Dispersion Returned – Overall AI Visibility Holds While the Gaps Widen Again
Six weeks into the CiteCompass NZ Election 2026 AI Visibility Monitor, the headline number appears remarkably stable.
The average AI Visibility Score across the 30 monitored party-policy profiles was 10.4/100 when rounded to one decimal place, the same reported figure as Week 5.
Beneath that stable headline, however, Week 6 recorded substantial redistribution.
Across the 30 individual profiles, 12 increased, 13 decreased and five were unchanged. The combined absolute movement across those profiles was 124 score points, even though the net change across the entire dataset was just -2 points.
And another pattern emerged:
The measured visibility gap widened across every policy domain in Week 6.
Cost of Living, Crime & Public Safety, Economy, Jobs & Growth, Health and Housing all recorded a larger spread between their highest and lowest measured scores than they did in Week 5.
At the study midpoint, the Week 6 picture is therefore not one of broad growth or decline.
It is one of increasing redistribution and dispersion within a persistently low-visibility information environment.
Week 6 reporting period: 21–28 September 2026 UTC.
AI Visibility Scores are CiteCompass research metrics measured on a 0–100 scale within the defined research framework. They measure observed visibility within generative AI research activity. They do not measure policy quality, electoral support or voter preference. The methodology applies an equivalent research framework across all six monitored parties. CiteCompass
Week 6 at a Glance
Average AI Visibility Score: 10.4/100
Exact average: 10.37
Week 5 exact average: 10.43
Profiles below 20: 27/30 — 90%
Profiles at 20 or above: 3/30
Profiles recording zero: 5
Highest measured Week 6 score: 41
Profiles increasing: 12
Profiles decreasing: 13
Profiles unchanged: 5
The important point is the contrast between the aggregate and the underlying data.
The overall average changed by less than one tenth of one point.
Individual profiles moved by as much as 15 points in a single week.
Week 6 Scorecard
| Policy domain | National | Labour | Green | NZ First | Te Pāti Māori | ACT |
|---|---|---|---|---|---|---|
| Cost of Living | 19 | 38 | 17 | 10 | 7 | 10 |
| Crime & Public Safety | 17 | 11 | 4 | 2 | 0 | 10 |
| Economy, Jobs & Growth | 17 | 14 | 0 | 5 | 8 | 1 |
| Health | 17 | 41 | 20 | 14 | 12 | 5 |
| Housing | 0 | 1 | 6 | 0 | 5 | 0 |
Important: highlighted figures simply identify the highest measured AI Visibility Score within each policy-domain row. They do not indicate policy quality, political support or voter preference.
Insight 1: An Almost Unchanged Average Conceals 124 Points of Movement
The Week 5 average was 10.4333.
The Week 6 average was 10.3667.
Both round to 10.4.
Looking only at the headline average would therefore suggest very little happened.
The individual profiles tell a different story.
Week 5 to Week 6 produced:
12 increases
13 decreases
5 unchanged profiles
The sum of the absolute movements across the 30 profiles was 124 score points.
Yet after all of those gains and losses offset one another, the combined Week 6 total was only two points below Week 5: 311 versus 313.
This is an increasingly important feature of the longitudinal study.
A stable aggregate AI Visibility Score does not imply a stable information environment.
Insight 2: The Cross-Party Visibility Gaps Widened in Every Policy Domain
Week 5 produced an interesting finding: compared with the Week 1 baseline, visibility ranges had narrowed across four of the five monitored policy domains.
Week 6 reversed that pattern sharply.
| Policy domain | Week 5 range | Week 6 range | Change |
|---|---|---|---|
| Cost of Living | 20 | 31 | +11 |
| Crime & Public Safety | 5 | 17 | +12 |
| Economy, Jobs & Growth | 13 | 17 | +4 |
| Health | 24 | 36 | +12 |
| Housing | 5 | 6 | +1 |
All five domains became more dispersed in Week 6.
The change is particularly pronounced in Health and Crime & Public Safety, where the difference between the highest and lowest measured profile increased by 12 points in a single week.
Cost of Living widened by 11 points.
This means Week 6 does more than redistribute visibility between individual profiles. It also reverses the convergence observed in Week 5.
There is an additional six-week perspective worth noting.
Compared with the Week 1 baseline, Week 6’s cross-party ranges are now larger in four of the five domains, while Health has returned to exactly its original 36-point range.
Again, this is a statistical observation about AI visibility rather than a statement about the underlying policies.
Insight 3: Cost of Living Falls Overall Despite One Large Increase
Cost of Living decreased from a domain average of 18.2 in Week 5 to 16.8 in Week 6.
The individual movements were:
| Party | Week 5 | Week 6 | Change |
|---|---|---|---|
| National | 28 | 19 | -9 |
| Labour | 23 | 38 | +15 |
| Green | 20 | 17 | -3 |
| NZ First | 18 | 10 | -8 |
| Te Pāti Māori | 8 | 7 | -1 |
| ACT | 12 | 10 | -2 |
Labour’s movement from 23 to 38 was the largest individual increase recorded during Week 6.
