AI discovery surface research: needs-driven economy insights

Research by Paul Hewett · In Marketing We Trust · January 2026. The strategic implications were published by B&T in What the RBA told us about marketing strategy.

The headline finding: AI surfaces are not just answering questions; they’re making recommendations, naming winners, and providing decision-ready verdicts. In a needs-driven economy where consumers are actively comparing providers for energy, insurance, and childcare, the brands that appear in these AI answers are winning consideration. The brands that don’t appear are invisible at the point of decision.

Executive summary

This research explores how AI-powered search surfaces respond to realistic Australian consumer queries (Google AI Overview, Google AI Mode, and ChatGPT) about purchase decisions in high-inflation essential categories.

What CMOs need to know

Challenger brands are eating your lunch. GloBird, 1st Energy, and Budget Direct are being recommended ahead of Origin, AGL, and NRMA in cost-focused queries. Clear value propositions beat brand awareness in AI surfaces.
AI Mode delivers verdicts, not just information. Google told a user, “you have a good deal; don’t switch”, and meant it. This is advice, not search results.
Comparison sites are your gateway to AI visibility. Canstar, Finder, and Energy Made Easy appeared in 12 of 13 queries. If you’re not winning on aggregators, you’re not winning in AI.
Measurement is harder than you think, but possible. GSC blends AI traffic with organic. ChatGPT traffic can be isolated in GA4 with proper setup.

What search marketers need to know

Long-tail, conversational queries are exploding. “My electricity bill was $650 last quarter for a 4 bedroom house; is that normal for NSW?” is the new search query. Your content needs to answer this.
ChatGPT reads your PDFs. The insurance comparison triggered 84 seconds of “thinking” and cited actual PDS documents. Technical SEO for documents now matters.
A local query bypassed AI Overview in our test. “Childcare centres Penrith” triggered a local pack, not an AI response. Local SEO and AI optimisation are different disciplines.
You can track ChatGPT referrals today. Set up a custom channel group in GA4 with the regex provided in this report. Stop flying blind.

Economic context

The inflation picture

Headline CPI rose 3.8% in the 12 months to December 2025, well above the RBA’s 2–3% target range. The RBA’s preferred measure of underlying inflation (trimmed mean) rose to 3.3%, confirming inflationary pressures remain elevated.

Category	Annual change (Dec 2025)	Consumer behaviour implication
Electricity	+21.5%	Active provider comparison; “should I switch?” queries
Insurance	+2.6%	Renewal shock; switching vs negotiating queries
Childcare	+11.2%	Local availability + pricing transparency queries
Housing	+5.5%	Mortgage/rent stress; cost benchmarking
Recreation & Culture	+4.4%	Discretionary spending under pressure
Rent	+3.9%	Cost of living baseline pressure
Food & Groceries	+3.4%	Budget benchmarking; “is this normal?” queries

Source: Australian Bureau of Statistics, CPI, December 2025. Insurance is the national CPI category, not a range of individual renewal quotes. The behaviour column is my interpretation.

The discretionary vs non-discretionary gap

Non-discretionary inflation (3.9%) continues to outpace discretionary inflation (3.5%). This 0.4 percentage point gap is critical: when essential costs rise faster than discretionary costs, household budgets tighten around necessities, and purchase behaviour shifts fundamentally.

The shift: from brand-led discovery (“I wonder what’s new”) to need-led comparison (“How do I reduce this cost?”)

The marketing implication: your customers are not scrolling social feeds waiting to be inspired. They’re searching, comparing, and asking AI systems for recommendations. The brands that appear in those answers win.

Research methodology

Approach

We tested 13 realistic consumer queries across three AI-powered discovery surfaces, capturing screenshots and documenting the complete response (brands cited, response depth, SERP structure, and measurement capability).

