The Named Index

Which software AI systems name, measured monthly

Research · October 2026

The homes AI named held 4–5 stars about as often as other rated homes in the same areas

A family looking for dementia care might put it to an AI assistant like this: my mum has dementia and can’t live at home any more — which nursing homes in [area] are good for dementia? We asked four AI systems that question, and two other ways of putting it, for ten areas of five Australian cities, and report three of them; every name the answers gave was matched against the government’s Star Ratings. Of the 290 rated homes they named for the area asked about, 211 held 4–5 stars (72.8%), against 73.8% expected from the homes in those same areas: about the same.

This is research on what AI systems say, not advice about choosing care. To find and compare aged care homes, use My Aged Care (myagedcare.gov.au, 1800 200 422, Mon–Fri 8am–8pm, Sat 10am–2pm). For dementia support, call the National Dementia Helpline on 1800 100 500 (24 hours, every day).

The one number

Finding

Of the rated homes the AI systems named for the area asked about, 72.8% held 4–5 stars, against 73.8% expected from the same areas. All ten areas: about the same.

The Overall Star Rating is not a measure of dementia care. It is the government’s overall rating of a home. Every rated home in the extract has 3, 4 or 5 stars overall, and none has 1 or 2, so “4–5 stars” means above the lowest overall rating any home has. In these ten areas, 485 of the 662 rated homes hold it (73.3%).

ChatGPT, Perplexity and Google AI Overviews together. Answers: runs that returned an answer. Named rated homes: distinct rated homes named for the area asked about.
AreasAnswersNamed no homeNamed rated homesAt 4–5 starsShareExpectedDifference (points)How it reads
All ten areas240529021172.8%73.8%−1.0about the same
The two Sydney areas490675988.1%87.0%+1.0about the same
The other eight areas191522315268.2%69.8%−1.6about the same

Expected: each named home counted at its own area’s share of rated homes with 4–5 stars. A home counts once, however often it was named. Unweighted, 485 of 662 rated homes in all ten areas hold 4–5 stars (73.3%), 134 of 154 in the two Sydney areas (87.0%) and 351 of 508 in the other eight (69.1%). A home named for its own area that has no overall rating (16 of the 306 named) is in the name counts below and never in the number.

How a row reads was fixed before any answer existed: a difference of at least 10 points reads higher or lower, and anything smaller reads about the same. Differences are computed before rounding. No significance is claimed, for this row or any other.

The location each system was given: Sydney for ChatGPT and Perplexity; Australia, with no city, for Google AI Overviews. ChatGPT and Perplexity were asked from Sydney and Google AI Overviews for Australia, so a difference between the named and expected shares in either group may partly reflect that setting. It is not a finding about the cities.

The column this page goes with carries a chart of the table above. In words:

THE NAMED INDEX · RESEARCH, OCTOBER 2026

The homes AI named held 4–5 stars about as often as other rated homes in the same areas

Of the rated homes the AI systems named for the area asked about, the share with 4–5 stars, against the share expected from the same areas. ChatGPT, Perplexity and Google AI Overviews.

The chart’s three rows.
AreasShare of named rated homes with 4–5 starsShare expected from the same areas
All ten areas72.8% (211 of 290 named rated homes)73.8%
The two Sydney areas88.1% (59 of 67 named rated homes)87.0%
The other eight areas68.2% (152 of 223 named rated homes)69.8%

Expected: each named home counted at its own area’s share of rated homes with 4–5 stars. A home counts once, however often it was named. ChatGPT and Perplexity were asked from Sydney and Google AI Overviews for Australia, so a difference between the named and expected shares in either group may partly reflect that setting. It is not a finding about the cities.

Star Ratings are not a measure of dementia care.
Based on Australian Institute of Health and Welfare material. Star Ratings, August 2026 extract (released 7 Aug 2026).
30 questions, 3 AI systems, 3 runs, 6 October 2026. namedindex.com/research/dementia-care-homes-2026-10

By system

The same three rows for each kind of system.

ChatGPT and Perplexity, asked from Sydney.
AreasAnswersNamed no homeNamed rated homesAt 4–5 starsShareExpectedDifference (points)How it reads
All ten areas180125918872.6%73.7%−1.1about the same
The two Sydney areas360575087.7%87.0%+0.7about the same
The other eight areas144120213868.3%69.9%−1.6about the same
Google AI Overviews, asked for Australia.
AreasAnswersNamed no homeNamed rated homesAt 4–5 starsShareExpectedDifference (points)How it reads
All ten areas6041197865.5%73.7%−8.2about the same
The two Sydney areas130272385.2%87.0%−1.9about the same
The other eight areas474925559.8%69.8%−10.0lower

One row reads lower: Google AI Overviews in the other eight areas, where 55 of 92 named rated homes held 4–5 stars (59.8%) against 69.8% expected, a difference of −10.0 points, which meets the 10-point line set before the data. It does not change the headline, which reads the three systems together, as was fixed in advance. No significance is claimed for it.

What else the answers show

Runs. Each of the 30 questions was asked three times of each system: 270 runs. 240 returned an answer. In the other 30, Google showed no AI Overview at all. That is not an answer that named no home, and those runs are outside every count on this page.

Answers that named no home: 5 of 240. ChatGPT 0 of 90, Perplexity 1 of 90, Google AI Overviews 4 of 60.

The names given. Across all ten areas the answers gave 1,400 names, each looked up in the Star Ratings extract. A provider’s brand counts as a name of its own only when the answer offers it without naming any of its homes; where the answer names homes, each home it identifies is counted.