But five of the six Cost of Living profiles decreased.
Collectively, the domain lost eight score points and its average fell from 18.2 to 16.8.
That creates an interesting contrast: one profile increased substantially while the domain overall became less visible.
Week 6’s Cost of Living average of 16.8 is also the lowest recorded for this domain during the first six weeks.
The sequence has been:
19.0 → 17.3 → 17.2 → 21.3 → 18.2 → 16.8
Insight 4: Economy, Jobs & Growth Moves in the Opposite Direction
Economy, Jobs & Growth produced the largest positive domain-level movement in Week 6.
Its average increased from 4.3 to 7.5.
The individual movements were:
| Party | Week 5 | Week 6 | Change |
|---|---|---|---|
| National | 13 | 17 | +4 |
| Labour | 8 | 14 | +6 |
| Green | 0 | 0 | 0 |
| NZ First | 2 | 5 | +3 |
| Te Pāti Māori | 0 | 8 | +8 |
| ACT | 3 | 1 | -2 |
Five of the six profiles were either higher or unchanged.
Across the domain, the six scores gained 19 points in aggregate, moving the domain average from 4.3 to 7.5.
That is the highest Economy, Jobs & Growth domain average recorded during the first six weeks.
The six-week sequence is:
6.2 → 3.7 → 6.5 → 7.2 → 4.3 → 7.5
This illustrates another recurring feature of the data: domain visibility can move materially from week to week even when the study-wide average moves very little.
Insight 5: Health Remains Remarkably Stable Overall — But Not Underneath
Health provides perhaps the clearest example of aggregate stability masking internal movement.
Its domain average changed from 17.8 in Week 5 to 18.2 in Week 6.
The individual movements were:
| Party | Week 5 | Week 6 | Change |
|---|---|---|---|
| National | 11 | 17 | +6 |
| Labour | 34 | 41 | +7 |
| Green | 27 | 20 | -7 |
| NZ First | 13 | 14 | +1 |
| Te Pāti Māori | 12 | 12 | 0 |
| ACT | 10 | 5 | -5 |
The 41 recorded for the Labour Health profile is the highest individual score in Week 6 and the first score above 40 recorded during the six-week study.
Yet the overall Health average barely changed.
In fact, Health has been strikingly stable at domain level throughout the first six weeks:
18.3 → 17.2 → 18.0 → 17.7 → 17.8 → 18.2
All six weekly Health averages have remained within a range of just 1.2 points.
At the same time, individual Health profiles have moved considerably.
Week 6 therefore strengthens one of the study’s recurring observations:
Domain-level stability can coexist with substantial redistribution between individual profiles.
Insight 6: Crime & Public Safety Reverses Much of Its Week 5 Convergence
Crime & Public Safety decreased from 10.0 in Week 5 to 7.3 in Week 6.
The individual movements were:
| Party | Week 5 | Week 6 | Change |
|---|---|---|---|
| National | 13 | 17 | +4 |
| Labour | 10 | 11 | +1 |
| Green | 9 | 4 | -5 |
| NZ First | 8 | 2 | -6 |
| Te Pāti Māori | 8 | 0 | -8 |
| ACT | 12 | 10 | -2 |
The six Week 5 scores had been unusually compressed, ranging from 8 to 13 — only a five-point spread. CiteCompass
One week later, they range from 0 to 17.
The cross-party range therefore expanded from 5 points to 17.
The domain also lost 16 aggregate score points, the largest domain-level decline during Week 6.
This is a useful reminder that the convergence identified in a single weekly observation does not necessarily persist.
Insight 7: Housing Remains the Persistent Low-Visibility Domain
Housing increased slightly from an average of 1.8 to 2.0.
The individual movements were:
| Party | Week 5 | Week 6 | Change |
|---|---|---|---|
| National | 5 | 0 | -5 |
| Labour | 0 | 1 | +1 |
| Green | 1 | 6 | +5 |
| NZ First | 0 | 0 | 0 |
| Te Pāti Māori | 5 | 5 | 0 |
| ACT | 0 | 0 | 0 |
Three Housing profiles recorded zero in Week 6.
The Green Housing score increased from 1 to 6, while National moved from 5 to zero.
Despite that redistribution, the domain itself remains extremely low compared with the other four policy areas.
The six-week Housing averages are:
1.0 → 2.2 → 0.2 → 1.3 → 1.8 → 2.0
Housing has therefore remained below 2.2 throughout the study.
Insight 8: The Number of Low-Visibility Profiles Increased Again
Week 5 recorded 25 of 30 profiles below 20/100.
In Week 6 that increased to 27 of 30.
That means:
90% of all monitored party-policy profiles scored below 20/100 in Week 6.
Only three profiles recorded 20 or above:
- Labour Cost of Living — 38
- Labour Health — 41
- Green Health — 20
The number of profiles recording zero remained at five.
However, their composition changed.
This supports the broader longitudinal observation that individual low or zero scores should not necessarily be interpreted as permanent characteristics of a profile. The measured information environment continues to change from week to week.