Surface	Queries tested	Character
Google AI Overview	4	AI-generated summary within standard Google Search
Google AI Mode	4	Conversational, multi-turn AI search experience
ChatGPT	5	Standalone LLM with web search and extended reasoning

Query design principles

All queries were designed to reflect realistic consumer intent in high-inflation categories: NSW geographic focus, household context embedded (“family of 4”, “4-bedroom house”, “25kWh/day usage”), decision-stage framing (“should I switch?”, “is that normal?”), specific data points ($650 quarterly bill, $400/week groceries, $800k house value).

These are designed scenarios, not a sample of observed consumer search logs. They let us inspect what the surfaces return for a realistic decision. The counts below describe these 13 tests, not market share or measured customer conversion.

Key findings

Finding 1: AI surfaces deliver verdicts, not just information

Every surface tested went beyond information retrieval to provide:

Surface	What it does	Example
Google AI Overview	Structures comparisons with clear positioning	“AGL for rewards, Origin for tech, EnergyAustralia for balance”
Google AI Mode	Gives explicit recommendations and verdicts	“Your $650 bill is quite standard; don’t switch, you have a good deal”
ChatGPT	Provides decision frameworks and personalised analysis	5-point checklist for insurance decision; PDS document citations

The AI is not neutral. It’s making choices about which brands to recommend. If you’re not in the recommendation, you’re not in the consideration set, regardless of your brand awareness or media spend.

The uncomfortable truth: a consumer who asks AI Mode, “should I switch from Origin?”, might be told, “no, you have a good deal”, and trust that advice. Your competitor’s superior positioning in AI surfaces can neutralise your entire acquisition funnel.

Finding 2: Challenger brands are winning cost-focused queries

Across all energy queries, smaller providers with clear value positioning dominated:

Provider	Times cited (of 6 energy queries)	How AI positions them
1st Energy	5	“Competitive rates”, “22% guaranteed discount”
GloBird Energy	4	“Cheapest”, “23% below Reference Price”
Momentum Energy	4	“Lower-cost, no-contract plans”
Kogan Energy	3	“Cost-competitive”, “as low as $1,491”
AGL	3	“Rewards leader”, Netflix perks (not price)
Origin	3	“Big 3 incumbent”, tech/app (not price)

In cost-conscious queries, AI surfaces recommend challengers on price and position incumbents on features/perks. Origin and AGL were rarely presented as the cheapest option in these tests. That is the positioning the answers gave them, not an independent tariff comparison.

For incumbent CMOs: your brand awareness isn’t protecting you in AI surfaces. If your value proposition is “we’re big and reliable”, you’ll lose to “we’re 23% cheaper” in a needs-driven economy. You need either (a) a credible cost story, or (b) feature differentiation strong enough to justify premium pricing.

For challenger CMOs: this is your moment. AI surfaces are amplifying clear, specific, quantified value claims. “23% below Reference Price” beats “great value” every time.

Finding 3: Comparison sites are the aggregator layer for AI

Comparison sites appeared in 12 of 13 queries:

Comparison site	Appearances	How AI uses them
Energy Made Easy (gov)	6	“Official”, “free”, “independent”; cited as the definitive comparison tool
Canstar	5	Awards, rankings, research; cited for credibility
Finder	4	Awards, best-of lists; “2026 Finder Award” mentioned explicitly
CareForKids.com.au	3	Childcare listings, vacancy data
CHOICE	3	Satisfaction surveys, research; cited for consumer advocacy
Compare the Market	2	General comparison recommendation

Your visibility on comparison sites directly influences your AI visibility. If Canstar ranks you highly, AI will cite that ranking. If you’re not on Energy Made Easy, you’re invisible in government-endorsed recommendations.

Comparison site optimisation is now a brand visibility strategy, not just a lead generation channel.

