When an Australian homeowner asks an AI assistant which solar installer to use, the answer is largely being written from a single website. In August 2026 we ran 16 real Australian solar queries through Google AI Mode, ChatGPT and Perplexity and traced every source the engines cited. SolarQuotes appeared in 14 of the 16 answers, with 25 separate citations across AI surfaces. The next most cited sources were Reddit with 16 and a roundup site with 12. Of the top eight cited domains, four were comparison or listicle sites and only one was an actual installer.
If you install solar in Australia, that concentration is the most important marketing fact of your year, because it means your presence in AI recommendations currently runs through infrastructure you do not own. This article covers what we measured, why it happened, and the two track strategy that follows from it.
What the concentration actually looks like
The pattern varies by engine but the direction is consistent. Perplexity and Google AI Mode cite comparison content directly and often, which is how a site reviewing hundreds of installers ends up in nearly every answer. ChatGPT behaves differently: it uses comparison sites as a discovery layer, finding candidate installer names there, then runs its own verification lookups against accreditation registers before deciding who to actually recommend. We watched it run literal searches in the pattern of a register name plus an installer’s brand name, and a third of its background queries targeted an accreditation body. The full verification behaviour is documented in our commercial solar study.
Either way, the funnel starts at the same place. The comparison layer decides which names enter consideration, and for residential solar in Australia that layer is dominated by one site, with a handful of others, a roundup site, Canstar and similar, taking the remaining slots. Reddit threads about installer experiences fill the trust gap around them.
One crucial nuance: this is a residential phenomenon. When we split the queries, aggregator and comparison sites took 35 per cent of ChatGPT’s residential citations and only 3 per cent of commercial ones, where installer websites took 97 per cent. Commercial solar answers are built almost entirely from installers’ own pages, because so few installers publish real commercial content that the engines have nothing else to cite. Residential visibility runs through the comparison layer; commercial visibility is still directly winnable on your own site.
Why one site ended up writing the shortlist
No conspiracy required. SolarQuotes has spent well over a decade publishing exactly what AI systems are built to reward: thousands of pages of specific, dated, structured content about installers, products and prices, a large verified review corpus tied to named businesses, and consistent, checkable facts. AI engines synthesising an answer about Australian solar reach for the deepest, most structured, most verifiable source available, and in this category that source is obvious. The lesson is not that comparison sites cheated. It is that the properties AI rewards, depth, specificity, verifiability, review volume, are buildable, and one participant built them for fifteen years while installers built brochure sites.
Track one: be excellent where the shortlist is written
Whatever your feelings about comparison platforms, the measured reality is that they are where AI discovers residential installers. Treating your presence there as a marketing asset rather than a directory listing is the immediate move.
That means a complete, current profile: accurate service areas, the brands you actually install, your CEC and NETCC standing stated exactly as the registers record it, because ChatGPT cross checks names against registers and a mismatch fails silently. It means review volume on the platform, built the same way you build Google reviews, by asking at the moment of customer satisfaction, with the routine we detailed in our review guide. And it means treating your rating there with the seriousness you treat your Google rating, because it is being read aloud to buyers by machines.
Do the same, proportionally, for the two or three other repeatedly cited sources in our data. And note that Reddit’s 16 citations mean the honest version of community reputation, being recommended by strangers in threads you did not start, now has direct AI distribution. You cannot manufacture that, but you can earn it, and service quality plus a memorable handover moment is how word of mouth digitises.
Track two: build the direct citation asset
The dependence problem with track one is obvious: rented visibility is repriceable and revocable. The second track is making your own site citable enough that engines quote you directly, which is exactly what already happens in commercial solar and increasingly in specific residential niches.
The gap is concrete. Comparison sites publish specific installed prices; most installer sites publish request a quote. Comparison sites publish structured product comparisons; installer sites publish brand logos. Comparison sites date and update content; installer sites have a blog last touched in 2023. Closing that gap looks like: real installed pricing by system size on your site, suburb specific install examples with numbers, product comparison content for the brands you actually fit, and the six verifiable business facts, licence, accreditation, years, service area, response time, reviews, in plain text. The playbook is the same one we laid out in AEO for solar companies, now with measured evidence of where it pays first: your llms.txt and schema make you machine readable, your pricing and project content makes you quotable, and the accreditation alignment makes you verifiable.
Commercial deserves its own sentence. If you do any commercial work at all, a genuine commercial capability page with one detailed case study enters you into answers where installer sites take 97 per cent of citations and only 23 installer domains nationally are being named at all. It is the single highest leverage page in Australian solar marketing right now.
