volume Is Not the Goal
A system publishing pages nobody searches for produces a large site with no traffic. The result is worse than a small site, because thin pages dilute the quality signal across the whole domain.
Topic selection therefore happens before generation, and the honest outcome on thin data is fewer pages built on real information. Scoring existing pages at volume runs through AI-scored site crawls across large page sets.
Test Two of Your Own Generated Pages
I suggest a test that takes about five minutes.
- Open two generated pages from the same template, side by side.
- Read both and highlight every sentence that differs by more than a name.
- Count the real figures on each page and check whether they differ.
- Ask whether a reader on one page would learn anything from the other.
Fewer than ten genuinely differing sentences means the two pages are one page. Rewriting the template will not solve a data problem.
Where the Standard Uniqueness Advice Breaks Down
If two of your generated pages were printed side by side, what would actually differ?
- The usual advice
- Vary the wording enough between pages and the duplicate content problem goes away.
- What holds up in practice
- Wording variation without substance variation is exactly the pattern the scaled content policy describes. What has to differ is the information, and rewriting the same information three ways produces three pages saying nothing new.
Consider a build using a synonym engine to vary phrasing across 2,000 location pages. Every page reads slightly differently and no two are textually identical.
Each page conveys the same information with different words. A system assessing substance finds 2,000 copies of one page, phrased 2,000 ways.
Ivan Sp varies the data rather than the phrasing. Where the data does not vary, the correct output is fewer pages rather than more synonyms.
Synonym spinning fails the policy and reads badly to humans at the same time. The pages that survive are the ones carrying different facts.
A generated set filed as near-duplicates consumes your crawl budget and returns nothing. The rules that prevent it cost an afternoon to write.
Four Things Commonly Assumed About Content Uniqueness
Which of these describes your uniqueness approach?
I meet these four assumptions whenever uniqueness is treated as a wording problem.
| Commonly assumed | What actually holds |
|---|---|
| Different wording means different content. | Uniqueness is measured in substance. Rewriting the same information three ways produces three pages saying the same thing. |
| A percentage difference threshold is enough. | Textual difference thresholds pass synonym spinning. What matters is whether the page carries information the others do not. |
| Every page needs the same sections. | Structure that varies with the data is itself a difference. Sections that appear only when data exists make pages genuinely distinct. |
| Numbers can be generated to add variety. | Invented figures are fabrication. Real figures from the source are the strongest differentiator available, and only if they are real. |
| Uniqueness can be fixed after generation. | Retrofitting difference across thousands of pages costs more than the original build. The rules belong before the first page. |
The rules take an afternoon to write and they decide whether the whole build was worth doing. Nothing later in the project can substitute for them.
A Worked Example: Two Uniqueness Rules Applied
Consider one template evaluated under two rule sets. The comparison below is illustrative, written to show what a substance rule changes. No client engagement is described here.
| Element | Wording-variation rule | Substance rule |
|---|---|---|
| Title tag | Same phrase, name substituted | Built from service, area, and a real attribute |
| Opening paragraph | Synonyms rotated | Built from three varying data fields |
| Body sections | Identical on every page | Present only where the data exists |
| Figures shown | None | Four real values per row |
| Internal links | Same five on every page | Generated from genuine relationships |
| Pages a reader could distinguish | Almost none | All of them |
| Likely index rate | Low | High |
I write the right-hand column as a rule before the build starts. Every row in it is checkable on a sample of twenty pages.
Synonym variation is repainting identical houses in different shades. From the street they are still the same house.
The section row does more work than it looks. Structure that varies with the data makes two pages genuinely different documents.
How the Rules Get Written
I grade the source material before defining any rule, since the source sets the ceiling.
- Audit the data source and list every field that genuinely varies between rows.
- Grade the source on how many words of real, differentiated content each row can support.
- Set a minimum count of varying elements that every generated page must satisfy.
- Specify which elements must draw on real data rather than on template phrasing.
- Specify which body sections appear conditionally, so structure varies with the data.
- Prohibit synonym substitution as a uniqueness mechanism, in writing.
- Prohibit any generated figure that does not come from the source.
- Generate a sample of twenty pages and read them side by side.
- Score the sample against the rules, and stop the build if the sample fails.
- Re-score a sample from each published batch, since data quality often varies across a source.
How This Engagement Is Scoped
I scope rule work as a phase before the build rather than as a separate engagement.
- What arrives
- A written rule set, a source grade, and a scored sample of twenty generated pages.
- Typical duration
- One to two weeks, before any template work begins.
- What the client supplies
- The data source and subject review of the sample batch.
- Included
- Source audit, rule definition, sample generation, and sample scoring.
- Not included
- Improving the data source, which is quoted separately where it is needed.
- What changes the estimate
- Several page types, or a source needing significant cleaning first.
- How it is quoted
- As a phase inside a programmatic build, or standalone after a scoping call.
Tools and Primary Sources
The references below define the policy the rules exist to satisfy. Each opens in a new tab.
- Google spam policies defines scaled content abuse and what separates it from legitimate automation.
- Creating helpful, people-first content is the standard each generated page has to meet individually.
- Consolidating duplicate URLs explains how near-identical pages are consolidated rather than indexed separately.
- Search Console Page Indexing report reports the near-duplicate statuses that tell you a rule set failed.
Common Questions
How much of a page can be templated?
Structure, navigation, schema, and framing can be templated safely. The substance a reader came for has to differ, and the split usually lands around half the visible words.
Does spinning text with synonyms count as unique?
No. Synonym substitution produces different strings carrying identical information, which is the pattern quality systems are built to detect. Real differences come from real data.
What if a client insists on building pages the data cannot support?
The risk gets stated in writing before the build. Where the data genuinely will not support the page count, the honest recommendation is fewer pages, and that recommendation gets made even when it reduces the project.
How much unique content does a generated page actually need?
Enough that the page carries information the others do not. Word count is the wrong measure, and the right one is whether a reader on one page would learn something new from another.
Discuss This Work
The first call is a free 30-minute consultation with no pitch attached. You receive an honest assessment of where the site stands and what can realistically be built.