Finding 4: Google AI Mode demonstrates advisor-level capability

Capability	Example	Implication
Personalised benchmarking	Compared user’s $650 bill against NSW averages	Users get context, not just data
Multi-turn memory	Remembered household context across follow-up questions	Conversation, not query-response
Explicit verdicts	“Your bill is quite standard, if not slightly below average”	AI makes the judgement call
Behavioural advice	“Shift laundry and dishwashing to weekend daylight hours”	Goes beyond product to lifestyle
Push-back on switching	When asked for comparison after verdict, still provided comparison but reiterated, “you have a good deal”	AI has a point of view

AI Mode is becoming a decision advisor, not a search engine. Your content strategy must shift from “ranking for keywords” to “being cited as a recommendation”. If your content is purely promotional, it won’t be cited. If it’s genuinely helpful, comparative, and takes a position, it will.

Finding 5: ChatGPT performs deep document analysis

The insurance comparison query revealed ChatGPT’s sophistication:

84 seconds of visible “thinking” before response
Cited actual PDS documents with links to policy wording
Property-specific analysis (flood risk assessment for 1990s brick construction)
5-point decision framework tailored to the user’s scenario
Offer to personalise further with postcode and specific coverage requirements

ChatGPT cited policy documents while comparing insurers. Citations and elapsed thinking time do not establish how much of each document it processed, or whether every comparison was correct. If your PDS is harder to access, less clearly written, or has gaps compared to competitors, ChatGPT will surface that. Document quality is now a competitive advantage.

Finding 6: The local childcare query returned a map pack

The childcare query “childcare centres Penrith” triggered a local pack/map listing, not an AI Overview:

Query type	What triggers	Primary optimisation
General comparison (“cheapest electricity NSW”)	AI Overview	Content, structured data, comparison positioning
Provider vs provider (“AGL vs Origin”)	AI Overview	Brand content, comparison content, review aggregation
Local service + location (“childcare Penrith”)	Local Pack / Maps	Google Business Profile, reviews, local citations
Complex personal scenario (“my $650 bill, is it normal?”)	AI Mode	Helpful content, benchmarking data, decision frameworks

Local businesses need both GEO (for AI surfaces) and traditional local SEO (for map pack). They’re different disciplines.

Finding 7: More complex queries prompted deeper responses in this sample

Query complexity	Example	AI response	Depth
Simple comparison	“cheapest electricity NSW”	Structured list with prices	High
Head-to-head	“AGL vs Origin”	Feature comparison with positioning	High
Personal context	“$650 bill, is that normal?”	Benchmark + verdict + advice	Very High
Complex scenario	“Compare 3 insurers for 1990s brick house”	84 seconds thinking, PDS citations, decision framework	Very High

We observed deeper responses to queries with embedded personal context. This test did not measure compute use or establish a statistical relationship. Create content that answers complex, context-rich queries, not just simple keywords.

Measurement framework

State of GEO measurement

Two fundamentally different measurement questions exist for AI surfaces, and most brands are conflating them:

Question	What you’re measuring	Current state (Jan 2026)
“Are we visible in AI responses?”	Whether your brand is being cited, recommended, or mentioned when consumers ask AI for advice	No native, first-party visibility metric is exposed by Google, OpenAI, Anthropic or Perplexity. Manual audits and third-party tools are currently required.
“Are we getting traffic from AI?”	Whether consumers are clicking through from AI surfaces to your website	ChatGPT/Perplexity (and other LLMs) can be tracked as referrals in GA4 via referrer/UTM and custom channels. Google AI Overview/Mode clicks are included in GSC “Web” data and cannot be filtered separately.

The uncomfortable reality: you can be highly visible in AI responses (cited as a recommendation) but receive few clicks because the AI answered the user’s question without requiring a click-through.