Every name given, all ten areas, all three systems.
What the name matchedNamesShare
(a) a home in the area asked about1,14181.5%
(b) a home outside it21615.4%
(c) a provider’s brand, with no single home282.0%
(d) no match in the August 2026 extract151.1%

Homes in New South Wales, named for the other eight areas: 1. ChatGPT and Perplexity 1, Google AI Overviews 0.

How often a home was named again. Of every home a system named at least once for a question, the share it named in every answered run of that question. The same measure the monthly index uses.

By system. Mean stars: the mean Overall Star Rating of the rated homes each system named. All rated homes in the ten areas: 3.76.
SystemAnswersNamed no homeNamed again in every runMean stars, named
ChatGPT90012.4%3.78
Perplexity90156.9%3.70
Google AI Overviews60412.7%3.66
Mean Overall Star Rating by area: the rated homes named for the area, against all rated homes in it.
AreaNamed rated homesMean, namedRated homesMean, all
Northern Sydney373.97993.95
South-west Sydney303.93553.91
Melbourne's eastern suburbs373.651213.70
Melbourne's south-eastern suburbs293.621343.69
Brisbane's northside243.88393.82
The Moreton Bay region, north of Brisbane203.65313.68
Perth's northern and western suburbs363.83603.85
Perth's south-eastern suburbs293.76463.76
Adelaide's eastern suburbs193.74373.68
Adelaide's northern suburbs293.52403.52

The sources the answers cited

The answers cited 1,086 sources, classified by who runs the site.

Citations by kind of site.
SiteCitationsShare
Government sites16114.8%
Care providers’ own sites56652.1%
Aged care directories and listing sites29927.5%
Everything else605.5%

Government sites among the 20 most cited: myagedcare.gov.au 112 (10.3%), agedcarequality.gov.au 17 (1.6%), health.gov.au 12 (1.1%). Care providers’ and directories’ sites are given as shares only, not by name.

What this does not show

It is not a judgement of any home’s dementia care. The Overall Star Rating is not a measure of dementia care, and this page names no home and no provider.

It is not advice, and it is not a ranking. It counts what the answers named; it does not say which home anyone should choose.

Answers change from run to run. The same question asked again can name different homes, as the repeat rates above show.

One run date, three phrasings, ten areas. Every answer was collected on 6 October 2026. Other wordings, other areas or another day could give other counts.

No significance is claimed. The rows read higher, lower or about the same by a rule fixed before the data, not by a statistical test.

An unmatched name means only that it had no match in the August 2026 extract. Unmatched names were not checked against My Aged Care.

Data and method

The 30 questions are three phrasings, each asked for ten areas. One is worded the way a search is: best aged care homes for dementia in [area]. One asks for what families look for: which aged care homes in [area] have a secure dementia unit? One is the way people talk to an assistant: my mum has dementia and can’t live at home any more — which nursing homes in [area] are good for dementia? No question names a provider or a home.

The systems. ChatGPT, on gpt-5.6-luna; Perplexity, on its fast preset, which reported serving openai/gpt-6-luna; and Google AI Overviews, which offers no choice of model. The location each was given: Sydney for ChatGPT and Perplexity; Australia, with no city, for Google AI Overviews. Every run was on 6 October 2026.

Google AI Mode was asked the same 90 questions and is left out: 69 of its 90 answers stored no home, because the homes were shown in a form this study’s capture did not keep.

The ratings. The Star Ratings quarterly data extract – August 2026, released 7 August 2026, published under CC BY 4.0. Its ratings are as the file has them; My Aged Care can show a home a different rating today. Based on Australian Institute of Health and Welfare material. Star Ratings quarterly data extract – August 2026, released 7 August 2026.

The matching rule. Both sides are normalised: lowercased, punctuation stripped, and generic words such as “aged care”, “residential” and “nursing” dropped. A name matches a home in the area asked about if it matches its name exactly; or if every word of it appears in exactly one area home’s name, provider and suburb taken together; and only then is it tried against homes outside the area. A name that matches more than one home is reviewed, never guessed. Every name lands in one of four groups: (a) a rated home in the area asked about, (b) a home outside it, (c) a provider’s brand with no single home, or (d) no match in the August 2026 extract. A brand counts under (c) only when the answer offers it without naming any of its homes.

A random 10% sample of the matches was checked against the answers and the Star Ratings extract in a separate review by an AI assistant, not by a person. It found 9 of the 168 sampled names placed in the wrong group, all of them provider brands, so every match to a provider rather than a home was reviewed again before any figure was computed. An unmatched name means only that it had no match in the August 2026 extract.

Fixed before the data. The questions, the areas, the comparison set of rated homes, the matching rule, the one number and how it reads, and the analysis script were written and committed before any answer existed. After the run, and before any figure was computed, four things were settled: Google AI Mode was left out, for the reason above; unmatched names were not checked against My Aged Care, so an unmatched name means no match in the extract; a provider’s brand counts as a name of its own only when the answer names none of its homes; and every remaining brand and unmatched name was reviewed again, and ten were placed or moved. The analysis script was committed before the answers existed. After the run, and before any figure was computed, two changes were made: the run numbering was corrected (the script counted runs from 1, where the store numbers them from 0), and its list of systems was cut to the three reported, after Google AI Mode was left out for the reason given above. No rule, figure definition or headline rule changed.

Run an aged care home? Email contact@namedindex.com and we’ll send you, free, what the AI systems in this test said about your homes. We use your email only to reply.

Edbert Ongko is the founder of The Named Index, a free monthly index at namedindex.com. The business offers a paid service to software vendors and is considering one for care providers. No vendor was a client in the August or September 2026 editions, no care home in this research is a client or paid to be included or left out, and position in the index cannot be bought.