The Mathematics Behind the Week 6 Redistribution
A useful way to understand Week 6 is to look at how each policy domain contributed to the overall movement.
| Policy domain | Week 5 total | Week 6 total | Change |
|---|---|---|---|
| Cost of Living | 109 | 101 | -8 |
| Crime & Public Safety | 60 | 44 | -16 |
| Economy, Jobs & Growth | 26 | 45 | +19 |
| Health | 107 | 109 | +2 |
| Housing | 11 | 12 | +1 |
| Total | 313 | 311 | -2 |
This explains why the overall average barely changed.
The +19 movement in Economy, Jobs & Growth and smaller gains in Health and Housing were almost completely offset by decreases in Crime & Public Safety and Cost of Living.
The result was considerable internal movement but almost no aggregate movement.
Six Weeks In: The Overall Trend
The overall average AI Visibility Score has developed as follows:
| Reporting week | Overall average |
|---|---|
| Week 1 | 9.9 |
| Week 2 | 9.5 |
| Week 3 | 10.2 |
| Week 4 | 11.1 |
| Week 5 | 10.4 |
| Week 6 | 10.4 |
The highest study-wide average remains the 11.1 recorded in Week 4.
Week 6 is approximately 0.47 points above the Week 1 baseline, but below the Week 4 peak.
The six-week evidence therefore does not show a sustained upward trajectory in overall political-party AI visibility.
Instead, the overall level has remained within a relatively narrow band while the distribution underneath it has repeatedly changed.
What Week 6 Adds to the Study
At the halfway point, four statistical patterns now stand out.
First, overall AI visibility remains low. The Week 6 average is 10.37/100, and 27 of the 30 monitored profiles score below 20.
Second, aggregate stability continues to conceal substantial profile movement. Week 6 recorded 124 points of absolute profile movement despite a net change of only -2.
Third, dispersion has returned. All five policy domains have wider cross-party score ranges than they did in Week 5.
Fourth, policy domains are behaving differently. Economy, Jobs & Growth reached its highest average of the six-week study while Cost of Living reached its lowest. Health remained exceptionally stable at domain level despite substantial movement underneath it. Housing remained persistently low.
The Week 6 finding can therefore be summarised as:
At the study midpoint, overall political-party AI visibility remains almost unchanged, but the distribution has become considerably more dispersed — with substantial week-on-week movements occurring beneath a stable aggregate score.
What the Data Does — and Does Not — Tell Us
The CiteCompass NZ Election 2026 AI Visibility Monitor is a 12-week longitudinal observational study examining how publicly available political information is surfaced, represented, sourced and cited by generative AI systems. All six monitored parties are assessed using an equivalent research architecture, profile framework, dynamic-query methodology, monitoring cadence, measurement framework, QA process and reporting methodology. CiteCompass
The scores do not assess whether a policy is good, bad or preferable. They do not measure voter sentiment, political support or electoral prospects, and the research does not establish that changes in AI visibility cause changes in voter behaviour. CiteCompass
The information environment is dynamic. Party information, media coverage, government and institutional material, other third-party sources and the AI systems themselves can change over time.
For that reason, the Week 6 report describes what changed in the measured visibility data without attributing individual movements to a specific political announcement, technical change, campaign decision or content intervention unless separate evidence supports that conclusion.
Our Governing Research Principle
Observe. Measure. Report. Do not advocate.
About the NZ Election 2026 AI Visibility Monitor
The CiteCompass NZ Election 2026 AI Visibility Monitor is an independent 12-week longitudinal observational study examining how leading generative AI platforms surface, represent and cite publicly available political-policy information during the lead-up to the 2026 New Zealand General Election.
The study monitors six parliamentary political parties across five policy domains: Cost of Living; Economy, Jobs and Growth; Health; Housing; and Crime and Public Safety.
All parties are monitored using an equivalent research architecture, profile framework, dynamic-query methodology, AI-platform framework, monitoring cadence, measurement framework, data-retention process, QA process and reporting methodology.
Individual research queries may change over time as part of CiteCompass’s dynamic methodology designed to model changing information needs and the evolving election information environment.
The study is observational rather than a controlled experiment.
Political Independence
CiteCompass is politically independent.
The research does not assess which political party has the best policies, endorse or oppose a political party or candidate, recommend how New Zealanders should vote, predict election results, constitute political polling, measure voter sentiment or claim that AI visibility determines voter behaviour.
CiteCompass reports measured changes in how publicly available political information is represented, surfaced, sourced and cited within generative AI systems. (CiteCompass)
CiteCompass has also stated that it will not provide AI-visibility optimisation services to any monitored political party or candidate for the duration of the research programme. (CiteCompass)
Research Information
Week 1 baseline: Research commenced 25 August 2026
Week 5 reporting period: 14–21 September 2026 UTC
Study duration: 12 weeks
Profiles monitored: 30 party-policy combinations
Methodology: Version 1.0
Research and Media Enquiries
Journalists, researchers, political parties and other interested parties are welcome to contact CiteCompass regarding the research framework, methodology, published findings or objective factual corrections.