The measurement reality check

Surface	Platform	Can track clicks?	Can filter separately?	Can track visibility?	Key limitation
Google AI Overview	GSC	Yes	No, blended with organic	No, requires third-party tools	Cannot isolate AI Overview traffic or visibility
Google AI Mode	GSC	Yes	No, blended with organic	No, requires third-party tools	Follow-ups count as new queries
ChatGPT	GA4	Yes (referral)	Yes, via source filter or custom channel	No, requires manual auditing	Requires setup; some traffic may appear as Direct
Perplexity	GA4	Yes (referral)	Yes, via source filter or custom channel	No, requires manual auditing	Lower volume; no UTM auto-append
Claude	GA4	Yes (referral)	Yes, via source filter or custom channel	No, requires manual auditing	Very low volume currently

Isolating identifiable AI referrals in GA4

Filter Session source in Traffic acquisition or an Exploration. A custom channel group is useful for repeatable reporting, but is not required to isolate a source. For a channel rule using “Source matches regex”, a starting pattern is:

^(.*\.)?(chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com)$

Check it against the source values your property actually receives. This identifies attributed referrals, not all AI influence: missing referrers, app transitions and untagged visits can still appear elsewhere. See Google’s custom channel-group instructions.

What to measure now

Metric	Question it answers	Where to find it
ChatGPT referral volume	Are we getting traffic from AI?	GA4 → Traffic acquisition → AI Traffic channel
Long-tail query trends in GSC	Are longer, conversational queries appearing?	GSC → Performance → Query → Custom regex: ^(\S+\s+){9,}\S+$. Query length alone does not identify AI usage.
Conversion rate by channel	Is AI traffic higher quality?	GA4 → Traffic acquisition or Exploration → key events by session source
Manual AI visibility audit	Are we being cited in AI responses?	Search your brand + competitor queries in AI Mode and ChatGPT monthly
AI Overview presence (third-party)	Which keywords trigger AI Overview with our brand?	Semrush, Profound, Conductor

Important: third-party measurement tools lack maturity, so quality and accuracy vary significantly. Many tools pull from APIs and do not reflect what users actually see.

Recommendations

For CMOs: strategic actions

This week:

Audit your AI visibility. Open AI Mode and ChatGPT. Search “[your brand] vs [competitor]” and “best [your category] NSW”. Are you being recommended? With what positioning?
Check your comparison site presence. Log into Canstar, Finder, and your category’s key aggregators. How are you positioned?
Review your value proposition clarity. AI surfaces reward specific, quantified claims. “23% below Reference Price” gets cited. “Great value” doesn’t.

This quarter:

Brief your agencies on AI visibility. Most SEO strategies optimise for blue links. Ensure your partners understand GEO and can audit your AI citation presence across surfaces.
Invest in comparison content that takes a position. AI surfaces don’t want “here are three options”; they want “here’s which option is best for [scenario]”.
Shift measurement downstream. AI clicks may be fewer but higher quality (users arrive informed, further down funnel).

For search marketers: technical and content priorities

Immediate:

Use GA4 session-source filters or set up an AI traffic channel to group identifiable ChatGPT/Perplexity referrals.
Audit PDF and document accessibility. ChatGPT reads policy documents, PDS files, and technical specs.
Monitor long-tail query trends in GSC. Conversational queries (10+ words, question format) can inform content planning; their length does not establish that AI caused them.

Strategic:

Create content for complex, context-rich queries. Build scenario-based guides, calculators, and comparison tools.
Optimise for aggregator visibility. Treat aggregator optimisation as a GEO tactic.
Quantify everything. “Save up to 23%” beats “competitive rates”.
Build decision frameworks into your content. “5 things to check before switching” content gets extracted into AI responses.
Track AI referral conversion rates separately. Early indicators suggest AI referral traffic converts at higher rates.

Sources

Australian Bureau of Statistics, Consumer Price Index, Australia, December 2025
Australian Bureau of Statistics, CPI September Quarter 2025
Reserve Bank of Australia, Monetary Policy Board Statement, 9 December 2025
Reserve Bank of Australia, Statement on Monetary Policy, November 2025
IMWT Discovery Intelligence: proprietary research, January 2026
Google Search Central: What are impressions, position, and clicks?
Search Engine Land: Google AI Mode traffic data comes to Search Console (June 2025)
Analytics Playbook: How to Track Traffic from AI Tools in GA4