What Actually Moves Rank in KZN

A checklist tells a business to add more images, more schema, a visible phone number and a longer homepage, all at once, because each one is individually good practice. Our relational analysis of 529 KZN domains suggests that advice is sometimes backwards.

Nine signal-pair combinations cross the significance bar in the current dataset. Some confirm that doing two things well together beats doing either alone. Others show the opposite: two individually reasonable signals, stacked together, correlating with worse observed rank than either one on its own.

The finding that leads

The strongest result in the current analysis concerns two content signals: how often a domain publishes, and how original that content is. Domains were split into four groups — publish often with generic content, publish rarely with original content, both, or neither — and compared on average observed rank position.

The “both” group averaged position 5.2. Publishing often with generic content, on its own, averaged position 7.7 — worse than doing neither at all (position 5.5). Publishing rarely but originally landed at position 6.3. Frequency without originality was the weakest group in the comparison.

A related pairing tells a similar story from a different angle: originality combined with real topical breadth (eight or more distinct topics covered) reached position 5.5, ahead of originality alone at position 6.3. Breadth without originality — publishing widely about generic subjects — was the weaker group at position 7.1.

Neither finding is a claim that posting a fixed number of articles will move a specific business up a fixed number of positions. It is a description of what the current dataset shows: publishing frequency by itself is not associated with better observed rank, and only combines with genuine originality to produce one.

Where more effort works against you

Four further findings run in the opposite direction. In each case, two signals that independently look like reasonable technical practice correlate with worse average rank when a domain carries both at once:

• A high homepage image count combined with a strong technical score averaged a worse position than a strong technical score alone.
• A visible phone number combined with a strong technical score averaged a worse position than a strong technical score without a visible phone number.
• Wide schema-markup variety combined with a visible phone number averaged a worse position than either alone.
• Full image alt-text coverage combined with a high homepage image count produced the weakest average position of any combined-condition group in the current analysis.

None of this means images, schema markup or a visible phone number are undesirable — each is a normal, sensible thing for a business site to carry. What the data suggests is that stacking several of these signals at once does not compound the way a generic technical checklist implies. In this dataset, the heaviest combinations of conventional “good practice” signals are associated with a weaker average position than doing one of them properly.

A fifth pairing — homepage word count combined with article-style content — showed the same direction but sits right at the edge of statistical significance and is mentioned here for completeness rather than as a confirmed pattern.

Organic and the Maps pack are not the same test

The same analysis run separately against Google’s local pack, rather than organic results, tells a related but distinct story. Two signals recur as sources of interference in both models — but each pairs with something different depending on which result type is being observed.

Technical score interferes with homepage image count and phone-number visibility in organic results, but with publishing volume and content originality in the Maps pack. Image alt-text coverage interferes with image count in organic, but with publishing volume and originality in the local pack — two separate interference pairings, not one repeating.

The signals that cause friction are consistent across both models. What they collide with is not. A business optimising only for organic search may be making decisions that do not transfer to how it appears on the map, and the reverse also holds.

What a separate rank-change comparison adds

A second, independent analysis compared domains that improved in observed rank against domains that declined between two snapshots roughly a week apart, testing 26 individual features. Only one reached statistical significance on its own: domains that declined carried nearly double the homepage image count of domains that improved (an average of 30 images against 16).

That single confirmed result lines up with the interference finding above — a high image count keeps appearing on the less favourable side of the comparison across two independently run analyses. The other 25 features tested moved in a direction but did not clear the significance bar at this sample size, and are not presented as findings.

What we looked at and are not publishing

Two categories of pattern appear in the underlying analysis and are deliberately excluded here, because checking them changed the conclusion.

The pattern layer reports several KZN industries moving in near-perfect statistical lockstep with each other. On inspection, each of those industries is tracked through a single search query with 17 to 18 observations. A correlation that close across single-query time series inside a 70-search network is much more likely to reflect a handful of shared, network-wide volatility events than a genuine relationship between those markets. It would need substantially more tracked searches per industry before that kind of claim could be supported.

The same layer also produces domain-level competitive narratives — which specific sites displaced which others in a given search, repeated across multiple observations. That is a different kind of claim from a rank-position chart: it characterises a named business’s competitive trajectory rather than simply reporting where it appeared. We have not published that layer while the same governance question that applies to naming a domain’s business type remains open.

What happens next

The pattern layer also flags several dates where movement spiked across many tracked searches simultaneously — consistent with a broader change in how results were being served, rather than any single business’s activity. That signal needs no characterisation of any domain to be useful on its own, and the underlying detection can be rebuilt directly from already-published position data. It is not part of this analysis yet, but it is a natural next addition.

The broader reasoning for treating correlational findings this cautiously is set out in Why TVS Does Not Call Every Correlation a Ranking Factor, and the underlying methodology is introduced in The Correlation Engine and Why We Built Market Intelligence.

For the practical service layer built on this evidence, see SEO services in Durban and technical SEO audits.

Data note: this article uses the relational-interaction and rank-change analysis output generated from the 25 August 2026 SERP run. The findings are observational and describe association, not a guaranteed effect of changing any single signal on a specific site. This article is intended to be refreshed as new SERP runs accumulate; the date above reflects the most recent verification.