What this means for lead buying
This data reframes the perennial aggregator lead debate, which we covered from the cost side in aggregator leads vs direct. The comparison platforms are no longer just lead resellers you can opt out of; they are the discovery infrastructure AI reads. You can decline to buy leads from a platform while still needing your profile and reviews there to be excellent, because the profile now works for you in channels the platform does not even monetise. Presence and lead buying have become separable decisions. Make them separately.
What we would do this quarter, in order
Strategy sections are easy to nod along to and hard to start, so here is the concrete sequence for an installer acting on this data, ordered by leverage per hour.
Week one is alignment. Pull up your SolarQuotes profile, your Google Business Profile, your NETCC and SAA register entries and your website, side by side, and make the trading name, service areas and claims identical everywhere. This is an afternoon, and it protects everything downstream, because the verification lookups assistants run fail on mismatches you would never notice. The register mechanics are detailed in the accreditation check study.
Weeks two to four are the review push on cited platforms. Your Google reviews matter, but this data says your rating and volume on the comparison platforms AI actually cites matter just as much, and most installers have a fraction of their Google review count there. Point your existing post job review routine at the cited platform for the next quarter, and the relative gap closes fast because so few competitors are doing it deliberately.
Month two is the commercial page, if you do any commercial work at all. One capability page, one real case study with numbers. The 97 per cent installer citation share and 23 domain field in commercial answers makes this the single best effort to visibility ratio in the entire industry right now, and it holds until enough installers read data like this, which is a window, not a permanent condition.
Month three is the direct citation layer on your own site: published installed pricing by system size, two or three suburb anchored project pages with real numbers, and the six facts in plain text. This is slower to pay than the platform work, and it is the only part you own outright, which is exactly why it comes after alignment and before everything else you were planning to spend on.
Throughout, the monthly prompt panel runs in the background, ten minutes a month that converts all of the above from faith into a trend line.
Measure your own shortlist rate
Finally, stop guessing. Build a ten prompt panel of real buyer questions for your service area, best installer in your city, battery quotes, is brand X good, run it monthly through ChatGPT, AI Mode and Perplexity, and track how often you are named and cited. Answers are volatile between runs, we measured only 24 per cent citation overlap on identical queries, so rates over repeated runs are the only honest signal. The full measurement system is in how to track your AI visibility. Installers who measure this monthly know within a quarter whether the two track work is paying. Everyone else finds out when the phone goes quiet.
Frequently asked questions
Which sources do AI assistants cite for Australian solar recommendations?
In our 16 query study: SolarQuotes in 14 of 16 answers with 25 AI citations, Reddit 16, roundup and comparison sites taking most remaining slots. Four of the top eight cited domains were comparison or listicle sites; only one was an installer. Residential answers lean on the comparison layer, commercial answers cite installer sites 97 per cent of the time.
Does ChatGPT just repeat what comparison sites say?
No, and the difference matters. ChatGPT uses comparison content for discovery, then verifies candidates against accreditation registers by name before recommending, with a third of its background searches targeting registers. An installer listed on every comparison site but mismatched or missing on the registers gets dropped silently at the verification step.
Should installers pay comparison sites for leads then?
Separate decisions. Your profile, accuracy and reviews on cited platforms now influence AI answers regardless of whether you buy leads there, so presence should be excellent either way. Whether buying leads stacks up is a cost per acquisition question, covered in our aggregator versus direct analysis, and the answer varies by installer.
Can an installer’s own website get cited directly?
Yes, and in commercial solar it is the norm: 97 per cent of ChatGPT’s commercial citations were installer sites, because almost nobody publishes real commercial content and the engines cite what exists. Residentially it requires closing the specificity gap: published installed pricing, dated project examples with numbers, product comparisons and verifiable credentials in plain text.
What is the fastest win from this research?
Two, in a week: align your trading name exactly across your website, CEC and NETCC register entries and comparison profiles so verification lookups succeed, and publish a commercial capability page if you do any commercial work, because that field is 23 installer domains thin nationally.
Will this concentration on one comparison site last?
Probably not in its current extremity, which is a reason to act rather than wait. The engines are actively diversifying their retrieval, Reddit’s 16 citations in our data would have been unthinkable two years ago, and every installer who builds direct citation assets chips at the aggregator share. But the direction of change favours the same work either way: whether answers keep leaning on comparison platforms or shift toward installers’ own pages, the installer with aligned registers, strong platform reviews and a fact dense website wins both futures. The only strategy the shift punishes is the one most installers are currently running, which is having no deliberate presence in either place.
How do I know if any of this is working?
A fixed monthly prompt panel across ChatGPT, AI Mode and Perplexity, tracking naming and citation rates against a baseline. Single answers are noise, identical queries overlap only about a quarter in citations, so rates over repeated runs are the measurement. Expect citation movement in weeks and naming movement in months.
