Pages with almost identical text and a shared address pattern

Why it matters

Pages that change only a place, service or other label may give visitors little useful information of their own. Google describes doorway abuse as pages created to rank for similar searches and lead people towards another destination; this check cannot establish that intent.

What to do

1. Read the three recorded pages and compare what a visitor can learn or do on each.
2. If each page has a distinct purpose, give it relevant details in its main content. If the pages serve the same purpose, decide whether one useful page would serve visitors better.
3. Run a new complete audit after making the change. Changing only canonical, noindex or language declarations does not change this text pattern.

When to leave it as it is: Leave the pages when each serves a legitimate purpose and the repeated text is appropriate. The audit cannot judge that purpose; document why the pages are useful rather than changing words solely to pass a similarity limit.

What we read

We read complete original server HTML with status 200 and one explicit main region: main, role="main" or article. We also need one unambiguous title in the head and the original XML sitemap listing for each page. JavaScript is not run.

The three addresses and titles must share one literal substitution pattern with three distinct values. The address value stays within one path segment, and the corresponding title value consists of one to four normalized words.

We compare multisets of consecutive five-token sequences in the main text. Each of the three pairs must have Jaccard similarity of at least 90%, and unmatched sequence multiplicity must be below 10% in both directions. These are exact product limits, not Google thresholds. A chain of similar neighbours does not qualify as a three-page witness.

An optional sitemap lastmod describes a declared modification date. It is shown only when its value is unambiguous and usable; it neither establishes publication time nor causes this check to report the pages.

Example & evidence

Illustrative example — three original pages, not a live audit.

/plumber-berlin-offer  — Plumber in Berlin
/plumber-london-offer  — Plumber in London
/plumber-paris-offer   — Plumber in Paris

All three pages appear in the original XML sitemap. Their complete main regions contain the same 104-token text: all three direct comparisons have 100% similarity and no unmatched sequences. The distinct place values fit one address and title slot, so the pattern is reported for review.

After review, the pages keep their addresses and titles but receive substantially different information for each location. A later complete audit reads all three original pages and their XML listings and proves that the old three-page pattern no longer holds. The original pattern can then be marked fixed.

Confirm & check again

Check for yourself: Compare the main text, title and address of all three recorded pages. Decide whether the changing value reflects useful differences or merely labels otherwise identical information.

Check again: Full audit — run a new complete audit. The check retains the original three-page witnesses and their direct pairs under a stable address-template identity. Every old obligation needs fresh original page and XML evidence before it can be cleared. A new page or a new representative address cannot replace missing old evidence; incomplete coverage does not prove a fix.

Limits & sources

This is a lexical heuristic with high false-positive risk. It does not identify cities or brands, classify intent, prove AI authorship, predict a search-engine action or establish that Google considers the pages spam.

Only complete original server HTML with a unique main region and title and original sitemap listings can supply the proof. Missing, truncated or ambiguous evidence stays unknown. Pages outside the sitemap do not supply a witness, and disappearance from a sitemap does not by itself fix a previous witness.

The three shown pages are a directly proved witness, not the size of a transitive cluster. Bounded comparison or template search, cancellation and retained-obligation overflow cannot establish that the old pattern has ended.

This checks extracted server text rather than browser appearance. CSS and scripts can make what a visitor sees different. Canonical, noindex and language settings do not suppress this independent text-pattern review.

Confidence: Medium the report contains a directly proved three-page lexical pattern; whether those pages are inappropriate depends on their purpose and useful differences, which this check cannot judge.

Sources

• Google Search Central, Spam policies for Google web search — Doorway abuse — background on doorway abuse, including intent and similar search queries; it does not specify our numeric similarity limits or establish that a reported text pattern is a policy violation. Checked 8 October 2026.

Editorial review: 2 November 2026